AffordSim

A team led by Prof. Lan Xuguang at Xi'an Jiaotong University proposes AffordSim, which plugs open-vocabulary 3D affordance prediction into simulation-based trajectory generation: starting from natural-language task descriptions, it localizes functional regions on objects and then uses motion planning to produce executable robot manipulation demonstrations. The work also introduces a benchmark spanning 50 tasks, 7 manipulation skills and 5 robot arms, with code released openly, targeting the failure of generic grasping to preserve task-relevant alignment.