Performance & Cooperation Among Multiple Autonomous Space-based Agents

Autonomous planning of space-based unmanned craft is at the cutting edge of science and technology. Problems associated with planning arise when multiple independent agents each on separate spacecraft must work in a coordinated fashion for a combined mission. This proposal examines the ramifications of cooperating and independent autonomous agents controlling unmanned craft toward a common cause. In particular, the health of an individual agent as evidenced by state variables are not domain specific; health should be added to the representation of state of the agent which can be communicated to others agents so that they may infer the ability of that agent to carry out its tasks. This information is also used in the scheduling and planning of the remainder of the agents who can plan for an agent to go offline if its state indicates this is a possibility. Collaborating agents are also likely to make mistakes. Pattern analysis can be used to learn from past plan deviations to avoid those situations that are likely to cause failure. Finally having additional information about the state of the software is useful in determining why an agent failed if the actual satellite cannot be recovered in a timely manner.