Search papers, labs, and topics across Lattice.
This paper introduces AffordTrajDP, a dynamic framework for robotic manipulation that enhances affordance-guided imitation learning by constructing affordance trajectories through object-centric temporal propagation. By leveraging RGB-D observations and propagating anchor affordances based on the object's SE(3) pose, the method ensures temporally consistent and state-aware guidance, addressing the limitations of static affordances in precision-critical tasks. The approach achieves a 70.0% average success rate on the ManiSkill3 benchmark, outperforming existing methods by up to 17.8%, and demonstrates robustness in real-world applications across various robotic arms and tasks.
Dynamic affordance trajectories can significantly improve robotic manipulation success rates, achieving 70% on challenging benchmarks.
Affordance-guided imitation learning has shown impressive performance in robotic manipulation tasks by compressing visual perception into task-specific geometric constraints (e.g., fixed contact points). However, the commonly used static affordances can become inconsistent in precision-critical tasks or under object location perturbations, leading to post-contact trajectory drift. To address this issue, we propose AffordTrajDP, a dynamic framework that constructs affordance trajectories via object-centric temporal propagation to guide the progressive manipulation process. Specifically, given an RGB-D observation, our core insight is that a retrieved anchor affordance, which captures the desired contact point between the end-effector and the target object, can be propagated forward via affordance propagation, using the object's SE(3) pose as a natural propagation medium, to yield an affordance trajectory that provides temporally consistent, state-aware guidance throughout execution. AffordTrajDP achieves 70.0% average success rate on ManiSkill3, outperforming strong baselines by up to 17.8%. Real-world experiments on Galaxea A1 and UR7e robotic arms, covering StackCube, PickCup, AdapterInsertion, Ring-on-Peg, Put-in-Bowl, and USB Insertion, further validate robustness under object placement variations and appearance changes, with seen and unseen object instances evaluated on Galaxea A1, and ablations confirm the contribution of each proposed component.
