Blockchain helps teams of agents coordinate reliably during attacks
BE-CTDE: Blockchain-Enabled Multi-Agent Reinforcement Learning with Anomaly Detection and Token Economics
Multiagent SystemsCryptography and SecurityMachine LearningNetworking and Internet Architecture
Summary
When many smart devices or agents work together, sometimes some may cheat or act strangely, breaking the teamwork. The authors propose using blockchain technology to keep an honest record of what each agent does and to reward or punish them based on their behavior. This system links how well an agent behaves with how much influence it has on the overall learning process. They tested this idea on a path-planning task and found it helped the agents cooperate much better, even when many were attacking or misbehaving.
What this means in practice
- •For iot system engineers: Improve coordination and trustworthiness of connected devices by linking behavior verification to learning updates using blockchain and economic incentives.
- •For multi-robot teams: Deploy path planning with enhanced defenses against misbehaving robots through blockchain-based accountability and token-driven rewards and penalties.
- •For distributed gaming platform developers: Create multiplayer game environments resistant to cheating by combining blockchain tracking with learning-based strategy validation and incentives.$Commercial implications: Enables sale of cheat-resistant gaming platforms where player actions and rewards are transparently managed on blockchain.
Authors
Chih-Wei Huang, Hsin-Chieh Lee, Yun-Chen Yu, Ibrahim Althamary, Rongxing Lu, Phone Lin
Abstract
Multi-agent reinforcement learning supports cooperative decision-making in Internet of Things systems, but strategic deviations can disrupt coordination and undermine confidence in the training process. This paper proposes Blockchain-Enabled Centralized Training and Decentralized Execution (BE-CTDE), which uses blockchain as an active training-management layer. A behavioral verifier compares executed trajectories with shared-policy recommendations; its flags drive a token mechanism combining slashing, burning, rewards, and compensation. Retained balances determine each agent's contribution to the global policy update, connecting auditable behavior to learning influence. We derive a one-round incentive condition that accounts for action-dependent compensation and the payoff from deviation, and a finite expected exclusion-time bound under explicit conditional detection and accounting assumptions, showing perfect detection is unnecessary. An exact finite-population capacity bound characterizes the limits of the population-relative detector. In multi-agent path finding experiments, BE-CTDE outperforms a state-of-the-art learning-based path planner with substantially higher success rates under attack, even when adversaries form the majority. It also outperforms partially defended baselines in all configurations, with collision rates up to 35% lower than under detection without economic consequences and up to 43% lower than under conventional stake slashing. Together, these results characterize how behavioral verification and token feedback can support accountable cooperative learning.