Reliable autonomous navigation is a central requirement for future lunar rover swarms operating in cluttered, partially known terrain. Small rover swarms offer advantages in redundancy, coverage, and mission flexibility, but their navigation problem is more demanding than single-rover path planning: the system must avoid hazards, handle incomplete obstacle knowledge, prevent inter-agent interference, and select routes feasible for the group footprint. This thesis investigates how the Robust Artificial Potential Field (RAPF) framework can be extended toward reliable goal reachability for lunar rover swarms in obstacle-dense environments.

The work first improves the single-agent RAPF planner through systematic hyperparameter optimization, midpoint collision checking, and partial-trajectory repair. The resulting Improved RAPF (I-RAPF) planner reduces physically invalid path segments and avoids unnecessary full-path recomputation after local-minimum events. In controlled full-map experiments, I-RAPF achieves near-perfect reachability while reducing planning effort compared with the tuned RAPF baseline. In a sensing-based navigation loop, where obstacles are discovered incrementally, I-RAPF achieves an overall success rate of 99.84\%, compared with 97.96\% for Tuned RAPF, while also requiring fewer replanning events, lower planning effort, and shorter executed paths.

The thesis then extends I-RAPF to the swarm setting through a layered Coordinator-Guided Swarm architecture. The proposed method combines a formation-aware I-RAPF planner, a tunnel-flocking coordination layer, and a motion primitive execution layer. The formation-aware planner generates a shared path and formation-radius profile, representing both the desired route and the lateral space available to the swarm. The tunnel-flocking layer converts this route into local desired velocities, while the motion primitive layer maps these velocities to executable non-holonomic rover actions under local safety constraints.

Simulation benchmarks compare the Coordinator-Guided Swarm stack against a Self-Planned Swarm baseline, in which all agents use I-RAPF independently while sharing obstacle detections. The results show that shared sensing alone is not sufficient for robust collective traversal: independently planned routes can still cause incomplete collective arrival, traffic-like interference, and inconsistent route choices. Coordinator-guided formation-aware route management improves full-swarm convergence by providing a common traversal structure. Preliminary real-rover validation on the Lunar Zebro platform further demonstrates integration of I-RAPF into the physical navigation stack and identifies practical deployment limitations related to localization quality and experimental logging.

Overall, the thesis shows that RAPF can be strengthened into a reliable single-agent planner and extended to swarm navigation when route-level formation feasibility is combined with local coordination and primitive-based execution. The results support the central conclusion that robust lunar rover swarm navigation requires both shared obstacle awareness and shared route organization that accounts for the spatial requirements of the swarm.