SOLUTION OF AN INTEGRATED TRAVELING SALESMAN AND COVERAGE PATH PLANNING PROBLEM BY USING A GENETIC ALGORITHM WITH MODIFIED OPERATORS

Abstract

Coverage path planning (CPP) is a fundamental task that is conducted in many applications for machining, cleaning, mine sweeping, lawn mowing, and performing missions by using unmanned aerial vehicles such as mapping, surveillance, search and rescue, and air-quality monitoring. An approach for conducting CPP for a known environment with obstacles involves decomposing the environment into cells such that each cell can be covered individually. The visiting order of the cells can then be decided to connect those intracell paths together. Finding the shortest intercell path that visits every cell and returns to the origin cell is similar to the traveling salesman problem (TSP). However, an additional variation from TSP that should be considered is that there are multiple intracell paths for each cell. These paths result from different selections of entry and exit points in each cell and thus affect the intercell path. This integrated TSP and CPP problem is known as TSP-CPP and is similar to the TSP with neighborhoods (TSPN). To solve TSP-CPP, one must simultaneously determine the visiting order of sites with minimal repetition and the transition points of each visiting site. The current approaches for solving TSP-CPP are as follows: (i) adapting dynamic programming (DP) for TSP to TSP-CPP, which is excellent for obtaining the optimal route and (ii) determining the optimal route by conducting a brute force enumerative search on entry and exit point combinations for every cell and then solving each combination of entry and exit points with a TSP solver. For large numbers of cells, approaches (i) and (ii) both suffer from exponential complexity and are impractical for complex environments. In this study, we proposed an appropriate genetic algorithm implementation for TSP-CPP to achieve an optimal balance between time efficiency and path optimality to eliminate the curse of dimensionality in DP. Our approach is demonstrated to find the true optimal solution as DP in all simulation environments that can be solved by both DP and GA, and GA is one hundred times faster than DP approach for maps decomposed with large cell number.