PATH PLANNING USING REINFORCEMENT LEARING POLICIES WITH TIME CONSTRAINTS FOR LUNAR ROVERS EXPLORATION

The autonomous navigation of planetary rovers in extraterrestrial environments rep- resents one of the most challenging topics in Robotics and Mechatronics Engineering. On the lunar surface, areas of highest scientific interest often coincide with Perma- nently Shadowed Regions, where the absence of sunlight results in extremely low temperatures and prevents solar charging. These conditions introduce strict time constraints on mobility and make classical path-planning strategies insu!cient. This thesis presents a reinforcement-learning-based path planner designed to handle these limitations by integrating terrain elevation, danger zones, and energy constraints into a custom training environment. Several agents were trained and tested across multiple simulated scenarios. Results show that the learned policy performs well in simpler environments and can account for non-local criteria that classical plan- ners cannot capture. Performance decreases in more complex settings, mainly due to looping behaviours, but the approach demonstrates strong potential and estab- lishes a basis for future development of reliable RL-based navigation strategies for off-Earth exploration.