TCP4D Unifies Robot Tool-Center-Point Tracking and Trajectory Generation
TCP4D paper authors present a unified model that uses a shared 4D scene representation to track observed tool-center-point motion and generate future trajectories. The authors report that, from monocular video and initial 2D TCP queries, the tracker recovers metric SE(3) poses, gripper states, scene geometry and camera parameters. They also report 7.60 mm and 16.07 mm absolute trajectory errors on DROID and RoboLab, 69.19% mean success on RoboTwin, and a 41.5% improvement in target-robot success from joint training with source and augmented demonstrations; these figures are author-reported.
TCP4D combines robotic action annotation, cross-embodiment data augmentation and future-trajectory generation in one model, making it relevant to embodied AI research and robot-learning systems. The reported results have not been independently validated in the supplied source.
Computer Science > Robotics
arXiv:2611.23143 (cs)
[Submitted on 17 Nov 2026]
Title:TCP4D: A Unified Tool-Center-Point Tracking and Generation Model for Robot Manipulation
Authors:Haosong Peng, Xiangyang Liu, Zhengshen Zhang, Yunfeng Fan, Tianyu Qi, Xin Xia, Yufeng Zhan, Wenchao Xu
View a PDF of the paper titled TCP4D: A Unified Tool-Center-Point Tracking and Generation Model for Robot Manipulation, by Haosong Peng and 7 other authors