Autonomous Robot Navigation and Efficient Data Collection: Human-Agent Joint Learning and Reinforcement-Based Autonomous Navigation

Our system integrates human operators and robots in a joint learning process to improve robot manipulation skills, reducing human effort and attention during data collection while maintaining data quality for downstream tasks. Challenges in teleoperating a robot arm are addressed by a system that allows human operators to share control with an assistive agent, reducing human workload and ensuring efficient data collection with less human adaptation required. The paper focuses on applying reinforcement learning techniques to achieve autonomous navigation for robots, using Deep Q Networks (DQN) and Proximal Policy Optimization (PPO) to optimize path planning and decision-making processes in dynamic environments. Reinforcement learning techniques such as DQN and PPO are highlighted for their ability to handle high-dimensional state spaces and improve stability and sample efficiency in autonomous navigation. Integrating advanced learning techniques in robotic systems enhances efficiency and adaptability, contributing to more efficient and robust robotic systems with broader applications in various industries, leading to increased automation, reduced operational costs, and enhanced productivity.