ROS 2 Real-Time Perception-to-Control for Non-Holonomic Wheelchairs

SCAI Lab, ETH Zürich · Feb 2025 – Nov 2025

Built a real-time perception-to-control pipeline (C++/ROS 2) on NVIDIA Jetson, integrating dual Livox LiDARs, IMU-based state estimation (FAST-LIO), and multi-LiDAR point-cloud fusion and preprocessing.
Developed a reactive crowd-navigation and collision-avoidance stack (convex bounding-shape constraints) for powered wheelchairs, including ROS 2 teleoperation interfaces for manual control and on-vehicle testing.
Integrated pedestrian detection (M-detector) with RVO2-based reactive collision avoidance for safe navigation among many moving agents.