Scholarlink

<p>Autonomous navigation in dynamic environments poses significant challenges due to complex agent–environment interactions, real-time decision-making, and adaptability requirements. We propose the Reinforcement-Imitation Learning for Autonomous Navigation (RILAN) algorithm, which integrates reinforcement learning, imitation learning, and uncertainty estimation for robust and adaptable navigation. RILAN combines the strengths of expert demonstrations and self-learned knowledge, overcoming individual limitations. The modular RILAN architecture comprises an imitation learning module, reinforcement learning module, and an uncertainty-aware policy selector, enabling efficient information processing and dynamic policy selection. Extensive experiments compare RILAN to state-of-the-art algorithms, including E2E-IL, CIL, DQN, and PPO. Results show RILAN outperforms these algorithms in success rate, average time to goal, collision rate, and off-road rate, demonstrating the proposed approach’s efficacy for robust autonomous navigation in applications like self-driving vehicles and robotics.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Bridging the gap: combining expert guidance and self-learned knowledge for efficient and robust autonomous navigation in complex settings

  • 期刊论文

  • 发表日期：

    2025年2月24日

查看全文
  • Dharmendra Kumar

摘要

Autonomous navigation in dynamic environments poses significant challenges due to complex agent–environment interactions, real-time decision-making, and adaptability requirements. We propose the Reinforcement-Imitation Learning for Autonomous Navigation (RILAN) algorithm, which integrates reinforcement learning, imitation learning, and uncertainty estimation for robust and adaptable navigation. RILAN combines the strengths of expert demonstrations and self-learned knowledge, overcoming individual limitations. The modular RILAN architecture comprises an imitation learning module, reinforcement learning module, and an uncertainty-aware policy selector, enabling efficient information processing and dynamic policy selection. Extensive experiments compare RILAN to state-of-the-art algorithms, including E2E-IL, CIL, DQN, and PPO. Results show RILAN outperforms these algorithms in success rate, average time to goal, collision rate, and off-road rate, demonstrating the proposed approach’s efficacy for robust autonomous navigation in applications like self-driving vehicles and robotics.