Person-Following for Indoor Robots

ROS-based real-time person tracking pipeline fusing RGB-D and LiDAR — deployed on Jetson-class hardware at COGA Robotics.

Outcome. Built a real-time person-following system for indoor mobile robots — fusing RGB-D and LiDAR in ROS with Deep SORT tracking, validated under real-world lighting and occlusion conditions on Jetson-class edge hardware.

Role: Intern Researcher · Dates: Dec 2022 – Feb 2023 · Context: COGA Robotics · Stack: Python, ROS, Deep SORT


Overview

Indoor person-following demands reliable tracking across varied lighting, partial occlusion, and constrained compute. The system fuses complementary sensor modalities — RGB-D for visual identity, LiDAR for depth geometry — to maintain stable trajectories where either sensor alone would fail.

Highlights

  • Implemented ROS nodes with RGB-D/LiDAR calibration for accurate cross-sensor alignment
  • Integrated Deep SORT tracker leveraging fused sensor signals for continuous trajectory estimation
  • Conducted lighting and occlusion stress tests to surface failure modes and validate recovery behavior
  • Profiled and optimized runtime on Jetson-class devices for real-time edge deployment
  • Designed modular node interfaces and launch files for maintainable, production-ready deployment