α-fair heterogeneous-agent reinforcement learning for mixed-capability teams

arXiv·low signal

This paper (arXiv 2606.13076, submitted June 12) presents α-fair learning mechanisms for multi-agent teams whose members have different capabilities, tuning how reward and effort are distributed across stronger and weaker agents. It addresses a practical gap in cooperative multi-agent RL where naive optimization concentrates work on the most capable agent. The fairness lens is relevant to orchestration designs that mix premium and cheap models in one workflow.