Event
AdaPolicy: Action-Driven Anticipatory Robot Policy
1 reports · 1 sources
Fan Yang and co-authors submitted a preprint proposing a robot policy framework, AdaPolicy: through cascaded next-action-chunk prediction, it shapes perception representations during training that can anticipate the future, without encoding future observations, and adds no extra overhead at inference time. The authors report that the method achieves consistent improvements on the MetaWorld, RoboTwin 2.0, and LIBERO simulation benchmarks and in real-world deployment, with gains of up to 14 percentage points on millimeter-level precision tasks such as headphone assembly and shoelace threading. These results are as reported in the paper.
Report timeline
- AdaPolicy: Action-Driven Anticipatory Robot Policies via Next-Chunk Prediction
AdaPolicy proposes an action-driven anticipatory policy framework that shapes future-aware representations through cascaded next-chunk prediction, requiring no encoding of future observations during training and adding zero inference-time overhead. The method achieves consistent gains on MetaWorld, RoboTwin 2.0, LIBERO, and real-world deployment, with improvements of up to 14 percentage points on millimeter-precision tasks such as headphone assembly and shoelace threading.
arXiv cs.RO ·