Academic advisor: Xiaoming Hu
Sponsor: SSF
The goal of this project is to develop new methods and paradigms for mobile robotic systems using a multi-agent control approach. A crucial aspect that will be studied is how the set of information used by each agent in the system to elaborate solutions influences performance and safety requirements. For a multi-agent system where many decisions are made locally based on local information, a communication scheme whereby every agent knows everything about every other agent may not be supported in time compatible with safety and performance. Therefore, the choice of the information structure is a critical issue here.
There has been a tremendous research interest in multi-agent systems both in control and computer science. In robotics there is also an increased interest in applying the methodologies from multi-agent systems to distributed manipulation, which range from multi-fingered grasping, dual-arm manipulation, to general cooperative mobile robotics. For example, in industrial applications of high manipulability, there has been genuine interest in cooperative manipulation, since such tasks typically cannot be realized by a single arm. In many non-prehensile manipulation tasks such as pushing much research has been done using a multi-agent approach. In particular, some decentralized control algorithms have been developed. Clearly, all the aforementioned applications can be put into the general framework of multi-agent synchronization and cooperation for which there is a vast literature.