In our group we study and develop algorithms for autonomous agents. Although the research topics listed below are very diverse, the research challenges in each application domain are similar:

  • A balance must be found between the interests of the individual agents, and the welfare of the system as a whole, assuming each agent will first and foremost pursue its own goals.
  • Given bounds on the computational capabilities of the agents, and on their ability to gather information either through perceiving the environment or communicating with other agents, we are nonetheless keen to prove performance guarantees (lower bounds) on the performance of our algorithms and coordination mechanisms.
  • Our research is aimed at enabling solutions to real-world problems; in all of the real-world problems we study, uncertainty surrounding developments in the environment plays a major role. Therefore, our algorithms and coordination mechanisms should strive for robustness, so good quality solutions are produced even when unexpected incidents occur.
Wordcloud of Algorithmics publications 2012-2016
Automated guided vehicles at the Hamburg container terminal

In context-aware route planning, the aim is to find a set of routes for autonomous vehicles that are both efficient and conflict-free. Conflicts can be not only collisions, but also deadlock situations. Route planning is done by agents reserving resources of the infrastructures for the time intervals of their intended usage.

A NedTrain workshop

At NEDTrain, trains come into the shop for repairs and regular check-ups. Prior to inspection, there is obviously uncertainty surrounding the set of repair activities that must be undertaking. A good repair schedule is one that minimizes both the time the trains are in for repairs, and the manpower to perform the repairs in a timely fashion.

When the decisions of multiple organisations are related, the optimal approach would be to solve their combined planning and scheduling problem centrally. However, this is usually impossible and undesirable. We aim to find alternative mechanisms for reaching near optimal plans and schedules without sharing all information and such that each organisation maintains its autonomy.

We develop algorithms for multiagent planning under uncertainty and apply them in the smart grid and transportation domains.

Big data can be used to support planning and scheduling decisions. But which algorithms and models provide the optimal balance between computation time and solution quality is an important challenge for our research group.