Multi-agent meta-reinforcement learning based on contrastive predictive coding—For rapid adaptive traffic control after emergencies
Wei Jiang, Xianwei Huang, Genjie Wang, Ruting Li, Shanshan Tang · 2026
Read Multi-agent meta-reinforcement learning based on contrastive predictive coding—For rapid adaptive traffic control after emergencies by Wei Jiang, Xianwei Huang, Genjie Wang, Ruting Li, Shanshan Tang (2026) 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/plos-pone-0357779.txt.
Description
Wei Jiang and coauthors published a 2026 paper in PLOS ONE on traffic signals that can clear an ambulance through a junction without blocking everyone else. Standard controllers lean on hand-built traffic features and stumble on new emergency patterns. Their framework teaches intersection agents with meta-reinforcement learning, so policies adapt from few examples, and compresses traffic into compact states using contrastive predictive coding. Agents share information weighted by distance and delay, with a priority layer scaling responses to severity.
In simulation, the system cut emergency response time sharply while keeping regular vehicles' waits contained, and high-priority runs succeeded at very high rates. Removing the predictive coding module hurt performance most, marking it as the load-bearing piece. Analysis credited the full design with a large share of the response-time gain. It is presented as a controller that learns fast enough for emergencies it has never seen.
Catalog record
| CREATOR | Wei Jiang, Xianwei Huang, Genjie Wang, Ruting Li, Shanshan Tang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 72861 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |