Generation of Geodesics with Actor-Critic Reinforcement Learning to Predict Midpoints
This work addresses path and motion planning challenges in robotics and related fields, representing an incremental improvement with a novel method for a known bottleneck.
The paper tackles the problem of finding shortest paths on manifolds by generating geodesics through recursive midpoint prediction, using an actor-critic reinforcement learning approach. It demonstrates experimental outperformance over existing methods in planning tasks such as path planning for agents with complex kinematics and motion planning for multi-degree-of-freedom robot arms.
To find the shortest paths for all pairs on manifolds with infinitesimally defined metrics, we introduce a framework to generate them by predicting midpoints recursively. To learn midpoint prediction, we propose an actor-critic approach. We prove the soundness of our approach and show experimentally that the proposed method outperforms existing methods on several planning tasks, including path planning for agents with complex kinematics and motion planning for multi-degree-of-freedom robot arms.