The goal of this research is to advance the intelligence and practicality of robot-assisted Mobile Ad-Hoc Network (MANET) by integrating reinforcement learning into network development and wireless communication. Specifically, 1) Under the background of Delay tolerant network (DTN), how to improve the effectiveness of the real-time routing in a multi-autonomous robot-assist MANET under uncertain and complex environment, even in the more extreme case, under disturbance from attacker. 2) Under the background of harsh millimeter-wave (mmwave) communication, how to maximize the practical communication quality of multi-user mmWave MIMO systems with uncertain or interference environment through effectively using the mobility of multi-robots as dynamic relays and machine learning techniques.

Networking: We presents the SaRE-MANET (Situation-aware Robot Enhanced Mobile Ad-hoc Network) routing protocol that adopts online reinforcement learning to supervise the mobility of multi- robots as well as handle the uncertainty and complexity of a harsh DTN environment. First, a set of mission-oriented metrics is introduced to describe the interrelation between network quality and multi-robot mobility. Then, a distributed multi-agent reinforcement learning algorithm is developed. It online optimizes the SaRE-MANET routing protocol as well as the mobility of multi-robots by using practical mission-oriented metrics. By inheriting SaRE-MANET, a GT-SaRE-MANET (Game Theoretic Situation-aware Robot Enhanced Mobile Ad-hoc Network) routing protocol is proposed that adopts online game theoretic reinforcement learning to design the mobility of multi-robot as well as handle the uncertainty and potential physical attack or cyber attack.

Wireless communication: We investigate how to use advantaged machine learning to enhance the millimeter wave MIMO system by using robots as communication relays under uncertain environments, including unknown interference. A novel Multi-Robot Enhanced Intelligent Multi-User Millimeter-Wave MIMO (MREI-MU-MIMO) solution that adopts a dynamic codebook based beamforming training protocol and online reinforcement learning to supervise the mobility of multi-robot-relays as well as handle the serious effects of the uncertain environment. An advanced Game-Theoretic Based Intelligent Multi-User Millimeter- Wave MIMO (GT-MU-MIMO) system that, not only inherits the MREI-MU-MIMO framework, but also effectively avoids signal interference from unknown radio jamming attack, etc.

The effectiveness of all proposed designs has been demonstrated through computer-aid simulation and hardware experiments.