Max-Plus Approach Based Intelligent Coordinated Transmission for Robot Swarms

Communication is vital to complete tasks coordinately for robot swarms. In this paper, we investigate massive MIMO enabled robot swarms. Specifically, for the robot swarms, the transceiver beamforming not only needs to maximize the rate, but also has to restrict the interference on other receivers. Therefore, the transceiver design of robots is critical to optimize the sum-rate performance under the restriction of the interference on the a specific robot. Currently, only exhaustive search is able to provide the optimal solution for the problem, whereas its complexity is unacceptable. In this paper, to address the intractable issue, based on the max-plus approach, we consider each transmitter or receiver as an independent decision agent, and all robots coordinately choose the optimal joint beam combination by max-plus algorithm. In the multi-agent framework, each agent learns the policy of choosing analog beam by reinforcement learning (RL). Furthermore, to improve the learning efficiency of RL and reduce the transmission latency, we exploit the efficient ELM network to replace the deep network of deep RL, and propose a ELM-based RL method to conduct the transmission between robots in robot swarm. Analysis and simulation results reveal that, the proposed method is able to achieve a near-optimal sum-rate performance, while the complexity is acceptable.