Adaptive Reward Mechanisms based on Reinforcement Learning Techniques for Mobile Robots in Unknown Environment

This research addresses the static and dynamic unpredictable nature of their surroundings, which presents challenges for mobile robots operating in uncertain contexts, sometimes hindering their ability to learn appropriate policies. In this paper, a method based on reinforcement learning (RL) techniques is proposed to enhance the learning process of mobile robots through adaptive reward mechanisms. The system continuously modifies the rewards obtained during the learning process with the goal of improving the resilience and adaptability of robot behaviour. This investigation covers various RL algorithms, such as Deep Q-Networks (DQN), Improvement of Proximal Policy Optimization (IPPO) and Actor-Critic techniques, along with reward schemes that are flexible and customized to the specific demands of mobile robot activities. It assesses the effectiveness of these strategies in terms of learning efficiency, flexibility to changes in the environment and robustness to uncertainty through extensive testing in simulated scenarios. The results demonstrate to an adaptive reward method might enhance mobile robots learning skills and enable them to navigate and perform tasks more adeptly in demanding and dynamic environments.