B-spline Policy Enables Variable-Speed Robot Motion
du_yilun · x · 2026-07-16
This thread discusses why visuomotor policies are slow: a key bottleneck is action chunking, where the policy predicts waypoints at a fixed frequency, treating all task phases uniformly.
The author points out that real robot manipulation is not uniform: it should move faster in open spaces and slow down during contact-rich interaction phases. To this end, B-spline Policy represents actions as continuously sampleable curves, enabling execution at any control frequency and temporal rescaling, resulting in faster and smoother control.
Related event: B-spline Policy proposes faster, smoother robot manipulation(5 posts)→
More from Embodied
- DONIC hits 86.79% on bimanual dexterous tracking, crushing prior SOTA's 9.43% — liuziwei7 · 2026-10-27
- Bedrock Robotics raises $600M as its fully autonomous excavators do paid work on live job sites — Sethwinterroth · 2026-10-27
- AW-0 hits 80.8% on RoboTwin tasks, beating strongest baseline by 16 points — trevordarrell · 2026-10-27
- AW-0's code stream fixes world model prediction: picks mango where baseline grabs apple — trevordarrell · 2026-10-27
- AW-0 runs at 230ms on a single RTX 5090, 2.4x faster than autoregressive code planning — trevordarrell · 2026-10-27
- AW-0 debuts: one generative world action model plans in code, predicts, and controls robots — trevordarrell · 2026-10-27