Target Interception in Uncertain Environment Using ART2-Based Reinforcement Learning

Tracking moving targets using mobile robots is a crucial aspect of robotics. This paper presents a novel approach for tracking a moving target in an uncertain environment with various obstacles, even when the target’s trajectory and speed are continuously changing and unknown. The proposed method utilizes reinforcement learning, a widely used technique for motion planning problems. However, applying reinforcement learning in uncertain dynamic environments poses a challenge due to the continuous state space. To address this issue, our algorithm employs an ART2 neural network for classifying the state space. Additionally, to enhance the speed of reaching the target, a point is predicted based on the target’s speed, direction, and the robot’s speed. The robot then selects its next move to approach this predicted point while avoiding contact with both static and dynamic obstacles. Simulation results demonstrate the efficiency of the algorithm, as the robot successfully reaches the target without colliding with any obstacles.