Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces MoMo, a two-stage imitation-learning framework that utilizes spatiotemporal action tokenization to enable robots to adapt their manipulation behaviors based on task and motion-mode conditions. The framework was evaluated across six real-robot manipulation tasks, demonstrating that varying the motion-mode condition led to distinguishable behaviors in terms of joint speed, acceleration, and end-effector approach pitch. Notably, MoMo successfully transferred unseen motion modes during tasks demonstrated in only one mode, showcasing compositional generalization and the reusability of motion modes across different manipulation tasks.
Robots can learn to adapt their manipulation styles dynamically, achieving compositional generalization across diverse tasks with a single framework.
To operate effectively across diverse contexts, robots must not only perform manipulation tasks accurately but also adapt how their actions unfold to the task, object, and interaction setting. We ask whether this execution-level variation can be learned as a reusable behavioral factor shared across tasks. We present \textbf{MoMo}, a two-stage imitation-learning framework consisting of a spatiotemporal action tokenizer and a behavior-cloning transformer that takes task and a continuous motion-mode condition as inputs. Across six real-robot manipulation tasks, varying this condition produces steady, dynamic, and intermediate behaviors that human raters can distinguish and that differ in joint speed, acceleration, and end-effector approach pitch. On tasks demonstrated in only one mode, MoMo transfers the unseen requested mode while largely preserving task success. Together, these results provide evidence of compositional generalization to unseen task--mode combinations and show that motion mode can be reused across tasks to control how a manipulation skill is performed.