Optimal Path Planning of Swarm Robots Using Multi Agent Deep Reinforcement Learning in Dynamic Environments

The problem of optimal path planning among swarm robots under dynamic conditions is a critical problem since obstacles cannot be predicted, inter-robots can collide and coordinated navigation is required. The traditional approaches to path planning, including A-, D-, potential field-, rapidly exploring random trees (RRT), and particle swarm optimization (PSO), are not always effective in the multi-agent dynamic environment, which results in inefficiency of trajectories and the higher probability of collisions. To overcome these difficulties, this paper suggests a Multi-Agent Deep Reinforcement Learning-based Swarm Path Planning (MADRL-SPP) framework, which would allow the swarm robots to navigate in an adaptive way, with zero collisions, and consuming minimal energy. The suggested MADRL-SPP framework describes every robot as an intelligent agent that interacts with the environment and other agents and benefits collective trajectories using the method of reward-based learning. The simulations that are undertaken using MATLAB are performed within dynamic operating environments, where obstacles are moving, the speed of the robots is heterogeneous and there are communication constraints. Performance analysis takes into account efficiency of the paths, the collision rate, convergence rate, the energy use and the scalability. Comparative analysis shows that MADRL-SPP is greatly superior to the traditional methods, such as A-, D-, potential field, RRT-, and PSO by being up to 32 percent more efficient in path, 45 percent less colliding, and converges quicker in dynamic conditions. The suggested framework can provide a scalable and powerful approach to real-time multi-agent navigation, which demonstrates the prospects of the combination of deep reinforcement learning with swarm robotics in intricate and unpredictable settings.