Person Following Robot

This project implements a robust person-following behavior for a robot (simulated or real) using YoloX with ByteTrack for person detection/tracking and a PID-based controller for smooth following.

Key Features

• Person Detection & Tracking: Uses YoloX for high-performance detection and ByteTrack for robust multi-object tracking.
• Depth-Based Localization: accurate distance estimation using RGB-D camera data (e.g., RealSense).
• Smooth Control: Implements a PD (Proportional-Derivative) controller for both linear and angular velocities to ensure smooth following and rotation.
• Safety: Includes handling for "Person Lost" scenarios and collision avoidance logic.

Architecture

The system consists of three main nodes:

1. person_tracker.py:

   • Subscribes to RGB and Depth camera streams.
   • Runs YoloX inference to detect persons.
   • Uses ByteTrack to assign consistent IDs to detected persons.
   • Calculates the 3D position of the target person relative to the robot.
   • Publishes the target pose to tracked_person/position.

2. person_follower.py:

   • Subscribes to tracked_person/position.
   • Calculates the error in distance and angle.
   • Computes velocity commands (cmd_vel) using a PD controller to keep the person at a desired distance and centered in the view.

Installation & Dependencies

Ensure you have ROS 2 installed (Humble/Iron/Rolling).

1. Dependencies:

   pip install yolox loguru scikit-image thop
   sudo apt install ros-<distro>-vision-opencv ros-<distro>-cv-bridge

2. Build: This is a ROS 2 package. Build it using colcon:

   colcon build
   source install/setup.bash

Usage

1. Launch the System:

   ros2 launch person_following_robot follow_person.launch.py

2. Configuration:

   • Adjust control gains (linear_kp, angular_kp, etc.) in person_follower.py to tune the response.
   • Camera intrinsics and model paths are configured in person_tracker.py.

Credits

• Original Author: Special thanks to malwaru for the original implementation and inspiration.