Follow-Me in Micro-Mobility with End-to-End Imitation Learning
Sahar Salimpour, Iacopo Catalano, Tomi Westerlund, Mohsen Falahi, Jorge Peña Queralta
TL;DR
The paper tackles follow-me navigation for assistive micro-mobility in crowded environments by proposing an end-to-end imitation-learning framework that maps $UWB$ observations to velocity commands $v$ and $\\omega$. Implemented on DAAV's holonomic wheelchair, the system employs two $UWB$ tags for robust leader tracking and 2D/3D LiDAR for obstacle avoidance, trained from real demonstrations with a moving window of 2 s. Among regression models, the LSTM with angle+range inputs delivers superior angular tracking and stable following, validated through real-world trials and production integration. The work demonstrates the viability of end-to-end imitation learning to optimize user-centered metrics like comfort in dynamic indoor and airport environments.
Abstract
Autonomous micro-mobility platforms face challenges from the perspective of the typical deployment environment: large indoor spaces or urban areas that are potentially crowded and highly dynamic. While social navigation algorithms have progressed significantly, optimizing user comfort and overall user experience over other typical metrics in robotics (e.g., time or distance traveled) is understudied. Specifically, these metrics are critical in commercial applications. In this paper, we show how imitation learning delivers smoother and overall better controllers, versus previously used manually-tuned controllers. We demonstrate how DAAV's autonomous wheelchair achieves state-of-the-art comfort in follow-me mode, in which it follows a human operator assisting persons with reduced mobility (PRM). This paper analyzes different neural network architectures for end-to-end control and demonstrates their usability in real-world production-level deployments.