Request to add paper: Enhancing Autonomous Vehicle Training with Language Model Integration and Critical Scenario Generation

[KethanR]

Hello HighwayEnv Team,

The Transport Systems and Logistics Group at Imperial College London leveraged your HighwayEnv environment to develop a closed-loop autonomous vehicle training framework to enhance AV safety resilience (enhances the agent’s performance and improves the learning rate). It uses real-world traffic dynamics, critical scenario generation, surrogate safety measures, and LLM analysis to target an RL agent's learning gaps to expose the agent to challenging scenarios during training.

Needless to say, we would be grateful if you add our paper to the list of publications & preprints.

Regards,
Kethan

P.S. A huge "thank you" for adding your environment to the open-source world!

[eleurent]

Hi Kethan,
Thanks for the PR, and congrats to the team for the very interesting paper! (those are really substantial performance improvements!)

[KethanR]

Hi Kethan, Thanks for the PR, and congrats to the team for the very interesting paper! (those are really substantial performance improvements!)

Kind of you to go through our paper, and for acknowledging the improvements. We will continue using HighwayEnv as a preliminary testbed for further experiments. Cheers for accepting the PR! @eleurent