Understand this story
The Ding Ning team at Tsinghua University, together with Beijing University of Posts and Telecommunications, Peking University, and Nature Will, built HandsOn, a simulation platform and evaluation benchmark for cross-embodiment dexterous manipulation. The benchmark contains 66 manually verified tasks and 18,000 human simulation teleoperation demonstrations, and supports 5 types of robotic arms, 6 types of dexterous hands, and about 200 object assets. Among the Core-50 tasks that provide demonstrations, the best-performing trained multi-task policy, OpenWAM, achieved a success rate of 29.1%; on Transfer-16, which has no target-task demonstrations, the average success rate of all 8 trained multi-task policies was below 1%. GPT-6, controlled zero-shot through the move_eef interface, achieved success rates of 5.2% and 1.9% on Core-50 and Transfer-16, respectively. There are currently no further developments.
AI-generated from public reports
Developments
1 confirmed developments
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 26
36Kr · AI
Tsinghua team builds the HandsOn dexterous manipulation benchmark, with 66 tasks comprehensively evaluating robots' two-handed capabilities
The team led by Ding Ning at Tsinghua University, together with Beijing University of Posts and Telecommunications, Peking University, and Nature Will, built HandsOn, a simulation platform and evaluation benchmark for cross-embodiment dexterous manipulation, comprising 66 manually verified tasks and 18,000 human simulation teleoperation demonstrations, and supporting 5 types of robotic arms, 6 types of dexterous hands, and about 200 object assets. Among the Core-50 tasks that provide demonstrations, the best-performing trained multi-task policy, OpenWAM, achieved a success rate of 29.1%; on Transfer-16, which has no target-task demonstrations, the average success rate of all 8 trained multi-task policies was below 1%. GPT-6, controlled zero-shot through the move_eef interface, achieved success rates of 5.2% and 1.9% on Core-50 and Transfer-16, respectively.