ChunkFlow: Towards Continuity-Consistent Chunked Policy Learning
2026-07-14 • Robotics
Robotics
AI summaryⓘ
The authors address a problem in vision-language action models where dividing tasks into chunks causes inconsistent predictions at the boundaries, leading to errors over time. They propose ChunkFlow, a method that carefully manages these chunk boundaries during both training and execution to create smoother, more reliable actions. Their approach includes special training losses and techniques to handle errors from previous steps, improving stability without slowing down the system. Tests on datasets and robots show better performance balancing success and speed.
vision-language modelsaction recognitionchunked policiestemporal coherenceoverlap blendingscheduled samplingAWAC fine-tuningsmoothness assumptions
Authors
Zhao Yang, Yinan Shi, Mingyuan Yao, Wenyao Xue, Yawei Jueluo, Longjun Liu
Abstract
Vision-language action (VLA) models increasingly adopt chunked action heads to satisfy real-time constraints; however, this introduces boundary jitter: overlapping regions between consecutive chunks often yield inconsistent predictions, degrading temporal coherence and the task success rate. Existing methods, such as inference-time blending, merely reweight mismatched proposals without correcting underlying errors, leading to residual accumulation under biased or noisy histories. We propose ChunkFlow, a seam-aware training-and-execution framework for chunked policies that aligns chunk structure with boundary execution. It partitions each chunk into frozen, editable, and future zones, applies deterministic overlap blending at execution, and trains raw predictions with seam and first- and second-order continuity losses. History corruption and scheduled sampling improve robustness to executed-history errors, while an AWAC fine-tuning stage adapts the policy without removing these structural regularizers. Under mild smoothness assumptions, pre-blending seam discrepancies provably decay with increasing overlap. Experiments on CALVIN, LIBERO, and real robots show an improved success-stability trade-off with low-latency inference. Project page: https://cytoderm-ai.github.io/chunkflow.