Optimized Area Coverage Path Planning In Complex Environment Using Geometric Q Learning Algorithm

Recently, autonomous vehicles have been used in dynamic environments for various automation applications, e.g., area coverage. Thus these missions require sophisticated learning planning to overcome various challenges, e.g., unforeseen obsta-cles. This work proposes to use a modified Q-learning geometric learning algorithm to generate an optimal path by iteratively exploring and exploiting the state-action space. Moreover, the re-inforcement learning approach obtains informed decisions about the best actions using the online experience and interaction with the surrounding environment. The proposed planning approach is model-free, i.e., it does not require a prior estimate of the distribution of threats on the map. While the encountered threats are avoided by updating the cost matrix during the mission. By integrating geometric learning, the algorithm incorporates spatial awareness, considering distances, obstacles, and other geometric attributes. Thus it improves the efficiency and robustness of the path planning process in complex environments. In this work, the simulation results validate the proposed planning algorithm. Various scenarios, e.g., area coverage, have been considered using different relevant information of the threats and targets.