AffordSim Integrates 3D Affordance Prediction into Robot Operation Data Generation

Professor Lan Xuguang's team at Xi'an Jiaotong University presents AffordSim, a system that connects natural language task descriptions to executable robot trajectories by predicting 3D functional regions on objects before motion planning. The framework reduces the need for manual contact point specification and supports evaluation across 50 tasks, seven operation skills, and five robotic arms. It also incorporates real-world scene backgrounds and camera viewpoint variations into simulated demonstrations to improve zero-shot transfer to physical robots.