Total: 1
The paper presents a novel sample-based algorithm, called C*, for real-time coverage path planning (CPP) of unknown environments. The C* algorithm is built upon the concept of Rapidly Covering Graph (RCGs). The RCG is constructed incrementally via progressive sampling during robot navigation, which eliminates the need for cellular decomposition of the search space. The RCG has a sparse-graph structure formed by efficient sampling and pruning techniques, which produces non-myopic waypoints of the coverage trajectory. While C* produces the desired back and forth coverage pattern, it adapts to the TSP-based locally optimal coverage of small uncovered regions, called coverage holes, that are surrounded by obstacles and covered regions. Thus, C* proactively detects and covers the coverage holes in situ, which reduces the coverage time by preventing the longer return trajectories from distant regions to cover such holes later. The algorithmic simplicity and low computational complexity of C* makes it easy to implement and suitable for real-time onboard applications. It is analytically proven that C* provides complete coverage of unknown environments. The performance of C* is validated by 1) extensive high-fidelity simulations and 2) real laboratory experiments using autonomous robots. A comparative evaluation with seven existing CPP methods demonstrate that C* yields significant performance improvements in terms of coverage time, number of turns, trajectory length and overlap ratio, while preventing the formation of coverage holes. Finally, C* is evaluated on two different applications of CPP using 1) energy-constrained robots and 2) multi-robot teams.