Reachability-Based Trajectory Safeguard (RTS): A Safe and Fast Reinforcement Learning Safety Layer for Continuous Control

Reinforcement Learning (RL) algorithms have achieved remarkable performance in decision making and control tasks by reasoning about long-term, cumulative reward using trial error. However, during RL training, applying this trial-and-error approach to real-world robots operating safety critical environment may lead collisions. To address challenge, letter proposes a Reachability-based Trajectory Safeguard (RTS), which leverages reachability analysis ensure training operation. Given known (but uncertain) model of robot, RTS precomputes Forward Reachable Set the robot tracking continuum parameterized trajectories. At runtime, agent selects from receding-horizon way robot; FRS is used identify if agent's choice safe or not, adjust unsafe choices. The efficacy method illustrated static environments on three nonlinear models, including 12-D quadrotor drone, simulation comparison with state-of-the-art motion planning methods.