Research/ PLANETARY ROBOTICS
Safe Deep Reinforcement Learning-Based Controller (SDRLC) for Autonomous Navigation of Planetary Rovers
Ravi Kiran Jana, Rudrashis Majumder, K S Bharathwaj, Suresh Sundaram
Motion Planning & NavigationOpenAlex topic: Robotic Path Planning Algorithms · Learning-based MethodsOpenAlex topic: Reinforcement Learning in Robotics · Guidance, Dynamics & ControlOpenAlex keyword: control barrier functions
Surface exploration and data collection by planetary rovers are challenging due to unknown complex planet terrains. This paper focuses on developing a Deep Reinforcement Learning (DRL)-based controller for rovers to enable safe operation. The necessary control input for safe and efficient vehicle maneuver is derived using the Control Barrier Function (CBF)-based safety protocols. Deep Deterministic Policy Gradient (DDPG) algorithm is used as a DRL framework to find the optimal exploration policies for the rover. Numerical simulations on different vehicle models show the efficacy of the proposed safety method for planetary rovers.