Xi'an Jiaotong University team proposes AffordSim to generate robot manipulation data

Professor Lan Xuguang's team at Xi'an Jiaotong University proposed AffordSim, which connects open-vocabulary 3D affordance prediction to simulated trajectory generation: starting from natural language task descriptions, it locates object functional regions, then combines cuRobo motion planning to generate executable manipulation demonstrations. The team also built an evaluation benchmark covering 50 tasks, 7 categories of manipulation skills, and 5 types of robotic arms, with an asset library containing more than 500 rigid and articulated objects. The team states that a policy trained only with AffordSim demonstrations has been deployed to a real Franka FR3, testing two placing tasks, one pouring task, and one hanging task, 10 trials each, without fine-tuning on real robot data. The results are limited by the coverage of the affordance model; deformable objects, dual-arm manipulation, and in-hand manipulation are not yet included in the framework.

AI 根据报道生成 · 7 hours ago更新