A Generic Domain-Configurable Planner for Autonomous Operation of Multi-agent Space Systems

Automated planning is the area of Artificial intelligence (AI) which provides reasoning for the selection of actions executed by machines so as to attain a specified goal.  A recent trend in the research community is to apply automated planning in multiple agent systems that work together in a coordinated fashion in order to attain highly complex space mission goals.  Even though automated planning and  scheduling algorithms are mature in industrial scenarios and robotics, little consideration has been  given for its application to multiple-agent space applications. Traditional classical planning algorithms might fail in multi-agent planning due to the sheer size of the search space and pruning down it will result in highly computational requirements.  Our research goal is to develop a planner for highly cooperative multiple-agent autonomous space  systems which can be used for different scenarios such as satellites and rovers.  We propose a Hierarchical Task Network Planner (HTN) for multi-agent planning and BDI framework for agent based modelling and describe a prototype being developed.  The planner is made domain  configurable for different multi-agent missions by keeping the heuristics and planning algorithm less dependent on the domain knowledge. In short developing a planner that can be applied to different missions will help in attaining fully reconfigurable autonomous system and also reduces the gap between AI planning methods and space engineering domain.

Thursday, 18 April, 2013 - 12:30 to 14:00
Raveesh Kandiyil
Presenter(s) biography: 

Raveesh Kandiyil is a PhD student at the Surrey Space centre and is being supervised by Prof.Dr.Yang Gao. His current research interests are Multi-agent planning, AI Planning and Spacecraft Autonomy.