B-spline Policy Accelerating Manipulation Policies via B-spline Action Representations

Jul 16, 2026

Upload by Ophelia

B-spline Policy (BSP) introduces a smooth continuous action representation for accelerating robot manipulation policies. Instead of predicting discrete action chunks, BSP parameterizes actions as B-spline curves using knots and control points, producing time-continuous trajectories that can be temporally scaled for faster execution by low-level controllers. The method integrates into standard policy learning pipelines by directly predicting B-spline parameters, with experiments across simulation and real-world tasks showing reduced completion time and maintained strong success rates.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages13

Reading Time33 min

Introduction

Problem: Inefficiency in action chunking

Motivation: Continuous action curves

Approach: B-spline Policy (BSP)

arXiv :2607 .09648v1 [ cs .RO] 10 Jul 2026 B-spline Policy: Accelerating Manipulation Policies via B-spline Action Representations Xiaoshen Han 1 ∗ Haoyu Xiong2 ∗ Haonan Chen 1 Chaoqi Liu 1 Antonio Torralba2 Yuke Zhu3 Yilun Du 1 1Harvard 2MIT 3UT Austin Abstract: In this work, we present B-spline Policy (BSP), an action representation designed for accelerating robot manipulation policies. Rather than predicting discrete-time action chunks, BSP parameterizes actions as continuous Bspline curves defined by a set of knots and control points. This representation yields smooth, time-continuous trajectories that can be temporally scaled and executed by low-level controllers at higher frequencies and speeds. We show that B-spline–parameterized actions can be seamlessly integrated into standard policy learning pipelines by directly predicting B-spline parameters. Experiments on simulated and real-world tasks demonstrate that BSP significantly reduces task completion time, achieving substantial improvements over baseline methods while maintaining strong success [rates. More results: B-spline-policy.github.io](rates. More results: B-spline-policy.github.io) Keywords: Fast manipulation; Visuomotor policy speedup; Action representation 1 Introduction Robotic manipulation via visuomotor policy learning has made remarkable progress in recent years [1, 2, 3, 4] . Yet, despite these advances, task execution speed remains a major bottleneck. In everyday manipulation tasks such as folding a T-shirt, humans typically complete the task in roughly 10 seconds, whereas even stateof-the-art robotic systems often require close to a minute [5, 1, 6] . This gap exposes a fundamental and largely unresolved challenge in visuomotor policy learning—efficiency: how can robots execute complex manipulation tasks quickly, rather than merely complete them successfully? Figure 1: B-spline policy. Rather than predicting a chunk of discrete-time actions, B-spline policy parameterizes actions as continuous B-spline[2]curves. This continuous representation enables temporal rescaling at execution time, allowing the policy to execute faster and more smoothly. A key source of inefficiency in modern visuomotor policies lies in the design of action chunking [4, 3] . Most existing approaches parameterize actions as fixed-length chunks sampled uniformly over time. While such chunking can stabilize long-horizon prediction and improve learning performance [7], it also introduces inherent limitations: • Uniform temporal resolution. Fixed-length chunks assume that every phase of a task requires identical control granularity. In practice, manipulation tasks exhibit strongly non-uniform temporal structure: fast motions such as reaching can be executed rapidly with coarse actions, whereas contact-rich phases such as grasping, alignment, and insertion demand fine-grained precision. Uniform chunking fundamentally misallocates representational capacity across time. ∗Equal contribution. [2]A B-spline defines a smooth curve by a set of knots and control points. Instead of drawing straight lines between waypoints, a B-spline creates a smooth path that follows their overall trend. Think of it like bending a flexible ruler near waypoints instead of connecting them with sharp corners. • Chunk discontinuities. Stitching together independently predicted action chunks at inference time can introduce discontinuities at chunk boundaries. While such boundary mismatches may be tolerable in low-speed or quasi-static settings, they become highly problematic during high-speed execution, where even small errors can lead to tracking failures or complete task failure. These limitations motivate a paradigm shift: representing robot actions as continuous curves rather than discrete waypoint chunks. Such a representation is used in classical motion planning [8, 9, 10, 11] and computer graphics [12, 13], where spline-based representations are widely favored for being smooth, c

FAQs

What is B-spline Policy (BSP) and what problem does it address?

BSP is an action representation that accelerates robot manipulation by replacing discrete-time action chunks with continuous B-spline action curves, targeting the execution-speed bottleneck in visuomotor policy learning.

How does BSP represent robot actions?

BSP parameterizes actions as continuous B-spline curves defined by knots and control points, enabling time-continuous trajectories rather than stitched fixed-length chunks.

How does BSP achieve faster execution without losing fidelity?

BSP supports temporal rescaling at execution time, letting each predicted B-spline segment be retimed and sampled at higher frequencies while an inference-time segment alignment mechanism enforces smooth transitions.

Jul 16, 2026

Upload by Ophelia