New approach helps agents share resources without overuse
Learning to Harvest Without Collapse in a Regenerative Commons: A Lagrangian Framework
Artificial IntelligenceComputer Science and Game Theory
Summary
When multiple people or agents share a resource, such as a fishery, they may overuse it, causing depletion. This paper introduces a way to program agents so that they explicitly preserve a set amount of the resource while still getting rewards for harvesting. The authors develop a mathematical framework that balances use and preservation by setting constraints on how much can be taken. They test their ideas in simulated fishing scenarios and show how these constraints influence harvesting and resource levels.
What this means in practice
- •For environmental resource managers: Design policies that ensure sustainable harvesting of shared natural resources under uncertain agent behaviors.
- •For multi-agent system developers: Create agent controllers that balance resource use and preservation using constrained learning techniques.
Authors
Jose Tupayachi, Xueping Li, Soham Das
Abstract
The tragedy of the commons poses a multi-agent safety problem: reward-seeking agents can deplete a shared resource, and cooperation among its users does not itself specify how much must be preserved. We make preservation an explicit requirement by formulating a regenerative commons as a constrained Markov game or a constrained multi-agent MDP with a designer-specified depletion budget. We develop a nonstationary Lagrangian framework that constructs a policy sequence from solutions of unconstrained games or cooperative control problems. Extending earlier time-average constructions, we introduce average-epoch solution concepts for reset episodes with discounted rewards and terminal costs. We prove a reward-independent feasibility certificate, cooperative feasibility and approximate optimality against feasible policy mixtures, and an extension to unbiased sampled costs. For self-interested agents, a constrained Nash certificate quantifies the price-dispersion term introduced by deviations that redistribute budget across epochs. Under the stated assumptions on solver accuracy and multiplier updates, these results give constrained policy-sequence guarantees using solutions of unconstrained problems. Experiments with constrained IPPO and MAPPO in a Gordon-Schaefer fishery examine how depletion budgets shape stock retention, harvest rewards, and price adaptation.