Collective Autonomous Vehicle Routing Needs Social Awareness for Coexistence with Human Drivers
Social marginal-cost rewards make autonomous-vehicle routing fairer to human drivers in mixed traffic.
The paper adds a marginal-cost social term to autonomous-vehicle routing rewards learned with multi-agent reinforcement learning. Marginal cost measures an AV's effect on other road users by comparing travel times with a counterfactual in which that AV is removed. On the Ingolstadt and Provins networks, selfish routing with 40% AVs raised other vehicles' travel times by 1.7% in Ingolstadt, falling to 1.1% with the social reward, while AV travel times also improved. Computing marginal cost only for the most influential AVs reduced simulations with similar performance; overweighting the social term hurt system performance.
- Marginal cost compares travel times with and without each AV.
- In Ingolstadt, social routing cut other vehicles' delay from 1.7% to 1.1%.
- Social AVs were also better off than selfish ones.
- Scoring only influential AVs cut compute with similar performance.
From the original article
1 min read at arxiv.orgCollective Autonomous Vehicle (AV) route choices in real-world traffic networks can create externalities for Human-Driven Vehicles (HDVs), as routes that benefit AVs may increase travel times for other road users. To address this, we introduce a social component, the marginal cost, to the selfish travel time AV reward.
By Anastasia Psarou · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.