PAC-ACT: Post-training Actor-Critic for Action Chunking Transformers
2026-07-10 • Robotics
RoboticsArtificial Intelligence
AI summaryⓘ
The authors designed PAC-ACT, a method to improve robot control policies for precise industrial tasks involving contact. They built on existing Action Chunking Transformer (ACT) models, fine-tuning them with reinforcement learning while keeping their original behavior stable. This approach helps robots handle force and pose changes better without slowing down or needing much memory. Tests showed PAC-ACT made robots safer and more successful at delicate tasks, like reducing excessive contact force significantly. The authors also found that their method helps robots explore well even with limited feedback and changing starting positions.
Precision Industrial ManipulationVision-Language-Action ModelsAction Chunking TransformerReinforcement LearningBehavior CloningActor-Critic ArchitectureContact Force ConstraintsPolicy OptimizationSparse RewardsRobotic Control
Authors
Yujie Pang, Zudong Li
Abstract
Precision industrial contact manipulation requires reliable robot policies under pose perturbations and contact-force constraints. Vision-language-action models offer broad generalization but often introduce high inference latency and GPU-memory cost, while vision-action chunking policies are more suitable for real-time industrial control. However, these policies are usually trained by behavior cloning and suffer from distribution shift in contact-rich tasks. This paper proposes PAC-ACT, a reinforcement-learning post-training framework for pretrained Action Chunking Transformer policies. PAC-ACT reformulates policy optimization at the chunk level, constructs an ACT-transferred actor-critic architecture, and introduces a hybrid behavior-prior constraint to preserve the pretrained action distribution during online fine-tuning. Experiments on industrial precision-contact benchmarks show that PAC-ACT improves task success, contact stability, and force safety while retaining low latency and low GPU-memory usage. On the Contour task, PAC-ACT significantly reduces peak contact force and decreases the proportion of force readings above 60 N by 46 times. Sparse-reward ablations further show that the proposed behavior-prior constraint enables effective exploration under randomized initial poses.