HCPG-Flow:Hierarchical Contact-Progress Guidance for Flow-Policy Robot Manipulation

2026-07-20

AI Summary

This work addresses the lack of reliable action selection mechanisms in existing flow-based policies for robotic manipulation, which are often undermined by inaccurate Q-value estimates from replay buffers. The authors propose a training-free, analytical action selector that dynamically alternates between approach and task-progress phases through a hierarchical, object-centric contact-progress guidance mechanism. Action proposals are scored and normalized based on first-order reductions of task-relevant distances, and robust selection is achieved via temperature-controlled embeddings. While preserving the objective of SAC-Flow, the method significantly improves task success rates—outperforming SAC-Flow on average across ten simulated tasks and achieving a 9.5 percentage point gain on the ManiSkill benchmark. In real-world experiments on four robotic tasks, it not only attains higher success rates but also reduces the number of steps to success by 17.4%.

Abstract

Flow policies can represent multimodal action distributions for robot manipulation, yet a robot must execute one action at each control step. When several proposals are sampled, critic-based ranking makes data collection depend on value estimates over candidate actions that may be weakly represented in replay. We introduce HCPG-Flow, an analytic rollout-time selector that augments SAC-Flow with hierarchical, object-centric contact-progress guidance while preserving its actor and critic objectives. HCPG switches from end-effector approach to task progress after contact, scores each proposal by the first-order reduction of a task-relevant distance, standardizes scores within the candidate set, and executes a temperature-controlled action embedding. Across ten simulated tasks, HCPG improves mean success over SAC-Flow on both benchmarks, including a 9.5 percentage-point gain on Maniskill. Four physical tasks further show high success with a 17.4% reduction in successful completion steps.Project page: https://hitxraz.github.io/HCPG-Flow/

Problem

• flow policies
• robot manipulation
• action selection
• critic-based ranking
• multimodal action distributions

Innovation

• flow policy
• contact-progress guidance
• hierarchical selection
• analytic rollout-time selector
• robot manipulation