Robots Learn Autonomous Navigation using Deep Reinforcement Learning | G2GCARI-CP | MARL Simulator

This video demonstrates 500 episodes from the deployment phase of Go-to-Goal with Collision Avoidance and Random Initialization (G2GCARI) experiment trained using common policy approach. Four randomly spawned cooperative non-holonomic robotic agents were trained using deep reinforcement learning to go to their goals without colliding with obstacles (walls) or other agents in the environment.