Search papers, labs, and topics across Lattice.
This paper introduces PAINT, a novel approach that enhances the consistency of robot action chunking during asynchronous execution by selecting an appropriate initial noise before generating the next action chunk. Unlike existing methods that rely on steering generation towards previously executed actions, PAINT reframes the problem as one of noise selection, enabling the flow-based policy to maintain coherence without modifications. The method demonstrates significant improvements in execution consistency and task performance across 12 simulated benchmarks and 6 real-world manipulation tasks, showcasing its versatility across different robotic embodiments.
Selecting the right initial noise can dramatically enhance the coherence of robot action sequences without the need for retraining or policy changes.
Action chunking enables robot policies to produce temporally coherent behavior, but generating multi-step action sequences with flow-based policies incurs latency that is incompatible with real-time control. Under asynchronous execution, the robot continues executing the current chunk while the next one is generated, causing even minor delays to create inconsistencies at chunk boundaries. Existing methods address this problem by steering generation toward the already executed action prefix. We instead show that prefix consistency can be achieved by selecting an appropriate initial noise before generation begins, allowing the unmodified flow ODE to produce a coherent next chunk. This reframes asynchronous inference as a noise selection problem rather than a trajectory steering problem. We introduce \textbf{PAINT}, a training-free method that finds this noise via backward Euler inversion and constructs the final chunk through a repainting rule. In summary, \texttt{PAINT} requires no gradients, retraining, or policy modification; yet it improves execution consistency and task performance across \textit{12 simulated benchmarks} and \textit{6 real-world manipulation tasks} spanning single-arm, bimanual, and humanoid embodiments. Website: ~\href{https://paint-action-chunking.github.io}{\texttt{https://paint-action-chunking.github.io}}.
