概要
Personal mobility devices, such as electric wheelchairs, play a crucial role in assisting individuals with mobility difficulties. However, joystick-based manual operation can be challenging, especially in crowded spaces and narrow pathways, highlighting the need for improved safety and comfort. This study proposes a path generation method using Inverse Reinforcement Learning (IRL) and Q-learning. By analyzing expert driving trajectories, we estimate a reward function that reflects the expert’s intentions during navigation. Based on the estimated reward function, optimal paths are generated. Verification experiments using subject trajectory data confirmed the effectiveness of the proposed method. Additionally, we demonstrated that the learned parameters can be applied to unknown environments where training data is unavailable. It is expected that the proposed method will enable adaptive path planning in large-scale commercial facilities and airports and provide more comfortable mobility assistance for users by granting autonomous driving in the future.
