Learning Bidding Strategies for Karma Economies in Realistic Traffic Settings with Multi-Agent Reinforcement Learning

Karma is a non-monetary resource-allocation mechanism that prioritizes users' needs rather than their financial power. Monetary pricing can effectively reduce congestion by imposing charges on specific road segments, but it may be unfair by favoring higher-income individuals. Prior work has shown that in this context, Karma can achieve similar efficiency while yielding fairer outcomes; however, it has been demonstrated only in a deterministic setting. Demonstrating Karma's applicability under more realistic traffic conditions is therefore important for real-world implementation. Additionally, experimental evidence suggests that humans may struggle to execute optimal bidding strategies in Karma economies. In this paper, we demonstrate the use of Multi-Agent Reinforcement Learning (MARL) to train automated bidding agents for travelers. In a microscopic traffic simulation case study, we show that MARL agents learn effective bidding strategies that yield fairer travel outcomes for drivers than those achieved under monetary pricing schemes.