�� ICRA ��Ұ��RoboScience ��ѧ�ġ��ԡ�Ѻ��

2026-06-07 11:14

��ϰ��

��ע

��ߣ��ҷ�

�༭��ΠD

��Ʒ��ϰ��

�� RoboScience ��ѧ�� ICRA ��ĵ��Ұ�С�

��˵��ѧ��ÿ�궼��һ��ʱ�̣��ȫ��Уʵ��ҡ��ҵ��˾��о��ͬʱ��Ŀ��Ͷ��ͬһ��ICRA һ��Ҫ��Ǹ��̨֮һ��

ICRA��ȫ�� IEEE International Conference on Robotics and Automation�� IEEE ��Զ��ѧ��µ��콢��飬Ҳ��ȫ��Զ��Ӱ��Ķ��ѧ��֮һ��

��һƪ��ԵĲ��ͨ��ͬ��󣬶��ȫ��ѧ��ǰ�صľ��ڻ��о��˵��ܹ��ICRA ��Ѿ��һ��Ҫ�Ͽɣ��ܹ��Ұ��ζ����Ž��ȫ��˼��·�ߵķ��

��̨�ϣ�ICRA 2026 ��Ƚ��——“��˲��˶�”�� Best Paper Finalist ���벮��ѧ��˹̹��ѧ��ʡ��ѧ��廪��ѧ�ȶ��г��ֵģ��һ֧�� RoboScience ��ѧ��ϯ��ѧ��ִ�� NUS �Ŷӣ��Ϊ��Bi-Adapt: Few-Shot Bimanual Adaptation for Novel Categories of 3D Objects Via Semantic Correspondence��

��Ѿ��Ŷ��ڶ��վ�� ICRA ��ĵ��̨��

һ��ǰ��ǵ��һƪ��ġ�D��R��O�� Grasp��ȫ��ǧƪͶ��ͻΧ�� ICRA 2025 “��˲��˶��Ľ�”��ICRA 2025 �ٷ��ʾ��ƪ��Ļ񽱵��ɣ��“�粻ͬ��Ŀɷ��ץȡ��”��˹��ס�

��仰˵��RoboScience ��ѧ��걻 ICRA ��Ϊ��ڻ��˲��ѵ��ϸ��𰸡�

ȫ�� 4000+ Ͷ�塢��ʲ�� 1%��ѵõ��Ψһ��ѡ——��µķ��ڻ��ѧ�粻��Ҫ�κζ��Ρ��֮�⣬��ִ��Ŷӽ�� ICRA 2026 һ�� 10 ƪ�о��ɹ��ɹ��ѡ��ץȡ��罻��ͳɱ��֪��滮�Ⱥ��ķ��ϵͳ�Ե��ƽ��ܵļ��߽硣

��˵��һ�ҳ��һ��й��˾��ڰ��Լ��ļ��·�߷ŵ�ȫ��ѧ���ҡ�Ҳ��ľ��ܼ��飬��ҳɹ�˵��ͬ٭��

�ص��ı��ᷢ�֣��ƪ��ڽ��ͬһ��⣺�û��˴�“һ��һ��”��߳��

D��R��O�� Grasp ��һ�� AI ��Կ�� 3 ָ��4 ָ��5 ָ��ͬ��̬��֣��籾��ץȡ�ϰ��壬�ɹ�� 87% ��ϡ��ʱ�䲻�� 1 �롣Bi-Adapt �û��ֻ��ʾ��ܰ��Ѿ�ѧ��˫��Э��Ǩ�Ƶ��δ��——��֮�⡣

һ��“��”��һ��“��”��Ʋ�ͬ�ļ��·��ͬһ��жϣ��ܵ��һ�̣�ʤ��“��”��

��Ҳ�� RoboScience ��ѧ��ҳ��һ��й��˾��⽲�Ĺ��º�Ѻ�µĶ�ע——“��Ʒ��ƿ��”��λ��ʼ�ˣ��ϯ��ѧ��֡�CEO ��Ұ��ֱ��ע��Ҫ��ˣ�ǰ��о��ķ��ۣ��ʹ��ģ��̵�ִ��

һ�Ա�֮��һ��룬ȴ����·��Ļ��ѧ�磬��ѹ��˵ķ��ǰ̨��걻 ICRA �Ͽɡ��һ��ϣ�RoboScience ��ѧ�Ѿ��ֻ��һ��򵥵ĳ��˾��֣��̨��й��

�ڹ�ȥ��ľ��˳����ͷ��Ҷ��ͬһ��ʽ��——VLA��Vision-Language-Action��Ӿ��ֱ��RoboScience ��ѧ��ҹ��߳��ͬ·�ߵĹ�˾��ǵĺ��ļ��ܹ�� VLOA��Vision-Language-Object-Action��Ǹ� O�� Object——��塣

��ֻ��һ��ĸ�Ĳ�𣬱��ȴ��һ��ȫ��ͬ��жϣ��Ҫ��“ͨ��”��Ҫѧ��л��ݻ��ȥ��ô��

VLA ��·��⣬ǡǡ��“��籾��”——�Ӹ�֪��ֱ��ھ�̬��̶��ĳ��ܹ��ֻҪ��һ��塢��һ��񡢻�һ̨��ͬ��͵Ļ��˱��壬��Ҫ��²ɼ��ݡ��ѵ��ģ�͡�˵��ˣ�VLA ��һ��“ʵ��-��”��Ӳ�󶨣��ͻ��“��”��

VLOA ��Ҫ��ģ��£��һ��ԣ�ָ��ˡ��塢��

��嵽�ܹ��ϣ��“��ģ��”��“ͨ�ò��ģ��”��ɣ��м��һ�� Object Trajectory��켣��Ľӿڰ��ߴ�ͨ��

�ϲ�ľ��ģ�ͣ��“��”��û��ڶ��֮ǰ��Ԥ��һ��δ��ĸ��ȥ��̬��α仯��˭�ᷢ��Ӵ��һ�ֽ�“3D ��ƹ켣”��м��ʾ——��ֱ�ۿ��˶�·��Ȼ��Լ�� 2D ��Ƶ��ﳣ��ҡ��崩ģ��

ͼע��ģ��3D ��ƹ켣

�²��ͨ�ò��ģ�ͣ��“�ı��”��켣��ɻ��˵ĹؽڽǶȡ��Ӵ��źţ��ﾫ׼��֡��“һ��һ��ģ��”��Ƭ��м��ѵ��ͬһ�׵ײ��

ͼע��ͨ�ò��ģ�͸��3D ��ƹ켣��

�м�� Object Trajectory�� VLOA ���һ��——��“��֪”��“ִ��”��׽���ϲ㲻�ع��ʲôӲ��²㲻�ع��Ҫ��ʲô��֮��“�� 3D ��ƹ켣”��ּ��ɶ��ֻ��ִ�е�ͨ��ӡ�

��ֽ��ֵģ�� VLOA ��£�

��壺�ӹ⻬��ϴ��ˮƿ��͸��ǩ�С��Ӳ��ʲ��ϣ�ͬһ��ģ��Զ��䣬��Ϊÿ��嵥��ѵ��

��񣺿��ŷ�Ҫ��ţ��Ӳ��Ҫ��̬ƽ�⡢ץ��ƬҪ��ѹ�顢��ע��Ҫ��׼��——��Щ��ȥҪ�ֱ𿪷��㷨�ľ�ϸ��񣬱�һ��ģ��ͳһ��£�

�籾�壺ģ��Ӳ��ȫ����һֻ��־��ֱ��á��磬ͬһ�ײ��Կ��޷�Ǩ�Ƶ� X-hand��12 ��ɶȳ��׼ֱ�� LEAP Hand��16 ��ɶ�ֱ��ȫ��ͬ��̬��֡�

��˵��İ��ȥ�� 5 �£�� VLOA��RoboScience ��ѧ��ȫ��Χ���ӡ��ߡ��ľ��——ƴ�Ҿߡ��¼��˻��˲��ѵ㣺��ڲ��˫��Эͬ��׼��ȡ��滮��ء�ģ�Ͷ��˵��ƴװ��;��Ϊ��ţ�ϵͳ��Զ��ָ�״̬��ɡ�

��ؼ��ǣ��“ͨ��”��ǿɳ��ͻ�Ƶġ�RoboScience ��ѧ��Ϊ��ģ�ͻ��۳�� 100 ��Сʱ��Ϊ��ĵĶ�ģ̬��Ƶ��ݣ��ÿ��ʮ��Сʱ��Ŀ�� 2026 ��׹��ǧ��Сʱ��ݼ��Ϊͨ�ò��ģ�ͣ��ж�ģ̬��ƽ̨ RoboMirage��ѻ�� 10B��100 �ڴΣ� ��켣��Ŀ�� 2026 ��ͻ�� 1T��1 ��ڴΣ��ģ�Ͷ��ڹ��֤�� Scaling Law——��ݹ�ģԽ�󣬷��Խǿ��ҳʿ�Ԥ��

��ζ�� RoboScience ��ѧѺ�µ��·�ߣ��һ��ܹ��˼��һ��Գ��ѩ�򡢹�ģԽ��Խ��Ĺ��ϵͳ��

��֮��VLOA �� VLA ��Ҳ��Զ�ĵط��ڣ��Ҫ�û��ң��硢��ı��硣

��ģ�ͺ� RoboScience ��ѧ��ƪ��񽱵��ķ��һ�𿴣��ʵ��һ��еġ��Ƕ��ͬһ��£��ѻ��˴�“ʵ��-��”��Ӳ��룬�ع�Ϊ“��ϵ-�켣”�Ŀɷ��ϵͳ��׵ײ㷶ʽ�� RoboScience ��ѧѺ�µ��߼�ֵ�Ķ�ע��

��ܽ��ѧ�Ƶ��ʱ��ˣ��һ�͵��˲ź��ͨ�ջ��

һ��Ŷ��ֻ�ó�ѧ��д�� paperȴ��Ʒ��ֻ�ó��̣��Ӳ��ȴд��ǰ�ص��㷨��ģ��“��Ҫǰ��ԭ��Ҫ��ģ��”�ķ��ע��Ҫ��ັ��˲�װ��ͬһ֧��

RoboScience ��ѧ��ϡȱ�ԣ�ǡǡ��˶��ˡ�

�ȿ��ϯ��ѧ��֡��й��“˹̹��ϵ”��һ��ؼ����Ͼ��ѧ��ʿ˹̹��ѧ��ʦ�ӻ��֪��ѧ�� Jeannette Bohg��ϵ�ʦ�� Leonidas J. Guibas——��ԺԺʿ��ͼ��ѧ�ͼ��δ��ĵ��֮һ��¼��¹��ѧ��NUS��ڡ�

˹̹��Ļ��Ȧ��ڹ�ȥ��˹۲��й��ܸ�ֵ�һ��ꡣ��껣�Hillbot ��ʼ�ˣ��ף��ͨ�ô�ʼ�ˣ��¬��ᣨ񷳹��ܴ�ʼ�ˣ��˺��ͬ��һ��ѧ��ϵ��Щ�˺��й��Ĵ��ͼ——Ҳ��ζ��Ȼ�ڹ��ǰ�ص�ѧ��Ȧӵ��һ��꼶��λ�á�

��ؼ��ǣ��ֵ��о��ߣ��Ӳ�ʿʱ��һֱ��ڻ��˲��“ͨ�û�”�ϣ��ڵ� UniGrasp�� D��R��O�� Grasp��Bi-Adapt��ٵ��¼�� ICRA 2026 �� T��R��O�� Grasp——��Կ��һ��ǳ��о��磺��ϳ��󡢲��ͳһ��׷��籾��ͨ�ñ��

��ֳ��һ�µ��о��ζ�� RoboScience ��ѧ�ļ��ʱƴ�յĲ����һλ��ѧ�߻��˽�ʮ��ʱ�䲻�ϴ�ĥ��ķ��۳��

�ٿ��Ұ��ϡȱ�ģ��Ǵ��пƴ��˹̹�� AI Lab ��ƻ�� AI Platform һ·�߳��Ĺ��̻��Ұ��Ʊ�ҵ��й��ѧ��ѧ��ϵ��˶ʿ��ҵ��˹̹��ѧ AI Lab��ʦ�Ǳ��Ϊ“AI ��”��

��ҵ��Ұ��ƻ�� AI Platform ��——��ְλ�ķ��Ҫ�� AI Ȧ��˲ŵ��ó��ĺ��ƽ̨��ҵ�ڳ�Ϊ“ƻ�� PyTorch �� CUDA”��֧��˶��ؼ� AI ��ƻ��̬�еĴ��ģ��أ��ƻ�� AI �ܹ��ʮ��̨�豸��ȶ��еĻ��ʩ��

��仰˵��Ұ��һ��“�� CEO”��Ǽȶ�ǰ��㷨��ֶ��ΰ��㷨װ��ʮ�ڼ��ҵϵͳ��ϡȱ��䡣

��ܵĲ�ҵ��ǡǡ��Ҫ��ҵ��֮��Զ�ѵģ��ȥ��Ƹ��Ī��¶��Ⱥ��“��ҵ��”��ǰ��ģ��Ҫ��ɿ��ȶ��С��Ĳ�Ʒ��Ĳ��ĳ��㷨��˼��ǹ��̵ػ�——ʲô��ˮ��ȶ�ιģ�͡�ʲô��֧��ʵʱ��ơ�ʲô��Ĺ��̷�ʽ��ڲ�ͬӲ��ϱ��һ��顣��Щ��о��Ա��ǿ���ǹ�ҵ�� AI ��ʦ��ܽ��⡣

Ҳ��ˣ��ֺ��Ұ��ϣ��һ��ҵ��Ϊ��ʵ�Ĵ��䣺˹̹��ѧ�� × ��ȹ��䡣