Ego-centric Learning of Communicative World Models for Autonomous Driving

Hang Wang, Dechen Gao, Junshan Zhang · 2025

Read Ego-centric Learning of Communicative World Models for Autonomous Driving by Hang Wang, Dechen Gao, Junshan Zhang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-08149.txt.

Description

Hang Wang and co-authors build CALL, a way for self-driving agents to learn together without drowning in messages. Multi-agent learning suffers when each car sees only part of the road. Sharing everything costs too much bandwidth. Their fix leans on world models. Each agent squeezes its state and intent into a small latent code.

That code is cheap to share with nearby agents. Each car learns in an ego-centric way, feeding on the shared codes to enrich its own world model. A sharper model predicts better, and better prediction plans better. The authors bound the gain from sharing on paper. Tests on driving tasks in the CARLA simulator show clear performance wins.

Catalog record

CREATORHang Wang, Dechen Gao, Junshan Zhang
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE92 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2506-08149

Browse the full library · About the Tower of Babel