Intelligent Path Control of Autonomous Agvs Using Reinforcement Learning Approach for Flexible Manufacturing Industries (Fms)

In the automated manufacturing industry, material handling and transportation are carried out using automated guided vehicles (AGVs) to improve production efficiency and accuracy. However, the path of the AGVs is predefined, ensuring the timely delivery of the materials between the workstations. Industrial scenarios where the movement of materials and human resources is highly dynamic increase the challenges of the AGVs to follow the predefined path and navigate. This paper proposes a novel hybrid navigation strategy for AGVs to estimate optimal collision-free paths using a dynamic programming approach that incorporates artificial intelligence techniques to increase their autonomy and flexibility. Costs for the available paths are predefined considering the distance, and the time is updated to the network, enabling the AGVs to choose an optimal path using Markov Decision Process (MDP). Once the optimal path is determined, the AGVs use the TEB planner to navigate to the intermediate goal points evaluated in the optimal path. The proposed algorithm also classifies the detected algorithm as a static and dynamic obstacle based on the laser scan data acquired and reacted accordingly to proceed further in its travel. Extensive simulations and real hardware experiment results show that the proposed path planning of the AGVs based on a dynamic programming approach not only effectively enhances single AGV path planning but also helps the other AGVs preceding it avoid the obstacle path, which effectively increases the overall performance of real-time industrial multi-robot material handling approaches.