Search papers, labs, and topics across Lattice.
This paper introduces HCPG-Flow, an innovative approach that enhances the SAC-Flow framework for robot manipulation by integrating hierarchical, object-centric contact-progress guidance. The method improves action selection during execution by transitioning from end-effector approaches to task progress assessments after contact, which allows for more effective ranking of action proposals. Experimental results demonstrate that HCPG-Flow achieves a 9.5 percentage-point increase in mean success on the Maniskill benchmark and reduces completion steps by 17.4% in physical tasks, showcasing its practical efficacy in real-world scenarios.
HCPG-Flow boosts robot manipulation success rates by over 9% by intelligently guiding action selection based on task progress rather than just end-effector positions.
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/
Tsinghua AI