Safeguarding Multi-Agent Reinforcement Learning for Connected and Automated Vehicles with Recursively Feasible Control Barrier Functions
Safeguarding Multi-Agent Reinforcement Learning for Connected and Automated Vehicles with Recursively Feasible Control Barrier Functions
This research proposes the Backup-Horizon CBF (BH-CBF) method to address safety guarantees in multi-agent reinforcement learning for connected and automated vehicles. The approach ensures recursive feasibility under control bounds and uses exact vehicle and road geometry, avoiding the conservatism of existing methods. It is applied as a reward during pre-training, a runtime filter at deployment, and a training loop filter for fine-tuning on unseen maps. Simulations on real-world roundabout and intersection maps show that BH-CBF prevents all collisions caused by unsafe MARL policies while maintaining safety guarantees at every step. In recorded data from a roundabout with 1264 human-driven vehicles over 115 km, the safeguarded policies reduced median travel time by 14% compared to human drivers without any collisions.
The paper introduces a recursively feasible control barrier function (BH-CBF) designed to ensure safety guarantees for multi-agent reinforcement learning in connected and automated vehicles, validated through simulations on real-world maps and recorded human-driven traffic.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
distance traveled
>115
View original evidence
e guarantee, or fall back to full braking far more often. In recorded roundabout traffic with 1264 human-driven vehicles over 115 km, our safeguarded policies shorten the median travel time by 14% compared with the human drivers, without any collision.
Open source S5
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
e guarantee, or fall back to full braking far more often. In recorded roundabout traffic with 1264 human-driven vehicles over 115 km, our safeguarded policies shorten the median travel time by 14% compared with the human drivers, without any collision.
Open source S5
F to safeguard a MARL policy over its life cycle: as a reward during pre-training, as a run-time filter at deployment, and as a filter in the training loop that keeps fine-tuning on unseen maps collision-free. In simulations on real-world maps of a roundabout and an intersection, BH-CBF avoids all collisions of unsafe MARL policies and keeps its safety guarantee at every step. Existing filters either collide, lose th
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org