Predictive Triggering for Multi-agent Systems
Master thesis project on intelligent resource allocation in distributed control of multi-agent systems
Multi-agent systems are composed of multiple intelligent agents acting and inter-acting autonomously in a complex environment. Applications include, but are not limited to, multi-robot systems, autonomous driving or quadcopter swarms flying in formation. In all three settings, it would be beneficial for the agents to communicate. In the case of multiple robots working together and the swarm of quadcopters, sharing information is important for coordination. In autonomous driving, communication between the vehicles allows for adaptive traffic control, which may lead to reduced fuel consumption and fewer traffic jams.
If information has to be exchanged over a common, typically wireless, network, communication becomes a shared and limited resource. Especially in settings with many agents, the bandwidth typically does not support simultaneous communication of all agents. Therefore communication has to be limited to the necessary instants and resources have to be allocated in a reasonable way.
The need for resource efficiency has lead to an increasing interest in event-based control algorithms since the beginning of the century. In event-based control, information is exchanged if certain events occur (e.g. a control error growing too large). In most of these algorithms, however, the agents decide instantaneously if they need to communicate, thus freed resources cannot be reallocated. In this project, we aim for algorithms that can predict future communication demands and, in this way, allow the communication system to reconfigure and allocate the limited resources to the agents which need to transmit information. At the same time, we want to guarantee good control performance.
We are looking for outstanding students who are eager to do their Master thesis on a challenging research project in the area of distributed control. The project involves fundamental theoretical developments in the area of networked and distributed control, and can also be extended to experimental validation on real-world platforms.
More information: