|
It is very challenging to develop a mission critical control for networked heterogeneous UAS that is intelligent and resilient even when implemented into a complex environment with practical constraints such as limited communication, uncertainty system dynamics etc. Meanwhile, lacking applicable decentralized control seriously limits the usage of networked UAS into critical military and civilian missions. Recently, many learning-based decentralized control algorithms have been developed. However, there are two significant limitations, i.e. 1) slow learning convergence speed which cannot catch the environmental changing rate and 2) The gap between mission planning and real-time control. Our proposed algorithm will overcome these challenges and reap the advantages from networked UAS.
Deeply integrating online fast reinforcement learning with real-time networked control, a novel mission critical decentralized resilient and intelligent control algorithm will be developed for network heterogeneous unmanned autonomous systems (UAS) in presence of limited communication, system uncertainties and harsh environment. Different from traditional decentralized control and learning algorithms, proposed design is a real-time applicable optimal and resilient solution that has particularly considered real-time mission completeness, the convergence speed of learning algorithm and the impacts from limited communication, system uncertainties and harsh environment.
|
