Cooperative Multi-AUV Data Collection Method Based on Staged Deep Reinforcement Learning With Distributed Negotiation | Synapse