BE-CTDE: Blockchain-enabled multi-agent reinforcement learning combining anomaly detection and token economy

BE-CTDE uses blockchain as a training management layer for multi-agent reinforcement learning, constraining policy deviation in collaboration. A behavior verifier compares actual trajectories with shared policy recommendations, and flags drive a token mechanism of slashing, burning, rewards, and compensation. In multi-agent pathfinding experiments, even when attackers are in the majority, the success rate under attack is substantially higher than that of state-of-the-art learning-based planners, and the collision rate is up to 35% lower than detection without economic consequences and up to 43% lower than conventional staking-based slashing.