AI-Based Autonomous Navigation System for Mobile Robots Using Deep Reinforcement Learning

This paper presented NavSim, an AI-Based Autonomous Navigation System for Mobile Robots that implements Deep Reinforcement Learning principles in a fully working, browser-accessible web application. The system provides a complete, reproducible experimental framework comparing DRL navigation strategies against classical planning baselines without requiring specialized robotics hardware.   Experimental results confirm the core finding: PPO achieves the highest navigation success rate (86.7% static, 57.5% dynamic), outperforming DQN and dramatically surpassing A*, whose reliance on static map assumptions causes catastrophic failure in dynamic environments (10% success, 90% collision rate). The 47.5 percentage point gap between PPO and A* in dynamic environments provides direct, quantitative evidence for the practical superiority of learned, sensor-reactive navigation policies over map-dependent classical planning.   The NavSim application demonstrates that meaningful DRL robotics experimentation is accessible without expensive hardware or complex middleware: the system requires only Flask and NumPy for the web demo, while the included Stable-Baselines3 training script provides a genuine path to real RL model training. Future work will extend the system to continuous action spaces, 3D environments, multi-robot coordination, and real TurtleBot3 hardware deployment via ROS integration.

1. Hart, P. E., Nilsson, N. J., & Raphael, B. (1968). A Formal Basis for the Heuristic Determination of Minimum Cost Paths. IEEE Transactions on Systems Science and Cybernetics, 4(2), 100–107. 2. Fox, D., Burgard, W., & Thrun, S. (1997). The Dynamic Window Approach to Collision Avoidance. IEEE Robotics & Automation Magazine, 4(1), 23–33. 3. Mnih, V., et al. (2015). Human-Level Control Through Deep Reinforcement Learning. Nature, 518(7540), 529–533. 4. Tai, L., Paolo, G., & Liu, M. (2017). Virtual-to-Real Deep Reinforcement Learning: Continuous Control of Mobile Robots for Mapless Navigation. Proceedings of IROS 2017. 5. Schulman, J., Wolski, F., Dhariwal, P., Radford, A., & Klimov, O. (2017). Proximal Policy Optimization Algorithms. arXiv:1707.06347. 6. Raffin, A., et al. (2021). Stable-Baselines3: Reliable Reinforcement Learning Implementations. Journal of Machine Learning Research, 22(268), 1–8. 7. Sutton, R. S., & Barto, A. G. (2018). Reinforcement Learning: An Introduction (2nd ed.). MIT Press. 8. Long, P., et al. (2018). Towards Optimally Decentralized Multi-Robot Collision Avoidance via Deep Reinforcement Learning. ICRA 2018. 9. Koenig, N., & Howard, A. (2004). Design and Use Paradigms for Gazebo, an Open-Source Multi-Robot Simulator. IROS 2004. 10. Flask Documentation. (2024). Flask Web Framework. https://flask.palletsprojects.com 11. Gymnasium Documentation. (2024). Farama Foundation. https://gymnasium.farama.org