Learning Multi-Agent Loco-Manipulation for Long-Horizon Quadrupedal Pushing
This addresses the limitation of manipulation capabilities in quadrupedal robots for real-world applications like search and rescue and industrial automation, representing a strong specific gain.
The paper tackles the problem of enabling multiple quadrupedal robots to perform long-horizon pushing tasks with obstacles, achieving a 36.0% higher success rate and 24.5% reduction in completion time compared to baselines.
Recently, quadrupedal locomotion has achieved significant success, but their manipulation capabilities, particularly in handling large objects, remain limited, restricting their usefulness in demanding real-world applications such as search and rescue, construction, industrial automation, and room organization. This paper tackles the task of obstacle-aware, long-horizon pushing by multiple quadrupedal robots. We propose a hierarchical multi-agent reinforcement learning framework with three levels of control. The high-level controller integrates an RRT planner and a centralized adaptive policy to generate subgoals, while the mid-level controller uses a decentralized goal-conditioned policy to guide the robots toward these sub-goals. A pre-trained low-level locomotion policy executes the movement commands. We evaluate our method against several baselines in simulation, demonstrating significant improvements over baseline approaches, with 36.0% higher success rates and 24.5% reduction in completion time than the best baseline. Our framework successfully enables long-horizon, obstacle-aware manipulation tasks like Push-Cuboid and Push-T on Go1 robots in the real world.