人工智能
机器人
计算机科学
一般化
接口(物质)
过度拟合
学习迁移
夹持器
机器人学习
人机交互
机器人学
计算机视觉
弹道
运动(物理)
空格(标点符号)
机械臂
传输(计算)
抓住
机器人控制
转移问题
机器学习
认知机器人学
运动规划
社交机器人
机器人运动学
作者
Tong Wu,Shoujie Li,Junhao Gong,Changqing Guo,X. J. Li,Shilong Mu,Wenbo Ding
标识
DOI:10.1109/lra.2026.3656802
摘要
Robotic foundation models trained on large-scale manipulation datasets have shown promise in learning generalist policies, but they often overfit to specific viewpoints, robot arms, and especially parallel-jaw grippers due to dataset biases. To address this limitation, we propose Cross-Embodiment Interface (CEI), a framework for cross-embodiment learning that enables the transfer of demonstrations across different robot arm and end-effector morphologies.CEIintroduces the concept offunctional similarity, which is quantified using Directional Chamfer Distance. Then it aligns robot trajectories through gradient-based optimization, followed by synthesizing observations and actions for unseen robot arms and end-effectors. In experiments,CEItransfers data and policies from a Franka Panda robot to16different embodiments across3tasks in simulation, and supports bidirectional transfer between a UR5+AG95 gripper robot and a UR5+Xhand robot across6real-world tasks, achieving an average transfer ratio of 82.4%. Finally, we demonstrate thatCEIcan also be extended with spatial generalization and multimodal motion generation capabilities using our proposed techniques. Project website:https://cross-embodiment-interface.github.io/.
科研通智能强力驱动
Strongly Powered by AbleSci AI