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

Noosaga Placements

• Robot Manipulationsubfield95%
  The paper focuses on accelerating robot manipulation policies through a new action representation, evaluated on manipulation tasks like picking, cleaning, and stacking.
  B-spline Policy (BSP), an action representation designed for accelerating robot manipulation policiesExperiments on simulated and real-world tasks demonstrate that BSP significantly reduces task completion time
• Imitation Learningframework95%
  The paper builds on imitation learning by converting demonstration trajectories into B-spline targets and training policies to predict B-spline parameters from observations.
  We focus on learning a robot policy π(a | o) from demonstration datasetsTo train B-spline policy, we first convert discrete demonstration trajectories into continuous B-spline trajectories
• Robot Learningsubfield90%
  The method integrates with imitation learning backbones (Diffusion Policy, ACT) and focuses on policy learning from demonstrations.
  We show that B-spline-parameterized actions can be seamlessly integrated into standard policy learning pipelinesWe integrate B-spline policy with two imitation-learning backbones: (i) Diffusion Policy and (ii) Action Chunking with Transformers (ACT)
• Deep Reinforcement Learningframework90%
  The method integrates with deep imitation learning architectures like Diffusion Policy and ACT, which are deep learning-based approaches.
  We integrate B-spline policy with two imitation-learning backbones: (i) Diffusion Policy and (ii) Action Chunking with Transformers (ACT)
• Learning-Based Manipulationframework85%
  The paper is situated within learning-based manipulation, proposing a new action representation to improve speed and smoothness of learned manipulation policies.
  B-spline Policy (BSP), an action representation designed for accelerating robot manipulation policiesExperiments on simulated and real-world tasks demonstrate that BSP significantly reduces task completion time
• Robot Controlsubfield70%
  The paper addresses low-level control aspects such as high-frequency execution, temporal rescaling, and segment alignment for smooth control.
  The low-level controller then samples this continuous segment at a much higher frequency to generate dense robot commandsWe introduce an inference-time segment alignment mechanism that enforces smooth transitions between the predicted B-spline segments

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 B-spline 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: https://b-spline-policy.github.io

Paper Context