UPPSC AE 2024 Combined State Engineering Services Examination

Basics of AI and Robotics | Chapter-1

1. What is the best definition of Artificial Intelligence (AI)?

A) AI refers to machines that can think and make decisions exactly like humans
B) AI is the simulation of human intelligence in machines programmed to think and learn
C) AI is the use of robots to replace human workers in factories
D) AI is the study of how computers are built and programmed

2. Who is considered the "Father of Artificial Intelligence"?

A) Alan Turing
B) John McCarthy
C) Geoffrey Hinton
D) Marvin Minsky

3. When was the term "Artificial Intelligence" first coined?

A) 1943
B) 1956
C) 1975
D) 1985

4. What is the Turing Test used for?

A) To measure the speed of a computer

B) To determine if a machine can exhibit intelligent behavior similar to humans
C) To check if a machine has emotions
D) To identify different types of AI

5. Which of the following is NOT a type of AI?

A) Narrow AI
B) General AI
C) Super AI
D) Hardware AI

6. Which of the following is an example of Weak AI?

A) Self-driving cars
B) An AI system that can reason and learn like humans
C) A humanoid robot with human-level intelligence
D) AI that can independently develop new scientific theories

7. Which of these is a major application of AI in healthcare?

A) Online shopping recommendations

B) Automated disease diagnosis using machine learning
C) GPS-based route optimization
D) AI-generated video game characters

8. Which AI technique is used in voice assistants like Siri and Alexa?

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
A) Genetic Algorithms
B) Natural Language Processing (NLP)
C) Neural Networks
D) Supervised Learning

9. Which of the following is an example of Artificial Intelligence in finance?

A) Predicting stock market trends using machine learning

B) Sending emails through an automated system
C) Managing physical currency transactions in banks
D) Providing manual financial consultancy

10. What is the primary goal of Machine Learning (ML)?

A) To manually program computers for every task

B) To enable computers to learn from data and improve over time
C) To replace all human intelligence
D) To increase hardware performance

11. Which of the following best describes the difference between AI and Machine Learning?

A) AI is a broader concept, while ML is a subset of AI

B) AI and ML are the same
C) ML is broader than AI
D) AI is only used in robotics, while ML is used in software

12. Which AI technique is commonly used in self-driving cars?

A) Expert Systems
B) Convolutional Neural Networks (CNN)
C) Traditional Database Management
D) Relational Databases

13. What is Deep Learning?

A) A type of learning where humans manually program AI systems

B) A subset of Machine Learning that uses neural networks to learn from large datasets
C) AI that does not require data to function
D) A programming language for AI

14. Which of the following is an example of an AI-powered chatbot?

A) Microsoft Word
B) Google Chrome
C) OpenAI’s ChatGPT
D) Adobe Photoshop

15. What is Natural Language Processing (NLP) primarily used for?

A) Analyzing and understanding human language

B) Controlling robotic movements
C) Improving database performance
D) Writing programming code

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
16. Which of the following is an example of AI in cybersecurity?

A) AI-powered intrusion detection systems

B) Using traditional firewalls for network security
C) Manual password management
D) Offline virus scanning

17. Which company developed AlphaGo, the AI system that defeated human champions in the
game of Go?

A) Microsoft
B) IBM
C) Google DeepMind
D) OpenAI

18. What is an Expert System in AI?

A) A system that mimics the decision-making ability of a human expert

B) A robot that physically performs expert-level tasks
C) A software that only stores and retrieves data
D) A system that replaces human workers completely

19. What is Reinforcement Learning (RL)?

A) Learning by memorizing past data

B) Learning by interacting with the environment and receiving rewards or penalties
C) Learning without any external input
D) Learning only from labeled datasets

20. Which AI technique is commonly used in fraud detection systems?

A) Genetic Algorithms
B) Machine Learning (ML)
C) Traditional Databases
D) Hard-coded Rules

1. Answer: B) AI is the simulation of human intelligence in machines programmed to think and

learn

Explanation:

A) AI refers to machines that can think and make decisions exactly like humans – AI
simulates human intelligence, but it does not think exactly like humans.

B) AI is the simulation of human intelligence in machines programmed to think and learn –

Correct, AI involves programming machines to process data, recognize patterns, and make
informed decisions.

C) AI is the use of robots to replace human workers in factories – AI is used in robotics, but
AI itself is not limited to automation in factories.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
D) AI is the study of how computers are built and programmed – AI focuses on intelligence
simulation, not just computer programming.

2. Answer: B) John McCarthy

Explanation:

A) Alan Turing – Turing was a pioneer in computing and proposed the Turing Test for AI but is
not the "father" of AI.

B) John McCarthy – Correct, he coined the term "Artificial Intelligence" in 1956 and
contributed significantly to AI research.

C) Geoffrey Hinton – A leading researcher in deep learning, but AI existed long before his
contributions.

D) Marvin Minsky – A key AI researcher but not the one who coined the term.

3. Answer: B) 1956

Explanation:

A) 1943 – AI concepts existed, but the term "Artificial Intelligence" was not used.

B) 1956 – Correct, the term was coined by John McCarthy at the Dartmouth Conference.

C) 1975 – AI research was growing, but the term was already established.

D) 1985 – AI was already a well-known field by then.

4. Answer: B) To determine if a machine can exhibit intelligent behavior similar to humans

Explanation:

A) To measure the speed of a computer – The Turing Test is not about speed.

B) To determine if a machine can exhibit intelligent behavior similar to humans – Correct,

proposed by Alan Turing, it tests whether a machine can imitate human intelligence.

C) To check if a machine has emotions – AI does not yet truly "feel" emotions.

D) To identify different types of AI – The test does not categorize AI types.

5. Answer: D) Hardware AI

Explanation:

A) Narrow AI – Correct, this refers to AI designed for specific tasks (e.g., Siri, Chatbots).

B) General AI – AI that can perform any intellectual task like a human, but it does not exist yet.

C) Super AI – AI that surpasses human intelligence, currently theoretical.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
D) Hardware AI – Incorrect term, AI is software-based and not classified as "hardware AI."

6. Answer: A) Self-driving cars

Explanation:

A) Self-driving cars – Correct, they use AI for a specific task and fall under Weak AI.

B) An AI system that can reason and learn like humans – This would be General AI, which
does not exist yet.

C) A humanoid robot with human-level intelligence – Would be Super AI, which is still
theoretical.

D) AI that can independently develop new scientific theories – This is beyond the current AI
capabilities.

7. Answer: B) Automated disease diagnosis using machine learning

Explanation:

A) Online shopping recommendations – This is an AI application but in e-commerce, not

healthcare.

B) Automated disease diagnosis using machine learning – Correct, AI is widely used in

medical imaging, diagnostics, and predictive healthcare analytics.

C) GPS-based route optimization – Used in navigation, not healthcare.

D) AI-generated video game characters – Used in gaming, not healthcare.

8. Answer: B) Natural Language Processing (NLP)

Explanation:

A) Genetic Algorithms – Used for optimization, not for voice recognition.

B) Natural Language Processing (NLP) – Correct, NLP allows AI to understand and respond to
human language.

C) Neural Networks – Used in AI but not specifically for speech processing.

D) Supervised Learning – A type of machine learning, but not directly responsible for voice
understanding.

9. Answer: A) Predicting stock market trends using machine learning

Explanation:

A) Predicting stock market trends using machine learning – Correct, AI analyzes vast
amounts of financial data to identify patterns and predict trends.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
B) Sending emails through an automated system – This is automation, not AI.

C) Managing physical currency transactions in banks – This involves traditional banking, not
AI.

D) Providing manual financial consultancy – This is a human task, not AI-based.

10. Answer: B) To enable computers to learn from data and improve over time

Explanation:

A) To manually program computers for every task – Traditional programming requires manual
coding, unlike ML, which learns from data.

B) To enable computers to learn from data and improve over time – Correct, ML allows
systems to learn patterns from data and refine their predictions.

C) To replace all human intelligence – AI enhances human tasks but does not replace
intelligence completely.

D) To increase hardware performance – ML is a software concept, not hardware-related.

11. Answer: A) AI is a broader concept, while ML is a subset of AI

Explanation:

A) AI is a broader concept, while ML is a subset of AI – Correct, ML is a specific type of AI

that learns from data.

B) AI and ML are the same – ML is a part of AI, but AI includes more than just ML.

C) ML is broader than AI – AI encompasses ML, not the other way around.

D) AI is only used in robotics, while ML is used in software – AI has applications beyond

robotics.

12. Answer: B) Convolutional Neural Networks (CNN)

Explanation:

A) Expert Systems – Used in rule-based AI, but not ideal for real-time visual recognition in self-
driving cars.

B) Convolutional Neural Networks (CNN) – Correct, CNNs process images and help AI
recognize objects, making them essential for autonomous vehicles.

C) Traditional Database Management – AI in self-driving cars involves real-time decision-

making, not static databases.

D) Relational Databases – Used for structured data storage, not AI decision-making.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
13. Answer: B) A subset of Machine Learning that uses neural networks to learn from large
datasets

Explanation:

A) A type of learning where humans manually program AI systems – Deep Learning (DL)
does not require manual programming for each decision.

B) A subset of Machine Learning that uses neural networks to learn from large datasets –
Correct, DL is a branch of ML that enables AI models to learn from vast amounts of data using
multiple neural network layers.

C) AI that does not require data to function – DL depends heavily on data.

D) A programming language for AI – DL is not a programming language but a method of

training AI models.

14. Answer: C) OpenAI’s ChatGPT

Explanation:

A) Microsoft Word – A word processor, not an AI chatbot.

B) Google Chrome – A web browser, not an AI chatbot.

C) OpenAI’s ChatGPT – Correct, ChatGPT is an AI chatbot designed to interact with users in

natural language.

D) Adobe Photoshop – An image editing software, not a chatbot.

15. Answer: A) Analyzing and understanding human language

Explanation:

A) Analyzing and understanding human language – Correct, NLP enables AI to process and
interpret human language.

B) Controlling robotic movements – Robotics uses AI, but NLP is for language understanding.

C) Improving database performance – Databases use different optimization techniques, not

NLP.

D) Writing programming code – AI can assist in coding, but NLP focuses on human language.

16. Answer: A) AI-powered intrusion detection systems

Explanation:

A) AI-powered intrusion detection systems – Correct, AI can detect anomalies and threats in
real time.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
B) Using traditional firewalls for network security – Firewalls provide security but don’t
typically use AI.

C) Manual password management – AI is not involved in manual password handling.

D) Offline virus scanning – Many antivirus programs now use AI, but traditional virus scanning
does not.

17. Answer: C) Google DeepMind

Explanation:

A) Microsoft – Microsoft works on AI but did not develop AlphaGo.

B) IBM – IBM developed Watson but not AlphaGo.

C) Google DeepMind – Correct, DeepMind (a subsidiary of Google) created AlphaGo, which

beat human Go champions.

D) OpenAI – OpenAI focuses on NLP and reinforcement learning but did not create AlphaGo.

18. Answer: A) A system that mimics the decision-making ability of a human expert

Explanation:

A) A system that mimics the decision-making ability of a human expert – Correct, Expert
Systems are AI programs designed to solve complex problems by reasoning like a human expert.

B) A robot that physically performs expert-level tasks – AI can assist robots, but Expert
Systems focus on decision-making, not physical tasks.

C) A software that only stores and retrieves data – Databases do this, but Expert Systems
analyze and provide reasoning-based decisions.

D) A system that replaces human workers completely – AI assists humans but does not fully
replace them.

19. Answer: B) Learning by interacting with the environment and receiving rewards or penalties

Explanation:

A) Learning by memorizing past data – This describes supervised learning, not RL.

B) Learning by interacting with the environment and receiving rewards or penalties –

Correct, RL is a type of ML where an agent learns by taking actions and receiving feedback.

C) Learning without any external input – RL requires interaction with the environment.

D) Learning only from labeled datasets – This describes supervised learning, while RL is based
on trial and error.

20. Answer: B) Machine Learning (ML)

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
Explanation:

A) Genetic Algorithms – Used for optimization, not typically fraud detection.

B) Machine Learning (ML) – Correct, ML models analyze transaction patterns and detect
anomalies for fraud prevention.

C) Traditional Databases – Databases store information but do not analyze patterns like AI.

D) Hard-coded Rules – While rule-based systems exist, ML is more efficient for fraud detection.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
Basics of AI and Robotics | Chapter-2 | Machine Learning

1. What is Machine Learning?

A) A way to program computers manually

B) A subset of AI that enables computers to learn from data
C) A new programming language
D) A type of database management

2. Which of the following is a type of Machine Learning?

A) Supervised Learning
B) Unsupervised Learning
C) Reinforcement Learning
D) All of the above

3. In supervised learning, the model is trained using:

A) Labeled data
B) Unlabeled data
C) Random data
D) No data

4. Which of the following is an example of supervised learning?

A) Clustering
B) Regression
C) Association Rule Mining
D) Dimensionality Reduction

5. What is Overfitting in ML?

A) A model that performs well on training data but poorly on new data
B) A model that generalizes well to unseen data
C) A model that is too simple to capture patterns
D) A model with fewer parameters than required

6. What is the purpose of Cross-Validation in ML?

A) To improve model accuracy on unseen data

B) To increase training time
C) To generate more data
D) To remove outliers from data

7. What is a Decision Tree in Machine Learning?

A) A tree-like structure for decision-making

B) A database indexing technique
C) A deep learning algorithm
D) A clustering method

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
8. In which ML approach does an agent learn by interacting with an environment?

A) Supervised Learning
B) Unsupervised Learning
C) Reinforcement Learning
D) Transfer Learning

9. What is the primary goal of Unsupervised Learning?

A) Predict future outcomes

B) Group similar data points
C) Classify data based on labels
D) Train a chatbot

10. Which ML model is best for processing images?

A) Decision Trees
B) Convolutional Neural Networks (CNNs)
C) Support Vector Machines (SVMs)
D) K-Means Clustering

11. Which algorithm is commonly used for spam email detection?

A) K-Means Clustering
B) Naïve Bayes
C) K-Nearest Neighbors (KNN)
D) Principal Component Analysis (PCA)

12. Which of the following is NOT a Machine Learning algorithm?

A) Random Forest
B) Support Vector Machine
C) Binary Search
D) Gradient Boosting

13. What is the curse of dimensionality in Machine Learning?

A) More features improve model performance

B) Increasing the number of features can lead to overfitting and computational challenges
C) Models with high dimensions always perform better
D) It is a feature selection technique

14. What is the main goal of Reinforcement Learning?

A) Minimizing loss function

B) Learning from labeled data
C) Maximizing cumulative rewards
D) Grouping similar data points

15. What is the primary purpose of Principal Component Analysis (PCA)?

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
A) Data classification
B) Data clustering
C) Dimensionality reduction
D) Hyperparameter tuning

16. What type of problem does Linear Regression solve?

A) Classification
B) Clustering
C) Regression
D) Reinforcement Learning

17. Which activation function is commonly used in deep learning?

A) ReLU
B) Sigmoid
C) Tanh
D) All of the above

18. What does the term "bias" refer to in Machine Learning?

A) The error due to overly simplistic assumptions in a model

B) The ability of a model to generalize well
C) The error caused by high variance
D) The process of selecting the best features

19. Which of the following is NOT a loss function in ML?

A) Mean Squared Error (MSE)

B) Cross-Entropy Loss
C) Huber Loss
D) PageRank

20. What does "bagging" do in ensemble learning?

A) Uses multiple models to reduce variance

B) Applies boosting to weak learners
C) Creates a deep neural network
D) Selects the best individual model

Solutions

1. Answer: B) A subset of AI that enables computers to learn from data

Explanation:
A) A way to program computers manually – Traditional programming requires explicit
rules, whereas ML learns from data.
B) A subset of AI that enables computers to learn from data – Correct, ML allows
machines to improve their performance based on experience.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
C) A new programming language – ML is not a programming language; it uses Python,
R, etc.
D) A type of database management – ML works with data but is not a database system.
2. Answer: D) All of the above
Explanation:
A) Supervised Learning – Uses labeled data to train models.
B) Unsupervised Learning – Finds patterns in unlabeled data.
C) Reinforcement Learning – Uses rewards/punishments to optimize decision-making.
D) All of the above – Correct, these are the three main ML categories.
3. Answer: A) Labeled data
Explanation:
A) Labeled data – Correct, supervised learning requires labeled examples.
B) Unlabeled data – Used in unsupervised learning.
C) Random data – Randomness does not help models learn properly.
D) No data – Learning requires data.
4. Answer: B) Regression
Explanation:
A) Clustering – Unsupervised learning, groups similar data.
B) Regression – Correct, predicts continuous values using labeled data.
C) Association Rule Mining – Extracts rules from data (e.g., market basket analysis).
D) Dimensionality Reduction – Reduces dataset features.
5. Answer: A) A model that performs well on training data but poorly on new data
Explanation:
A) A model that performs well on training data but poorly on new data – Correct,
overfitting happens when a model memorizes training data instead of learning patterns.
B) A model that generalizes well to unseen data – This is good generalization, the
opposite of overfitting.
C) A model that is too simple to capture patterns – This is underfitting.
D) A model with fewer parameters than required – Overfitting usually occurs with too
many parameters.
6. Answer: A) To improve model accuracy on unseen data
Explanation:
A) To improve model accuracy on unseen data – Correct, cross-validation prevents
overfitting by evaluating the model on different data splits.
B) To increase training time – Cross-validation may increase training time, but that is
not its purpose.
C) To generate more data – It evaluates data but does not generate it.
D) To remove outliers from data – Not a primary function of cross-validation.
7. Answer: A) A tree-like structure for decision-making
Explanation:
A) A tree-like structure for decision-making – Correct, decision trees split data based
on conditions.
B) A database indexing technique – Not related to ML.
C) A deep learning algorithm – Decision trees are not deep learning.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
D) A clustering method – Decision trees are supervised learning, clustering is
unsupervised.
8. Answer: C) Reinforcement Learning
Explanation:
A) Supervised Learning – Uses labeled data.
B) Unsupervised Learning – Finds patterns in unlabeled data.
C) Reinforcement Learning – Correct, the agent learns by receiving rewards or
penalties.
D) Transfer Learning – Applies knowledge from one task to another.
9. Answer: B) Group similar data points
Explanation:
A) Predict future outcomes – This is the goal of supervised learning.
B) Group similar data points – Correct, unsupervised learning discovers patterns and
clusters.
C) Classify data based on labels – Requires labeled data (supervised learning).
D) Train a chatbot – Can involve various ML types, not just unsupervised learning.
10. Answer: B) Convolutional Neural Networks (CNNs)
Explanation:
A) Decision Trees – Good for tabular data, not images.
B) Convolutional Neural Networks (CNNs) – Correct, CNNs are specifically designed
for image recognition.
C) Support Vector Machines (SVMs) – Can process images but less effective than
CNNs.
D) K-Means Clustering – Used for clustering, not image processing.
11. Answer: B) Naïve Bayes
Explanation:
A) K-Means Clustering – Used for unsupervised clustering, not classification.
B) Naïve Bayes – Correct, this probabilistic classifier is highly effective for text-based
classification like spam filtering.
C) K-Nearest Neighbors (KNN) – Can be used, but is not as efficient as Naïve Bayes for
spam filtering.
D) Principal Component Analysis (PCA) – Used for dimensionality reduction, not
classification.
12. Answer: C) Binary Search
Explanation:
A) Random Forest – A widely used ML algorithm for classification and regression.
B) Support Vector Machine (SVM) – A powerful ML algorithm for classification tasks.
C) Binary Search – Correct, it is an algorithm for searching in sorted lists, not an ML
algorithm.
D) Gradient Boosting – A powerful ML algorithm used in boosted decision trees.
13. Answer: B) Increasing the number of features can lead to overfitting and computational
challenges
Explanation:

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
A) More features improve model performance – More features do not always improve
models; they may cause overfitting.
B) Increasing the number of features can lead to overfitting and computational
challenges – Correct, too many features make models complex and computationally
expensive.
C) Models with high dimensions always perform better – High dimensionality often
reduces model performance.
D) It is a feature selection technique – No, but feature selection helps combat the curse
of dimensionality.
14. Answer: C) Maximizing cumulative rewards
Explanation:
A) Minimizing loss function – More relevant to supervised learning.
B) Learning from labeled data – Used in supervised learning, whereas reinforcement
learning learns from actions and feedback.
C) Maximizing cumulative rewards – Correct, RL agents optimize actions to achieve
the highest reward over time.
D) Grouping similar data points – A goal of unsupervised learning (clustering).
15. Answer: C) Dimensionality reduction
Explanation:
A) Data classification – Classification uses models like Decision Trees and SVMs.
B) Data clustering – PCA is not used for clustering, but it can help improve clustering
performance.
C) Dimensionality reduction – Correct, PCA reduces the number of features while
preserving important information.
D) Hyperparameter tuning – Hyperparameter tuning optimizes model performance, not
dimensionality.
16. Answer: C) Regression
Explanation:
A) Classification – Linear Regression predicts continuous values, while classification
predicts discrete labels.
B) Clustering – Clustering groups similar data points, unrelated to Linear Regression.
C) Regression – Correct, Linear Regression models continuous relationships between
variables.
D) Reinforcement Learning – Not related to regression.
17. Answer: D) All of the above
Explanation:
A) ReLU (Rectified Linear Unit) – Most common in deep learning due to efficiency.
B) Sigmoid – Used in binary classification problems.
C) Tanh – Similar to Sigmoid but scales between -1 and 1.
D) All of the above – Correct, all are activation functions used in deep learning.
18. Answer: A) The error due to overly simplistic assumptions in a model
Explanation:
A) The error due to overly simplistic assumptions in a model – Correct, high bias
leads to underfitting.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
B) The ability of a model to generalize well – This refers to low bias and low variance
models.
C) The error caused by high variance – High variance leads to overfitting, not bias.
D) The process of selecting the best features – Feature selection is a separate process.
19. Answer: D) PageRank
Explanation:
A) Mean Squared Error (MSE) – Used in regression problems.
B) Cross-Entropy Loss – Used in classification problems.
C) Huber Loss – A robust loss function for regression.
D) PageRank – Correct, PageRank is an algorithm used by Google Search for ranking
web pages, not a loss function.
20. Answer: A) Uses multiple models to reduce variance
Explanation:
A) Uses multiple models to reduce variance – Correct, bagging combines predictions
from multiple models (e.g., Random Forest).
B) Applies boosting to weak learners – Boosting, not bagging, focuses on improving
weak models sequentially.
C) Creates a deep neural network – Bagging is unrelated to neural networks.
D) Selects the best individual model – Bagging uses multiple models, not just one.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination

Basics of AI and Robotics | Deep Learning, NLP, Computer Vision

1. Which of the following is a characteristic of Deep Learning?

A) It requires less data than traditional machine learning methods.
B) It involves a deep neural network with many layers.
C) It is only used for image processing tasks.
D) It is a type of supervised learning technique.
2. What is the purpose of an activation function in a neural network?
A) To convert the input into output
B) To introduce non-linearity into the network
C) To initialize the weights
D) To scale the output values
3. Which activation function is commonly used in the hidden layers of a neural
network?
A) Sigmoid
B) Tanh
C) ReLU
D) Softmax
4. What is a Convolutional Neural Network (CNN) best suited for?
A) Time series prediction
B) Image recognition and processing
C) Natural language processing
D) Structured data analysis
5. Which of the following is a common issue in training deep neural networks?
A) Overfitting
B) Underfitting
C) Lack of available training data
D) Both A and B
6. What is the vanishing gradient problem?
A) The gradient becomes too large during backpropagation
B) The gradient diminishes as it is propagated back through the layers
C) The learning rate is too high
D) The weights are initialized to very large values
7. What is the primary advantage of using Dropout in deep learning?
A) It reduces overfitting by randomly deactivating neurons during training
B) It increases the size of the dataset
C) It speeds up training by eliminating certain features
D) It improves convergence by reducing the learning rate

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
8. Which type of neural network is particularly useful for sequence data, such as time
series or text?
A) Convolutional Neural Network (CNN)
B) Recurrent Neural Network (RNN)
C) Feedforward Neural Network
D) Radial Basis Function Network
9. What is the primary function of the Softmax activation function?
A) To transform data into a binary output
B) To introduce non-linearity into the network
C) To normalize the output values into probabilities
D) To prevent the vanishing gradient problem
10. What is transfer learning in deep learning?
A) Training a model from scratch
B) Using a pre-trained model for a different but similar task
C) Transferring weights from one model to another
D) Fine-tuning a model by using a smaller dataset
11. Which of the following best describes a Neural Network?
A) A mathematical model used for predicting values
B) A network of interconnected neurons that mimics human brain functions
C) A method for training deep models using gradient descent
D) A type of machine learning algorithm based solely on decision trees
12. What does a Convolutional Neural Network (CNN) primarily perform?
A) Image recognition and processing
B) Time series forecasting
C) Natural language processing
D) Regression analysis
13. What does a Recurrent Neural Network (RNN) excel at?
A) Handling sequential data
B) Performing non-linear regression
C) Detecting edges in images
D) Solving multi-class classification problems
14. Which of the following is an essential property of Convolutional Neural Networks
(CNNs)?
A) They use layers of neurons that are fully connected.
B) They use convolutional layers to detect patterns in images.
C) They always work with sequential data like text or speech.
D) They are only used for binary classification tasks.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
15. What is the key advantage of using an LSTM (Long Short-Term Memory) over a
basic RNN?
A) LSTM has fewer parameters, making it more efficient.
B) LSTM is better at learning long-term dependencies in sequential data.
C) LSTM is easier to train than a CNN.
D) LSTM is faster at processing text data than RNN.
16. In the context of deep learning, what is “backpropagation”?
A) The method of adjusting the weights in a neural network during training.
B) A type of regularization technique to avoid overfitting.
C) A method for preprocessing input data before training.
D) A technique used to split data into training and testing sets.
17. What is a key feature of transformers in Natural Language Processing (NLP)?
A) Transformers use recurrent layers to capture dependencies.
B) They rely on attention mechanisms to process sequences.
C) Transformers are only useful for sentiment analysis tasks.
D) Transformers are exclusively used for image-related tasks.
18. Which of the following is a common challenge when training deep learning models?
A) The network becomes too small to learn meaningful features.
B) It is difficult to handle unstructured data.
C) Training deep networks can be very computationally expensive.
D) The learning rate is always too low to converge.
19. Which of the following is the primary function of the pooling layer in a CNN?
A) To reduce the spatial dimensions of the input
B) To enhance the edge-detection capabilities of the network
C) To apply an activation function to the input
D) To flatten the input for the fully connected layer
20. What is the vanishing gradient problem in deep neural networks?
A) Gradients become too large, causing instability.
B) Gradients become too small, hindering weight updates during backpropagation.
C) Gradients fail to converge, making training impossible.
D) Gradients become negative, causing incorrect weight updates.

Solutions

1. Answer: B) It involves a deep neural network with many layers.

Explanation:

A) It requires less data than traditional machine learning methods – Deep learning typically
requires large datasets to train models effectively, more than traditional machine learning
techniques.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
B) It involves a deep neural network with many layers – Correct, deep learning uses deep
neural networks with multiple layers to learn complex patterns.

C) It is only used for image processing tasks – Deep learning is not limited to image processing;
it can be applied to natural language processing, speech recognition, and more.

D) It is a type of supervised learning technique – Deep learning can be used for both
supervised and unsupervised learning tasks.

2. Answer: B) To introduce non-linearity into the network

Explanation:

A) To convert the input into output – The output is produced by a neuron, but the activation
function helps in non-linear transformations.

B) To introduce non-linearity into the network – Correct, activation functions allow the
network to learn complex patterns by introducing non-linearities.

C) To initialize the weights – Weights are initialized before training and are not directly linked to
activation functions.

D) To scale the output values – While some activation functions scale outputs, their main
function is to add non-linearity.

3. Answer: C) ReLU
Explanation:

A) Sigmoid – The sigmoid function is mainly used in output layers for binary classification tasks.

B) Tanh – Tanh can be used in hidden layers but is less efficient than ReLU in deep networks.

C) ReLU – Correct, ReLU (Rectified Linear Unit) is widely used in hidden layers because it
helps avoid the vanishing gradient problem and speeds up training.

D) Softmax – Softmax is typically used in the output layer for multi-class classification tasks, not
hidden layers.

4. Answer: B) Image recognition and processing

Explanation:

A) Time series prediction – Recurrent neural networks (RNNs) are more suitable for time series
tasks.

B) Image recognition and processing – Correct, CNNs are specifically designed to handle grid-
like data, such as images, and are widely used for computer vision tasks.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
C) Natural language processing – RNNs and transformers are typically used for language tasks,
not CNNs.

D) Structured data analysis – Structured data tasks are more suitable for other types of models
like decision trees or gradient boosting.

5. Answer: D) Both A and B

Explanation:

A) Overfitting – This occurs when the model learns too well on the training data and performs
poorly on unseen data.

B) Underfitting – This occurs when the model is too simple to capture the underlying patterns in
the data.

C) Lack of available training data – While this is an issue, it's not the only problem in deep
learning.

D) Both A and B – Correct, deep learning models can suffer from both overfitting and
underfitting depending on the complexity of the model and the amount of data.

6. Answer: B) The gradient diminishes as it is propagated back through the layers

Explanation:

A) The gradient becomes too large during backpropagation – This refers to the exploding
gradient problem, not the vanishing gradient problem.

B) The gradient diminishes as it is propagated back through the layers – Correct, the
vanishing gradient problem occurs when gradients become very small, making it difficult for the
network to learn in the earlier layers.

C) The learning rate is too high – A high learning rate can cause instability, but it is not related
to the vanishing gradient problem.

D) The weights are initialized to very large values – Large weight initialization may cause
other issues, but it is not the cause of the vanishing gradient problem.

7. Answer: A) It reduces overfitting by randomly deactivating neurons during training

Explanation:

A) It reduces overfitting by randomly deactivating neurons during training – Correct,

dropout helps prevent overfitting by randomly deactivating neurons in a layer during training,
forcing the network to learn more robust features.

B) It increases the size of the dataset – Dropout does not increase the size of the dataset; it is a
regularization technique.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
C) It speeds up training by eliminating certain features – Dropout can slow down training
slightly, but its primary role is regularization, not feature elimination.

D) It improves convergence by reducing the learning rate – Dropout does not directly affect
the learning rate.

8. Answer: B) Recurrent Neural Network (RNN)

Explanation:

A) Convolutional Neural Network (CNN) – CNNs are good for grid-like data (images), not for
sequential data.

B) Recurrent Neural Network (RNN) – Correct, RNNs are designed to process sequential data
by maintaining an internal state that captures information from previous steps in the sequence.

C) Feedforward Neural Network – These networks are not designed for sequence data.

D) Radial Basis Function Network – This network is used for function approximation and not
for sequential data.

9. Answer: C) To normalize the output values into probabilities

Explanation:

A) To transform data into a binary output – This is more closely related to the sigmoid
function.

B) To introduce non-linearity into the network – The Softmax function does not introduce non-
linearity; it normalizes the output.

C) To normalize the output values into probabilities – Correct, Softmax converts raw output
values (logits) into probabilities that sum to 1, often used in classification tasks.

D) To prevent the vanishing gradient problem – Softmax does not address the vanishing
gradient problem.

10. Answer: B) Using a pre-trained model for a different but similar task
Explanation:

A) Training a model from scratch – This is the opposite of transfer learning.

B) Using a pre-trained model for a different but similar task – Correct, transfer learning
involves taking a pre-trained model on one task and fine-tuning it for a new but related task.

C) Transferring weights from one model to another – This is part of transfer learning but not
the full definition.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
D) Fine-tuning a model by using a smaller dataset – Fine-tuning is part of transfer learning, but
the key concept is leveraging pre-trained models.

11. Answer: B) A network of interconnected neurons that mimics human brain functions
Explanation:

A) A mathematical model used for predicting values – While a neural network can predict
values, it is more than just a mathematical model.

B) A network of interconnected neurons that mimics human brain functions – Correct,

neural networks are designed to simulate the behavior of the human brain by using layers of
interconnected nodes (neurons).

C) A method for training deep models using gradient descent – Gradient descent is an
optimization algorithm, not a neural network itself.

D) A type of machine learning algorithm based solely on decision trees – Neural networks are
not based on decision trees; they belong to a different class of models.

12. Answer: A) Image recognition and processing

Explanation:

A) Image recognition and processing – Correct, CNNs are specifically designed for tasks like
image recognition and classification.

B) Time series forecasting – RNNs or LSTMs are better suited for time series data, not CNNs.

C) Natural language processing – While CNNs can be applied to text, RNNs and transformers
are more commonly used for NLP tasks.

D) Regression analysis – Regression tasks are typically performed by models like linear
regression, not CNNs.

13. Answer: A) Handling sequential data

Explanation:

A) Handling sequential data – Correct, RNNs are designed to handle sequential data such as
time series, text, or speech by maintaining a memory of previous inputs.

B) Performing non-linear regression – Non-linear regression is better handled by deep neural

networks, not specifically RNNs.

C) Detecting edges in images – Edge detection is typically done using CNNs, not RNNs.

D) Solving multi-class classification problems – RNNs can be used for classification tasks, but
they are not primarily known for multi-class classification.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
14. Answer: B) They use convolutional layers to detect patterns in images.
Explanation:

A) They use layers of neurons that are fully connected – CNNs typically use convolutional and
pooling layers, not just fully connected layers.

B) They use convolutional layers to detect patterns in images – Correct, CNNs are designed
to automatically detect spatial patterns in images using convolutional layers.

C) They always work with sequential data like text or speech – CNNs are often used with
images; RNNs and transformers are better for sequential data.

D) They are only used for binary classification tasks – CNNs can be used for both binary and
multi-class classification tasks.

15. Answer: B) LSTM is better at learning long-term dependencies in sequential data.

Explanation:

A) LSTM has fewer parameters, making it more efficient – LSTM models are more complex
than basic RNNs and typically have more parameters.

B) LSTM is better at learning long-term dependencies in sequential data – Correct, LSTMs

address the vanishing gradient problem in RNNs, enabling them to capture long-term
dependencies.

C) LSTM is easier to train than a CNN – LSTMs are generally harder to train than CNNs for
certain tasks, especially when compared to simpler models.

D) LSTM is faster at processing text data than RNN – LSTMs are usually slower due to their
complexity.

16. Answer: A) The method of adjusting the weights in a neural network during training.
Explanation:

A) The method of adjusting the weights in a neural network during training – Correct,
backpropagation is the process of computing the gradient of the loss function with respect to each
weight in the network, and adjusting the weights to minimize the loss.

B) A type of regularization technique to avoid overfitting – Regularization techniques like

dropout help prevent overfitting, but backpropagation is not one of them.

C) A method for preprocessing input data before training – Preprocessing typically includes
normalization, encoding, etc., but not backpropagation.

D) A technique used to split data into training and testing sets – Data splitting is done before
training and is not related to backpropagation.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
17. Answer: B) They rely on attention mechanisms to process sequences.
Explanation:

A) Transformers use recurrent layers to capture dependencies – Transformers do not use

recurrent layers; instead, they rely on attention mechanisms for capturing dependencies.

B) They rely on attention mechanisms to process sequences – Correct, transformers use

attention mechanisms, allowing them to handle long-range dependencies without the limitations
of RNNs.

C) Transformers are only useful for sentiment analysis tasks – Transformers can be applied to
a wide range of NLP tasks, not just sentiment analysis.

D) Transformers are exclusively used for image-related tasks – While transformers have been
applied to vision tasks (e.g., Vision Transformers), they were originally designed for NLP.

18. Answer: C) Training deep networks can be very computationally expensive.

Explanation:

A) The network becomes too small to learn meaningful features – This is not typically a
challenge when training deep networks; it’s more of an issue when using too simple models.

B) It is difficult to handle unstructured data – Deep learning models are particularly good at
handling unstructured data like images, text, and audio.

C) Training deep networks can be very computationally expensive – Correct, deep learning
models require large amounts of data and computational resources (e.g., GPUs) to train
effectively.

D) The learning rate is always too low to converge – The learning rate can cause issues if it’s
too high or too low, but it’s not always too low.

19. Answer: A) To reduce the spatial dimensions of the input

Explanation:

A) To reduce the spatial dimensions of the input – Correct, the pooling layer (e.g., max
pooling) reduces the spatial dimensions, thus reducing computational complexity and helping the
network generalize better.

B) To enhance the edge-detection capabilities of the network – Edge detection is typically

performed by convolutional layers, not pooling layers.

C) To apply an activation function to the input – Activation functions are applied to the output
of neurons, not in the pooling layers.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
D) To flatten the input for the fully connected layer – Flattening is done after the convolutional
and pooling layers, before passing to fully connected layers.

20. Answer: B) Gradients become too small, hindering weight updates during
backpropagation.
Explanation:

A) Gradients become too large, causing instability – This describes the exploding gradient
problem, not the vanishing gradient problem.

B) Gradients become too small, hindering weight updates during backpropagation –

Correct, the vanishing gradient problem occurs when gradients become extremely small, causing
weights in earlier layers to update very slowly.

C) Gradients fail to converge, making training impossible – While vanishing gradients can
slow down training, they don’t necessarily make it impossible.

D) Gradients become negative, causing incorrect weight updates – Gradients can be negative,
but it’s not the primary concern in the vanishing gradient problem.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination

Basics of AI and Robotics | Deep Learning, NLP, Computer

Vision
1. Which activation function is most commonly used in hidden layers of deep networks?
A) Sigmoid
B) ReLU (Rectified Linear Unit)
C) Tanh
D) Softmax
2. What is the key difference between a traditional neural network and a convolutional
neural network (CNN)?
A) CNNs use fully connected layers only
B) CNNs are designed to handle image data using convolutional layers
C) CNNs can only be used for regression tasks
D) CNNs do not require any form of pooling
3. Which of the following is a major advantage of Long Short-Term Memory (LSTM)
networks over vanilla RNNs?
A) LSTMs are faster to train than RNNs
B) LSTMs can capture long-term dependencies in sequences
C) LSTMs are only used for image data
D) LSTMs do not require a loss function during training
4. What is the role of a softmax activation function in the output layer of a neural
network?
A) To introduce non-linearity into the output
B) To produce a probability distribution for classification tasks
C) To normalize the input data
D) To calculate the error during backpropagation
5. What is the main disadvantage of using the sigmoid activation function?
A) It suffers from the vanishing gradient problem
B) It cannot be used in multi-class classification problems
C) It is computationally expensive to calculate
D) It outputs values that are not between 0 and 1
6. Which of the following is the primary use of Natural Language Processing (NLP)?
A) To detect patterns in images
B) To analyze and process human language
C) To perform regression on numerical data
D) To build recommendation systems
7. What is the main advantage of using transformers in NLP tasks?
A) They require less training data than traditional models
B) They are easier to interpret and explain
C) They capture long-range dependencies without using RNNs
D) They are less computationally expensive than CNNs
8. In a deep learning network, what is the purpose of dropout during training?

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
A) To increase the learning rate
B) To randomly deactivate neurons to prevent overfitting
C) To normalize the inputs to the network
D) To reduce the depth of the network
Answer: B) To randomly deactivate neurons to prevent overfitting
9. In a Convolutional Neural Network (CNN), what is the purpose of a pooling layer?
A) To reduce the dimensionality of the input image
B) To increase the depth of the feature maps
C) To apply the convolution operation on the feature maps
D) To introduce non-linearity in the network
10. What is the main advantage of ReLU over the sigmoid activation function in deep
learning models?
A) ReLU has a bounded output, while sigmoid does not
B) ReLU prevents the vanishing gradient problem
C) ReLU is computationally slower than sigmoid
D) ReLU can be used for binary classification only
11. What is an advantage of using a batch normalization layer in a neural network?
A) It reduces the size of the model
B) It speeds up the convergence of the network
C) It removes the need for an activation function
D) It prevents overfitting by adding noise to the data
12. In Recurrent Neural Networks (RNNs), what problem occurs due to the vanishing
gradient issue?
A) RNNs fail to learn short-term dependencies
B) RNNs become overfitted very quickly
C) RNNs struggle to learn long-term dependencies
D) RNNs do not provide predictions during training
13. Which of the following is a primary advantage of Transformer models over RNNs in
NLP tasks?
A) Transformers require less training data
B) Transformers process sequences in parallel rather than sequentially
C) Transformers are simpler to implement than RNNs
D) Transformers can only handle fixed-length inputs
14. In which type of neural network are Long Short-Term Memory (LSTM) cells
primarily used?
A) Feedforward Neural Networks (FNN)
B) Convolutional Neural Networks (CNN)
C) Recurrent Neural Networks (RNN)
D) Radial Basis Function Networks (RBFN)
15. What is the primary function of the attention mechanism in Transformers?
A) To determine which words in a sequence are important for prediction
B) To prevent overfitting during training
C) To convert the input sequence into a fixed-size representation
D) To generate feature maps for input data

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
16. In a neural network, what is the role of the backpropagation algorithm?
A) To initialize the weights of the network
B) To calculate the loss or error between predicted and actual values
C) To adjust the weights in the network based on the loss
D) To apply the activation function to the outputs
17. What is the main reason that Recurrent Neural Networks (RNNs) are not well-suited
for very long sequences?
A) RNNs do not have enough neurons to process long sequences
B) RNNs suffer from the vanishing and exploding gradient problems
C) RNNs are only suitable for fixed-length sequences
D) RNNs are computationally inefficient for long sequences
18. What is a common application of Natural Language Processing (NLP) in real-world
scenarios?
A) Image classification
B) Speech-to-text conversion
C) Object detection in images
D) Predictive maintenance in machines

Solutions
1. Answer: C) ReLU
Explanation:
A) Sigmoid – The sigmoid function is mainly used in output layers for binary classification
tasks.
B) Tanh – Tanh can be used in hidden layers but is less efficient than ReLU in deep networks.
C) ReLU – Correct, ReLU (Rectified Linear Unit) is widely used in hidden layers because it
helps avoid the vanishing gradient problem and speeds up training.
D) Softmax – Softmax is typically used in the output layer for multi-class classification tasks,
not hidden layers.
2. Answer: B) CNNs are designed to handle image data using convolutional layers
Explanation:
A) CNNs use fully connected layers only – CNNs typically use convolutional layers and
pooling layers, with fully connected layers towards the end.
B) CNNs are designed to handle image data using convolutional layers – Correct, CNNs
are specialized for image processing, utilizing convolutional layers to detect features.
C) CNNs can only be used for regression tasks – CNNs can be used for both classification
and regression tasks, not just regression.
D) CNNs do not require any form of pooling – Pooling is an essential component of CNNs
to reduce the spatial dimensions and computational complexity.
3. Answer: B) LSTMs can capture long-term dependencies in sequences
Explanation:
A) LSTMs are faster to train than RNNs – LSTMs are generally slower to train due to their
more complex architecture.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
B) LSTMs can capture long-term dependencies in sequences – Correct, LSTMs are
designed to handle long-term dependencies better than vanilla RNNs by using gating
mechanisms.
C) LSTMs are only used for image data – LSTMs are primarily used for sequential data
(e.g., time series, text), not image data.
D) LSTMs do not require a loss function during training – Like any other neural network,
LSTMs require a loss function for training.
4. Answer: B) To produce a probability distribution for classification tasks
Explanation:
A) To introduce non-linearity into the output – Non-linearity is introduced by activation
functions in hidden layers, not by softmax.
B) To produce a probability distribution for classification tasks – Correct, softmax
converts the raw output into a probability distribution for multi-class classification.
C) To normalize the input data – Normalization of input data is typically done during
preprocessing, not by the softmax function.
D) To calculate the error during backpropagation – Error calculation is done by the loss
function, not by the softmax function.
5. Answer: A) It suffers from the vanishing gradient problem
Explanation:
A) It suffers from the vanishing gradient problem – Correct, the sigmoid function can
cause gradients to shrink to very small values during backpropagation, leading to slow
learning in deep networks.
B) It cannot be used in multi-class classification problems – Sigmoid can be used in binary
classification, while softmax is more suitable for multi-class classification.
C) It is computationally expensive to calculate – Sigmoid is relatively inexpensive to
compute, but it may cause issues with vanishing gradients.
D) It outputs values that are not between 0 and 1 – The sigmoid function always outputs
values between 0 and 1, making it suitable for binary classification.
6. Answer: B) To analyze and process human language
Explanation:
A) To detect patterns in images – This is the role of CNNs, not NLP.
B) To analyze and process human language – Correct, NLP focuses on tasks like text
analysis, speech recognition, and language understanding.
C) To perform regression on numerical data – Regression tasks are usually handled by
machine learning models, not by NLP techniques.
D) To build recommendation systems – Recommendation systems are typically built using
collaborative filtering, not directly with NLP.
7. Answer: C) They capture long-range dependencies without using RNNs
Explanation:
A) They require less training data than traditional models – Transformers may require
large amounts of training data, especially for large models.
B) They are easier to interpret and explain – Transformers are more complex and harder
to interpret than some other models.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
C) They capture long-range dependencies without using RNNs – Correct, transformers
use attention mechanisms to model long-range dependencies in data without relying on
RNNs.
D) They are less computationally expensive than CNNs – Transformers tend to be
computationally expensive, particularly for large-scale models.
8. Answer: B) To randomly deactivate neurons to prevent overfitting
Explanation:
A) To increase the learning rate – Dropout does not affect the learning rate directly.
B) To randomly deactivate neurons to prevent overfitting – Correct, dropout randomly
drops neurons during training to prevent overfitting by forcing the network to generalize
better.
C) To normalize the inputs to the network – Input normalization is done through other
techniques like batch normalization or standardization.
D) To reduce the depth of the network – Dropout does not affect the depth of the network;
it only affects the neurons active during training.
9. Answer: A) To reduce the dimensionality of the input image
Explanation:
A) To reduce the dimensionality of the input image – Correct, pooling layers, such as max
pooling, reduce the spatial dimensions (height and width) of the input, helping to decrease
computational cost and reduce overfitting.
B) To increase the depth of the feature maps – This is not the purpose of pooling; pooling
reduces spatial dimensions, not depth.
C) To apply the convolution operation on the feature maps – The convolution operation is
performed by convolutional layers, not pooling layers.
D) To introduce non-linearity in the network – Non-linearity is typically introduced by
activation functions, not pooling layers.
10. Answer: B) ReLU prevents the vanishing gradient problem
Explanation:
A) ReLU has a bounded output, while sigmoid does not – ReLU has an unbounded output
(values range from 0 to infinity), while sigmoid has a bounded output between 0 and 1.
B) ReLU prevents the vanishing gradient problem – Correct, ReLU helps to mitigate the
vanishing gradient problem by allowing gradients to flow through the network more
effectively.
C) ReLU is computationally slower than sigmoid – ReLU is computationally faster than
sigmoid because it involves a simple thresholding operation (activation is just max(0, x)).
D) ReLU can be used for binary classification only – ReLU is typically used in hidden
layers, while sigmoid is often used in the output layer for binary classification.
11. Answer: B) It speeds up the convergence of the network
Explanation:
A) It reduces the size of the model – Batch normalization doesn’t reduce the size of the
model; it normalizes the activations across mini-batches.
B) It speeds up the convergence of the network – Correct, batch normalization stabilizes
the learning process by normalizing activations, which allows the network to converge faster.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
C) It removes the need for an activation function – Batch normalization does not replace
activation functions; it works alongside them.
D) It prevents overfitting by adding noise to the data – This is more akin to dropout; batch
normalization stabilizes training but doesn’t add noise.
12. Answer: C) RNNs struggle to learn long-term dependencies
Explanation:
A) RNNs fail to learn short-term dependencies – RNNs are good at capturing short-term
dependencies; the vanishing gradient issue primarily affects long-term dependencies.
B) RNNs become overfitted very quickly – While overfitting can be an issue in any model,
the vanishing gradient problem specifically prevents the model from learning long-term
patterns.
C) RNNs struggle to learn long-term dependencies – Correct, the vanishing gradient
problem causes the gradients to diminish during backpropagation, making it difficult for
RNNs to learn long-term dependencies.
D) RNNs do not provide predictions during training – RNNs do provide predictions during
training; the issue lies in the difficulty of learning long-term dependencies.
13. Answer: B) Transformers process sequences in parallel rather than sequentially
Explanation:
A) Transformers require less training data – Transformers often require large datasets,
especially when using models like BERT or GPT.
B) Transformers process sequences in parallel rather than sequentially – Correct, one of
the key advantages of transformers is their ability to process sequences in parallel,
improving training efficiency.
C) Transformers are simpler to implement than RNNs – Transformers are more complex
to implement compared to RNNs due to the attention mechanism.
D) Transformers can only handle fixed-length inputs – Transformers can handle variable-
length inputs due to their attention mechanisms, unlike RNNs that are typically limited by
sequence length.
14. Answer: C) Recurrent Neural Networks (RNN)
Explanation:
A) Feedforward Neural Networks (FNN) – LSTM cells are not used in feedforward
networks; they are specifically designed for sequential data.
B) Convolutional Neural Networks (CNN) – CNNs focus on processing spatial data (like
images) and don’t use LSTMs.
C) Recurrent Neural Networks (RNN) – Correct, LSTM cells are an advanced form of
RNNs, designed to overcome the vanishing gradient problem and capture long-term
dependencies in sequential data.
D) Radial Basis Function Networks (RBFN) – LSTMs are not used in RBFNs, which use
radial basis functions for classification.
15. Answer: A) To determine which words in a sequence are important for prediction
Explanation:
A) To determine which words in a sequence are important for prediction – Correct, the
attention mechanism assigns different attention scores to different words in a sequence,
allowing the model to focus on the most relevant words.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
B) To prevent overfitting during training – While attention improves model performance,
it does not directly prevent overfitting (which is usually handled by regularization techniques
like dropout).
C) To convert the input sequence into a fixed-size representation – This is typically done
by encoding methods, but not by attention directly.
D) To generate feature maps for input data – Feature maps are typically generated by
convolutional layers, not by attention.
16. Answer: C) To adjust the weights in the network based on the loss
Explanation:
A) To initialize the weights of the network – Weight initialization is typically done at the
beginning of the training process, not by backpropagation.
B) To calculate the loss or error between predicted and actual values – Loss calculation is
part of the forward pass, but backpropagation’s primary role is adjusting weights.
C) To adjust the weights in the network based on the loss – Correct, backpropagation
calculates the gradients of the loss with respect to the weights and adjusts the weights to
minimize the error.
D) To apply the activation function to the outputs – The activation function is applied
during the forward pass, not by backpropagation.
17. Answer: B) RNNs suffer from the vanishing and exploding gradient problems
Explanation:
A) RNNs do not have enough neurons to process long sequences – The issue with RNNs is
not the number of neurons, but how they handle long-term dependencies.
B) RNNs suffer from the vanishing and exploding gradient problems – Correct, RNNs
struggle to maintain gradients over long sequences, which leads to issues with learning long-
term dependencies.
C) RNNs are only suitable for fixed-length sequences – RNNs can handle sequences of
variable lengths.
D) RNNs are computationally inefficient for long sequences – While RNNs can be
computationally inefficient, the primary issue is related to gradients, not computation itself.
18. Answer: B) Speech-to-text conversion
Explanation:
A) Image classification – Image classification is a computer vision task, not an NLP task.
B) Speech-to-text conversion – Correct, speech-to-text conversion is a classic example of
an NLP task, where audio is transcribed into text.
C) Object detection in images – Object detection is part of computer vision, not NLP.
D) Predictive maintenance in machines – Predictive maintenance is an application of
machine learning, but it is not specifically related to NLP.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination

Basics of AI and Robotics | Fundamentals of Robotics

1. What is the primary function of an actuator in a robot?
To process the sensory data
B) To generate the movement of the robot
C) To provide power to the robot's system
D) To control the robot’s software
2. Which of the following is a common type of sensor used in robotics?
Microprocessor
B) Photodetector
C) LED
D) Microphone
3. What is the primary function of a robot's controller?
To process sensory data
B) To control the movement of actuators
C) To store the robot's software
D) To interact with human operators
4. What is the primary difference between an autonomous robot and a remote-
controlled robot?
Autonomous robots require a human operator
B) Autonomous robots can make decisions based on sensors
C) Remote-controlled robots can make decisions
D) Remote-controlled robots can operate without sensors
5. Which of the following is an example of an industrial robot?
Roomba vacuum cleaner
B) Robotic arm in an automotive assembly line
C) Drone for aerial photography
D) Humanoid robot for customer service
6. What is a key challenge in robot programming?
Designing the robot’s hardware
B) Ensuring the robot can handle uncertainty in its environment
C) Manufacturing the robot’s body
D) Developing the robot’s power supply
7. What is an example of a robot used for medical applications?
Boston Dynamics' Spot robot
B) A robotic surgical system like the Da Vinci Surgical System
C) A drone used for package delivery
D) A robotic vacuum cleaner
8. Which of the following describes the concept of “robot perception”?
The robot’s ability to move
B) The robot’s ability to understand and interpret sensory input
C) The robot’s ability to solve complex mathematical problems
D) The robot’s ability to communicate with humans

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
9. What is the role of a robot's "end effector"?
To control the robot’s movements
B) To provide sensory input
C) To perform tasks such as picking up objects or welding
D) To power the robot
10. What is the difference between a manipulator and an end effector in robotics?
A manipulator is a type of sensor
B) A manipulator moves objects, and the end effector holds or interacts with
objects
C) The manipulator powers the robot, and the end effector is responsible for
moving it
D) The manipulator stores the robot’s software, and the end effector performs
tasks
11. Which of the following is a key feature of autonomous robots?
They require constant human control
B) They can operate based on pre-programmed instructions only
C) They use sensors and algorithms to make decisions
D) They cannot interact with the environment
12. What is the main purpose of a robotic arm in an industrial setting?
To assist with complex computations
B) To perform repetitive or dangerous tasks
C) To design the robot’s software
D) To act as a sensory input device
13. What is one advantage of using robots in manufacturing?
Robots can replace human workers entirely
B) Robots can work without any maintenance
C) Robots can increase precision and efficiency
D) Robots can perform complex creative tasks
14. Which type of control system is most commonly used for robots with high
precision requirements?
Open-loop control
B) Closed-loop control
C) Manual control
D) Supervisory control
15. Which of the following is a characteristic of an industrial robot?
It has a humanoid form
B) It can perform multiple tasks autonomously
C) It requires constant manual operation
D) It is designed to perform a limited set of repetitive tasks
16. What is an example of a robot that uses reinforcement learning?
Robotic arm on an assembly line
B) Autonomous car navigating through traffic

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
C) Robotic vacuum cleaner following a pattern
D) Industrial robot performing quality checks
17. In robotic systems, what is the function of the "feedback loop"?
To improve the robot's programming
B) To update the robot's software
C) To continuously adjust the robot's actions based on sensor data
D) To control the robot’s power supply
18. Which of the following is an example of a mobile robot?
A robotic arm on an assembly line
B) A drone that delivers packages
C) A robotic surgical assistant
D) A stationary robotic vacuum cleaner
19. What is "inverse kinematics" in the context of robotics?
The study of robot joints
B) The mathematical approach to determine the required joint movements for a
desired end effector position
C) The process of converting analog data into digital
D) The design of robot software
20. What is one of the main uses of robotic vision systems?
To communicate with humans
B) To detect and interpret visual data from the environment
C) To process sensory input from tactile sensors
D) To store robotic programs
21. What type of actuator is commonly used in robotics for precise control of
movement?
Pneumatic actuator
B) Hydraulic actuator
C) Electric motor
D) Thermal actuator
22. What is the role of a robot’s "controller"?
To move the robot’s body
B) To provide power to the robot
C) To process input from sensors and control actuators
D) To store the robot’s program
23. In robotics, what is the purpose of a "gripper"?
To move the robot
B) To provide sensory feedback
C) To manipulate objects
D) To calculate robot movement
24. What is an example of an application of robot swarm technology?
A robot assembling components on a production line
B) Drones working together to monitor a large area

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
C) A robot vacuum cleaning a house
D) A robot performing surgery
25. Which of the following technologies is essential for autonomous robots to
perceive their environment?
GPS
B) Sensors
C) Actuators
D) Processors
26. What type of robot is used in medical surgeries?
Autonomous robot
B) Humanoid robot
C) Surgical robot
D) Industrial robot
27. What is "simultaneous localization and mapping" (SLAM) in robotics?
A method for storing a robot’s environment data
B) A technique used to avoid obstacles
C) A method for mapping an environment and tracking the robot’s location within
it
D) A method to program robot movements
28. In robotics, what is a "sensor fusion" system?
A system that combines multiple sensors’ data to improve the robot’s
understanding of the environment
B) A system that fuses the robot’s body parts together
C) A system that combines the robot’s power sources
D) A system that reduces the weight of the robot
29. What is the key difference between a humanoid robot and a mobile robot?
Humanoid robots can move, while mobile robots cannot
B) Humanoid robots are designed to look like humans, while mobile robots are
designed to move autonomously
C) Humanoid robots are stationary, while mobile robots are capable of movement
D) Mobile robots have more sensors than humanoid robots
30. Which of the following is an example of a robot that performs a repetitive task?
A surgical robot
B) A robotic arm on an assembly line
C) A robot vacuum cleaner
D) A robot performing in a play
31. What is the primary function of the end-effector in a robot?
To provide power to the robot
B) To control the robot's movement
C) To perform tasks or manipulate objects
D) To process the robot's sensory data

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
32. What is the purpose of a robot's "vision system"?
To provide power to the robot
B) To track the robot’s location
C) To process sensory data for the robot's navigation
D) To store the robot's program
33. What does the term "actuator" refer to in robotics?
A part of the robot that processes input data
B) A system that controls the robot’s sensors
C) A device that performs physical actions based on commands
D) A part that gives feedback to the robot’s controller
34. Which of the following is NOT a common type of robot actuator?
Electric motor
B) Hydraulic cylinder
C) Pneumatic piston
D) GPS system
35. In the context of robotic arms, what does "degrees of freedom" (DOF) refer to?
The number of sensors on the robot
B) The number of independent movements the robot can make
C) The number of actuators the robot has
D) The number of tasks the robot can perform

Solutions
1. Answer: B) To generate the movement of the robot
Explanation:

A) To process the sensory data – This is the role of sensors in robotics, not actuators.

B) To generate the movement of the robot – Correct, actuators are responsible for converting
electrical energy into mechanical movement, driving the robot’s motion.

C) To provide power to the robot's system – Power is provided by the power supply or battery,
not by actuators.

D) To control the robot’s software – Software is controlled by a computer or controller, not by

actuators.

2. Answer: B) Photodetector
Explanation:

A) Microprocessor – Microprocessors are used for computation and control, not for sensing.

B) Photodetector – Correct, photodetectors are commonly used to sense light or objects and are
essential for vision systems in robots.

C) LED – LEDs are used for display purposes, not for sensing.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
D) Microphone – Microphones can be used to sense sound, but they are not as common as other
sensors like photodetectors in most robotic applications.

3. Answer: B) To control the movement of actuators

Explanation:

A) To process sensory data – Sensory data is processed by a robot’s sensory processing system,
not the controller itself.

B) To control the movement of actuators – Correct, the controller processes inputs and sends
signals to the actuators to perform specific movements.

C) To store the robot's software – The software is stored in memory, but the primary function of
the controller is not storage.

D) To interact with human operators – Human interaction is typically handled through an

interface, not directly by the controller.

4. Answer: B) Autonomous robots can make decisions based on sensors

Explanation:

A) Autonomous robots require a human operator – This is incorrect; autonomous robots are
capable of operating without a human operator.

B) Autonomous robots can make decisions based on sensors – Correct, autonomous robots
use sensors to perceive their environment and make decisions based on that input.

C) Remote-controlled robots can make decisions – Remote-controlled robots rely on human

input, not on autonomous decision-making.

D) Remote-controlled robots can operate without sensors – Remote-controlled robots often

rely on sensors to ensure safe operation, though not as extensively as autonomous robots.

5. Answer: B) Robotic arm in an automotive assembly line

Explanation:

A) Roomba vacuum cleaner – Roomba is a service robot, not specifically an industrial robot.

B) Robotic arm in an automotive assembly line – Correct, robotic arms are commonly used in
industrial environments, such as automotive assembly lines, for tasks like welding and assembling
parts.

C) Drone for aerial photography – Drones are often used for recreational or commercial
purposes, not typically in industrial settings.

D) Humanoid robot for customer service – Humanoid robots are generally used in service
sectors, not industrial manufacturing.

6. Answer: B) Ensuring the robot can handle uncertainty in its environment

Explanation:

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
A) Designing the robot’s hardware – While important, hardware design is a separate issue from
programming.

B) Ensuring the robot can handle uncertainty in its environment – Correct, uncertainty (e.g.,
unexpected obstacles or changing conditions) is a significant challenge in robot programming,
especially for autonomous robots.

C) Manufacturing the robot’s body – Manufacturing is part of hardware design, not directly
related to programming.

D) Developing the robot’s power supply – Power supply design is important but is not a direct
programming challenge.

7. Answer: B) A robotic surgical system like the Da Vinci Surgical System

Explanation:

A) Boston Dynamics' Spot robot – Spot is a general-purpose robot designed for mobility and
inspection, not specifically medical applications.

B) A robotic surgical system like the Da Vinci Surgical System – Correct, the Da Vinci system
is an example of a robot used for minimally invasive surgeries.

C) A drone used for package delivery – Drones are used in logistics, not typically for medical
applications.

D) A robotic vacuum cleaner – This is a service robot, not used for medical purposes.

8. Answer: B) The robot’s ability to understand and interpret sensory input

Explanation:

A) The robot’s ability to move – This is related to the robot’s actuation, not its perception.

B) The robot’s ability to understand and interpret sensory input – Correct, robot perception
involves interpreting sensory data (e.g., from cameras, microphones, or touch sensors) to
understand the environment.

C) The robot’s ability to solve complex mathematical problems – This is a task for the robot’s
computing system, not directly related to perception.

D) The robot’s ability to communicate with humans – Communication is a form of interaction,

not perception.

9. Answer: C) To perform tasks such as picking up objects or welding

Explanation:

A) To control the robot’s movements – This is typically the role of the robot's actuators, not the
end effector.

B) To provide sensory input – Sensory input is handled by the sensors, not the end effector.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
C) To perform tasks such as picking up objects or welding – Correct, the end effector is the
part of the robot (such as a gripper, tool, or welding torch) that directly interacts with the
environment to perform specific tasks.

D) To power the robot – Power is supplied by the power system, not the end effector.

10. Answer: B) A manipulator moves objects, and the end effector holds or interacts with
objects
Explanation:

A) A manipulator is a type of sensor – A manipulator is a mechanical arm, not a sensor.

B) A manipulator moves objects, and the end effector holds or interacts with objects –
Correct, the manipulator is the robot’s arm or limb that moves the end effector (like a gripper or
tool) to perform tasks.

C) The manipulator powers the robot, and the end effector is responsible for moving it – The
manipulator does not power the robot, and the end effector does not move the robot.

D) The manipulator stores the robot’s software, and the end effector performs tasks – This is
incorrect; software storage is handled by memory, not the manipulator.

11. Answer: C) They use sensors and algorithms to make decisions

Explanation:

A) They require constant human control – Autonomous robots are designed to operate without
constant human control.

B) They can operate based on pre-programmed instructions only – Autonomous robots also
use sensors and make real-time decisions, not just rely on pre-programmed instructions.

C) They use sensors and algorithms to make decisions – Correct, autonomous robots use their
sensors to perceive the environment and make decisions based on algorithms.

D) They cannot interact with the environment – This is incorrect; autonomous robots interact
with the environment via sensors and actuators.

12. Answer: B) To perform repetitive or dangerous tasks

Explanation:

A) To assist with complex computations – A robotic arm does not perform complex
computations; this is done by the robot's controller or computer system.

B) To perform repetitive or dangerous tasks – Correct, robotic arms are often used in
industrial settings to perform repetitive tasks like welding or packaging, or dangerous tasks that
would be hazardous to humans.

C) To design the robot’s software – Software design is done by engineers, not the robotic arm.

D) To act as a sensory input device – The robotic arm does not serve as a sensor; sensors are
separate components in robotics.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
13. Answer: C) Robots can increase precision and efficiency
Explanation:

A) Robots can replace human workers entirely – This is not always true; robots can assist but
may not fully replace human workers, especially for tasks requiring creativity or decision-making.

B) Robots can work without any maintenance – Robots do require regular maintenance to
ensure optimal performance.

C) Robots can increase precision and efficiency – Correct, robots can perform tasks with high
precision and efficiency, reducing errors and increasing production rates.

D) Robots can perform complex creative tasks – While robots are excellent at repetitive and
precise tasks, creativity is still largely a human domain.

14. Answer: B) Closed-loop control

Explanation:

A) Open-loop control – Open-loop control systems are not ideal for precision tasks as they lack
feedback mechanisms.

B) Closed-loop control – Correct, closed-loop systems use feedback from sensors to

continuously adjust and correct the robot’s movements, providing high precision.

C) Manual control – Manual control requires human intervention, not suitable for high-precision
tasks.

D) Supervisory control – Supervisory control involves human oversight, not direct control of the
robot’s movements.

15. Answer: D) It is designed to perform a limited set of repetitive tasks

Explanation:

A) It has a humanoid form – Industrial robots typically do not have humanoid forms; they are
often designed with mechanical arms or other task-specific structures.

B) It can perform multiple tasks autonomously – While some robots can perform multiple
tasks, industrial robots are generally designed to perform specific repetitive tasks.

C) It requires constant manual operation – Industrial robots are typically autonomous,

requiring little human intervention once programmed.

D) It is designed to perform a limited set of repetitive tasks – Correct, industrial robots are
optimized for tasks like welding, packaging, and assembling in a repetitive, precise manner.

16. Answer: B) Autonomous car navigating through traffic

Explanation:

A) Robotic arm on an assembly line – Typically uses predefined programming, not

reinforcement learning.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
B) Autonomous car navigating through traffic – Correct, reinforcement learning is often used
for robots like self-driving cars, which must continuously learn from their environment and adapt
their behavior.

C) Robotic vacuum cleaner following a pattern – Generally follows programmed patterns, not
reinforcement learning.

D) Industrial robot performing quality checks – Industrial robots usually follow pre-
programmed instructions rather than using reinforcement learning.

17. Answer: C) To continuously adjust the robot's actions based on sensor data
Explanation:

A) To improve the robot's programming – Feedback loops help in adjusting actions but are not
directly involved in software programming.

B) To update the robot's software – Feedback loops are not responsible for software updates,
which are performed separately.

C) To continuously adjust the robot's actions based on sensor data – Correct, feedback loops
allow robots to adjust their behavior in real-time based on data received from sensors, improving
task performance.

D) To control the robot’s power supply – Power supply management is handled separately and
not by feedback loops.

18. Answer: B) A drone that delivers packages

Explanation:

A) A robotic arm on an assembly line – A robotic arm is stationary and does not move around
autonomously.

B) A drone that delivers packages – Correct, drones are mobile robots designed to move
through the air autonomously to deliver packages.

C) A robotic surgical assistant – This type of robot typically operates in a fixed environment,
such as an operating room.

D) A stationary robotic vacuum cleaner – While mobile within a house, it is not considered a
"mobile robot" in the context of autonomy and operational range like a drone.

19. Answer: B) The mathematical approach to determine the required joint movements
for a desired end effector position
Explanation:

A) The study of robot joints – Inverse kinematics specifically involves calculating the
movements required by the joints, not studying them.

B) The mathematical approach to determine the required joint movements for a desired end
effector position – Correct, inverse kinematics helps calculate the joint angles needed for the
robot's end effector to reach a target position.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
C) The process of converting analog data into digital – This is related to signal processing, not
inverse kinematics.

D) The design of robot software – Software design is a broader topic and is not specific to
inverse kinematics.

20. Answer: B) To detect and interpret visual data from the environment
Explanation:

A) To communicate with humans – Communication is typically handled by other interfaces, not

by vision systems.

B) To detect and interpret visual data from the environment – Correct, robotic vision systems
use cameras and sensors to detect objects, distances, and surroundings for decision-making.

C) To process sensory input from tactile sensors – Tactile sensors are separate from vision
systems and are used for touch-based feedback.

D) To store robotic programs – Program storage is handled by the robot's memory, not by its
vision system.

21. Answer: C) Electric motor

Explanation:

A) Pneumatic actuator – Pneumatic actuators are used for rapid movement but are less precise
than electric motors.

B) Hydraulic actuator – Hydraulic actuators provide strong force but are generally used in
heavy-duty applications and are less precise for fine movements.

C) Electric motor – Correct, electric motors are commonly used in robotics for precise control
of movement, as they can be easily controlled with high accuracy.

D) Thermal actuator – Thermal actuators are not widely used in robotics; they operate based on
temperature changes and are not suitable for precise control.

22. Answer: C) To process input from sensors and control actuators

Explanation:

A) To move the robot’s body – The controller does not directly move the robot’s body but sends
commands to actuators to control movement.

B) To provide power to the robot – Power is provided by the robot's power supply, not the
controller.

C) To process input from sensors and control actuators – Correct, the controller is the brain of
the robot, processing sensor data and controlling the actuators based on that input.

D) To store the robot’s program – The program storage is usually handled by the robot's
memory or processor, not the controller.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
23. Answer: C) To manipulate objects
Explanation:

A) To move the robot – Grippers do not move the robot; that function is performed by actuators
or wheels.

B) To provide sensory feedback – Grippers are not primarily responsible for sensory feedback;
sensors and cameras handle feedback.

C) To manipulate objects – Correct, a gripper is a tool or end-effector designed to pick up, hold,
and manipulate objects in various ways.

D) To calculate robot movement – The gripper does not calculate movement; movement is
controlled by the robot's actuator and controller.

24. Answer: B) Drones working together to monitor a large area

Explanation:

A) A robot assembling components on a production line – This typically involves individual

robots, not swarm behavior.

B) Drones working together to monitor a large area – Correct, robot swarm technology
involves multiple robots working together to achieve a task, such as monitoring a large area
through coordinated efforts.

C) A robot vacuum cleaning a house – This is usually a single robot and not an example of
swarm behavior.

D) A robot performing surgery – Surgical robots typically operate individually, not as part of a
swarm.

25. Answer: B) Sensors

Explanation:

A) GPS – While GPS can be useful for navigation in outdoor environments, it does not fully
allow robots to perceive their environment.

B) Sensors – Correct, sensors such as cameras, LIDAR, and ultrasonic sensors help robots
perceive their surroundings and gather data about obstacles, objects, and other environmental
factors.

C) Actuators – Actuators are used to execute movements or actions but do not help in perception.

D) Processors – Processors help in data processing, but they do not directly perceive the
environment. Sensors provide the necessary input for perception.

26. Answer: C) Surgical robot

Explanation:

A) Autonomous robot – Autonomous robots operate independently but are not specifically used
for surgeries.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
B) Humanoid robot – Humanoid robots resemble humans but are not typically used for surgery.

C) Surgical robot – Correct, surgical robots, such as the Da Vinci surgical system, assist
surgeons in performing complex procedures with high precision.

D) Industrial robot – Industrial robots are used in manufacturing, not in medical surgeries.

27. Answer: C) A method for mapping an environment and tracking the robot’s location
within it
Explanation:

A) A method for storing a robot’s environment data – SLAM is about creating a map and
determining position, not just storing data.

B) A technique used to avoid obstacles – SLAM does help with navigation, but its primary goal
is to build a map and track the robot’s position, not just avoid obstacles.

C) A method for mapping an environment and tracking the robot’s location within it –
Correct, SLAM helps robots build a map of an unknown environment while keeping track of
their location within it.

D) A method to program robot movements – SLAM does not program movements; it is focused
on localization and mapping.

28. Answer: A) A system that combines multiple sensors’ data to improve the robot’s
understanding of the environment
Explanation:

B) A system that fuses the robot’s body parts together – Sensor fusion is not related to body
parts but to combining data from multiple sensors.

C) A system that combines the robot’s power sources – This is unrelated to sensor fusion,
which focuses on data integration.

D) A system that reduces the weight of the robot – Reducing weight is an engineering concern,
not related to sensor fusion.

A) A system that combines multiple sensors’ data to improve the robot’s understanding of
the environment – Correct, sensor fusion integrates data from different sensors (e.g., cameras,
LIDAR, IMU) to enhance perception and decision-making.

29. Answer: B) Humanoid robots are designed to look like humans, while mobile robots
are designed to move autonomously
Explanation:

A) Humanoid robots can move, while mobile robots cannot – This is incorrect, as both
humanoid and mobile robots can move, but they have different designs and purposes.

B) Humanoid robots are designed to look like humans, while mobile robots are designed to
move autonomously – Correct, humanoid robots are designed to resemble humans, while mobile
robots focus on autonomous movement.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
C) Humanoid robots are stationary, while mobile robots are capable of movement – This is
incorrect; humanoid robots can also move.

D) Mobile robots have more sensors than humanoid robots – The number of sensors is not
defined by robot type, but by the robot’s intended task.

30. Answer: B) A robotic arm on an assembly line

Explanation:

A) A surgical robot – Surgical robots assist in complex medical tasks, not repetitive tasks.

B) A robotic arm on an assembly line – Correct, industrial robotic arms perform repetitive tasks
such as welding, painting, or assembling parts.

C) A robot vacuum cleaner – While it performs repetitive cleaning tasks, it's not the most
common example in an industrial or manufacturing context.

D) A robot performing in a play – This is more for entertainment and is not focused on
repetitive tasks.

31. Answer: C) To perform tasks or manipulate objects

Explanation:

A) To provide power to the robot – Power is provided by the robot's power system, not the end-
effector.

B) To control the robot's movement – Movement is controlled by the actuators and the robot's
controller, not the end-effector.

C) To perform tasks or manipulate objects – Correct, the end-effector is the part of the robot
that interacts with the environment, performing tasks such as gripping, welding, or other
specialized actions.

D) To process the robot's sensory data – Data processing is handled by the robot's controller or
processor, not the end-effector.

32. Answer: C) To process sensory data for the robot's navigation

Explanation:

A) To provide power to the robot – Vision systems are not related to providing power; they are
involved in perception and navigation.

B) To track the robot’s location – Location tracking is typically handled by localization systems
(like SLAM), not the vision system alone.

C) To process sensory data for the robot's navigation – Correct, the vision system helps the
robot process visual data (from cameras or LIDAR) to understand its environment and navigate
accordingly.

D) To store the robot's program – Program storage is usually handled by the robot's memory,
not the vision system.

Use code DrSwati for maximum discount on Unacademy Plus subscription.

 UPPSC AE 2024 Combined State Engineering Services Examination
33. Answer: C) A device that performs physical actions based on commands
Explanation:

A) A part of the robot that processes input data – This is the function of the robot's processor
or controller, not the actuator.

B) A system that controls the robot’s sensors – Sensors are managed by the robot's processor,
not the actuator.

C) A device that performs physical actions based on commands – Correct, actuators are
responsible for moving or controlling a part of the robot, such as its arms, wheels, or legs, based
on instructions from the controller.

D) A part that gives feedback to the robot’s controller – Feedback is typically provided by
sensors, not actuators.

34. Answer: D) GPS system

Explanation:

A) Electric motor – Electric motors are commonly used as actuators in robots to control
movement.

B) Hydraulic cylinder – Hydraulic actuators are used in robots that require large amounts of
force for specific tasks.

C) Pneumatic piston – Pneumatic actuators are commonly used in robots for fast movement and
light tasks.

D) GPS system – Correct, a GPS system is used for location tracking and navigation, not for
physical actuation of movement.

35. Answer: B) The number of independent movements the robot can make
Explanation:

A) The number of sensors on the robot – Sensors do not define degrees of freedom; they help
with perception.

B) The number of independent movements the robot can make – Correct, degrees of freedom
refer to the number of independent movements a robot can make, such as rotations or translations
in different directions.

C) The number of actuators the robot has – While actuators influence movement, degrees of
freedom are related to how many different types of movements are possible, not just the actuators.

D) The number of tasks the robot can perform – Degrees of freedom are not related to tasks;
they refer to movement capabilities.

Use code DrSwati for maximum discount on Unacademy Plus subscription.