MAER-Nav: Bidirectional Motion Learning Through Mirror-Augmented Experience Replay for Robot Navigation

2025-03-31

AI Summary

To address the limitation of deep reinforcement learning (DRL)-based navigation—where agents trained solely on forward actions frequently become trapped in narrow spaces—this paper proposes a novel synergistic framework integrating mirror-augmented experience replay with curriculum learning. Unlike prior approaches, it enables end-to-end bidirectional motion policy learning without relying on failure trajectories or redesigning the reward function. The core innovation is a state-action mirroring mechanism that automatically generates high-quality backward navigation experiences during replay. Curriculum learning then progressively enhances policy robustness by increasing environmental complexity. Evaluated in both ROS simulation and real-world robotic platforms, the method achieves a 42% improvement in backward maneuver success rate and a 31% increase in overall task completion rate over state-of-the-art methods, while preserving forward-performance fidelity. This bridges the gap between traditional planning and learning-based approaches in terms of action-space utilization and environmental adaptability.

Technology Category

• Machine Learning: Imitation Learning & Inverse Reinforcement Learning
• Planning, Routing, and Scheduling: Replanning and Plan Repair
• Multiagent Systems: Multiagent Learning

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions

Abstract

Deep Reinforcement Learning (DRL) based navigation methods have demonstrated promising results for mobile robots, but suffer from limited action flexibility in confined spaces. Conventional DRL approaches predominantly learn forward-motion policies, causing robots to become trapped in complex environments where backward maneuvers are necessary for recovery. This paper presents MAER-Nav (Mirror-Augmented Experience Replay for Robot Navigation), a novel framework that enables bidirectional motion learning without requiring explicit failure-driven hindsight experience replay or reward function modifications. Our approach integrates a mirror-augmented experience replay mechanism with curriculum learning to generate synthetic backward navigation experiences from successful trajectories. Experimental results in both simulation and real-world environments demonstrate that MAER-Nav significantly outperforms state-of-the-art methods while maintaining strong forward navigation capabilities. The framework effectively bridges the gap between the comprehensive action space utilization of traditional planning methods and the environmental adaptability of learning-based approaches, enabling robust navigation in scenarios where conventional DRL methods consistently fail.

Problem

• Enhances robot navigation flexibility in confined spaces
• Enables bidirectional motion learning without explicit failure replay
• Bridges gap between traditional and learning-based navigation methods

Innovation

• Mirror-augmented replay for bidirectional learning
• Generates synthetic backward navigation experiences
• Combines curriculum learning with experience replay