LMGN: A Unified Policy Learning Model for Last-Mile Grasp-Navigation
LMGN: A Unified Policy Learning Model for Last-Mile Grasp-Navigation
This paper introduces LMGN, a unified policy learning framework designed for legged robots with manipulators operating in complex environments. The authors propose an Explicit Distance and Directional Reachability (ED-DR) representation to model grasp-compatible base placements and approach directions. A semantically aligned Hierarchical Mixture-of-Experts (HMoE) policy is developed to handle distinct control behaviors across stages including base adjustment, pre-grasp alignment, grasp execution, and object lift. A new optimization method jointly aligns navigation and manipulation while ensuring stage-dependent expert routing and action-level consistency. Experiments conducted on the Habitat and Isaac Gym simulators demonstrate that LMGN outperforms state-of-the-art methods in terms of success rates for last-mile navigation and target grasping.
The authors present LMGN, a unified policy learning model for quadruped robots that integrates explicit grasp-reachability representations with hierarchical mixture-of-experts policies to address last-mile navigation and manipulation challenges.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Environment
- SimulationOpen source S5
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
HMoE policy. Extensive experiments on the Habitat and Isaac Gym simulators demonstrate that LMGN consistently outperforms state-of-the-art methods, denoting that integrating grasp-aware reachability with stage-aware policy learning significantly improves the success rates of last-mile navigation and target grasp for robots.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org