The Physics & Astronomy Colloquium Series presents Jundong Liu from Ohio University, discussing "Physics-Inspired Reinforcement Learning for Robotic Motions" on October 4.

Abstract: Collision avoidance is a crucial task in vision-guided autonomous navigation. Solutions based on deep reinforcement learning (DRL) have become increasingly popular. In this work, we propose several novel agent state and reward function designs to tackle two critical issues in DRL-based navigation solutions: 1) the smoothness of the trained flight trajectories, and 2) model generalization to handle unseen environments.

Formulated within a DRL framework, our model relies on physics-inspired rewards and smoothness constraints to ensure UAVs fly smoothly while significantly reducing the risk of collision. The proposed smoothness reward minimizes a combination of first-order and second-order derivatives of flight trajectories, which also drives the points to be evenly distributed, leading to stable flight speeds. Experiments demonstrate the effectiveness of our overall design and its individual components.