Researchers have introduced a new framework called ART (3D Affordance tRansfer neTwork) to address the challenge of transferring human-derived affordance knowledge to robots. Existing methods often fail because affordances are dependent on the agent's embodiment, and human hands differ significantly from robot end-effectors. To facilitate this transfer, a new dataset called HPAD (Human-Robot Paired Affordance Dataset) was created, featuring over 2,137 objects with both human-prior and robot-validated affordance labels. ART improves transferability by filtering unreliable human priors and reconstructing them into a robot-actionable format using graph-diffusion refinement, achieving an 86.5% success rate in real-world robotic manipulation tasks. AI
影响 This research could improve robot learning and manipulation by enabling more effective transfer of skills from human demonstrations.
排序理由 The cluster contains a research paper detailing a new model and dataset for a specific robotics task. [lever_c_demoted from research: ic=1 ai=1.0]
AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →