“Deep Reinforcement Learning for Autonomous Navigation in Unknown Environments”

Abstract

“Robotic and artificial intelligence face the challenge of autonomous navigation in unknown environments. Then, this research studies the application of Deep Reinforcement Learning (DRL) in intelligent path planning, as well obstacle avoidance. The efficiency of four DRL algorithms dependent on dynamic environment, including, Deep Deterministic Policy Gradient (DDPG), Proximal Policy Optimization (PPO), Soft Actor-Critic (SAC) and Q-learning was investigated and implemented. Finally, the experiments revealed, SAC achieved a success rate of 92.3%, PPO achieved a success rate of 89.7%, while DDPG achieved a success rate of 85.2% and Q-learning achieved a success rate of 78.9%. In shortening navigation time, the proposed models were superior in that SAC reduced path deviation by 24 percent over traditional approaches. It also shows the effect of sensor fusion and adaptive reward function, improving decision making accuracy by 34%. The findings show that hybrid learning models and real-time optimization can significantly increase navigation capability. Nevertheless, computational efficiency and spectral adaptability to rapidly varying environments continue to be challenging aspects. There is still much room for future research by developing real time learning frameworks to boost performance even more and devise energy efficient navigation strategies.