Distributed Agents for Autonomous Spacecraft

Space missions have evolved considerably in the last fifty years in both complexity and ambition. In order to enable this continued improvement in the scientific and commercial return of space missions new control systems are needed that can manage complex combinations of state of the art hardware with a minimum of human interaction. Distributed multi-agent systems are one approach to controlling complex multisatellite space missions. A distributed system is not enough on its own however, the spacecraft must be able to carry out complex tasks such as planning, negotiation and close proximity formation flying autonomously. It is the coupling of distributed control with autonomy that is the focus of this thesis. Three contributions to the state of the art are described herein. They all involve the innovative use of multi-agent systems in space missions. The first is the development of a multi-agent architecture, HASA, specifically for space missions. The second is to use embedded agents to autonomously control an interferometric type space telescope. The third is based on software agents that coordinate multiple Earth observation missions coupled with a global optimisation technique for data extraction. The HASA architecture was developed in reaction to the over generality of most multi-agent architectures in the computer science and robotics literature and the ad-hoc, case-by-case approach, to multi-agent architectures when developed and deployed for space missions. The HASA architecture has a recursive nature which allows for the multi-agent system to be completely described throughout its development process as the design evolves and more sub-systems are implemented. It also inherits a focus on the robust generation of a product and safe operation from architectures in use in the manufacturing industry. A multi-agent system was designed using the HASA architecture for an interferometric space telescope type mission. This type of mission puts high requirements on formation flying and cooperation between agents. The formation flying agents were then implemented using a Java framework and tested on a multi-platform distributed simulation suite developed especially for this thesis. Three different control methods were incorporated into the agents and the multi-agent system was shown to be able to acquire and change formation and avoid collisions autonomously. A second multi-agent system was designed for the GMES mission in collaboration with GMV, the industrial partner in this project. This basic MAS design was transferred to the HASA architecture. A novel image selection algorithm was developed to work alongside the GMES multi-agent system. This algorithm uses global optimisation techniques to suggest image parameters to users based on the output of the multi-agent system.