Control Barrier Functions
Project Objective
To investigate and implement novel formulations of Control Barrier Functions (CBFs) for multi-robot collision avoidance, focusing on methods that are both optimal and distributed in nature. The goal is to improve safety guarantees and scalability in decentralized robotic systems.
Research Supervisor
Gazebo Codebase:
Python Codebase:
Abstract
This project explores novel Control Barrier Function (CBF) formulations for multi-robot collision avoidance, focusing on scalability and smooth control under distributed settings. We implement two approaches: an optimal distributed CBF that allows robots to compute locally safe controls with minimal coordination, and a log-sum-exp relaxation that replaces the non-smooth minimum operator in safety constraints, resulting in better optimization performance.
The methods are evaluated in a Python-based simulator using second-order (double integrator) robot dynamics, with comparisons to a centralized CBF baseline. Key metrics include collision avoidance, control smoothness, and constraint satisfaction. To validate real-world applicability, the controllers are deployed in a Gazebo + ROS1 Noetic environment with TurtleBot3 agents operating in shared spaces. Compared to the baseline, our methods show improved safety and smoother trajectories, particularly in high-density scenarios.