Explore a Stanford seminar on multi-agent systems and traffic negotiation presented by Mrdjan Jankovic of Ford Research. Delve into the complexities of coding negotiation algorithms for autonomous driving and mobile robotics. Learn about Control Barrier Function (CBF) based methods for collision avoidance and their advantages in solving non-convex obstacle avoidance problems. Compare six different CBF-based control policies for a 5-agent, holonomic-robot system, examining their effectiveness in collision avoidance and preventing gridlocks. Analyze the correlation between gridlock prevention and system stability, illustrated through extensive simulations and a vehicle experiment. Gain insights into decentralized multi-agent controllers, centralized CBF controllers, and the PCCA algorithm. Investigate the causes of gridlocks from a stability perspective and explore potential solutions, including the impact of lower barrier bandwidth on traffic flow improvement.
Overview
Syllabus
Introduction
Objective - unstable feedback loop? ord
Why CBFs? Short answer - convex QP
CBF based safety filter
Barrier margin for robustness
Robust Control Barrier Functions
Turning obstacles into barriers
CBF based obstacle avoidance
Traffic flow and gridlocks
Avoiding interacting obstacles
Decentralized multi-agent controllers
Centralized CBF Controller
Co-optimization and CCS
PCCA algorithm guarantees
5 agents Monte Carlo Simulations
Comparison of CBF based methods
Deadlock resolution
Cause of gridlocks - stability?
DR: simulation perspective
Centralized and PCCA equilibrium analysis
PCCA: simulation perspective
Properties of CBF algorithms
Some MA unstable modes are undesirable
Lower barrier bandwidth may improve flow
Conclusion
Predictor-Corrector for Coll. Avoidance
Taught by
Stanford Online
