Daniel Polani

 

 

Title: James Bond Meets RoboCop: The Artificial Intelligence Challenge in Complex Multiagent Simulation Scenarios

 

Abstract: In recent research it has become increasingly clear that, to boost the
principled exploration of viable and powerful Artificial Intelligence (AI) strategies, particularly for robotics and autonomous agents, a strong incentive derives from using adequate scenarios. Suitable scenarios usually display different aspects or levels at which a problem has to be solved, where an increasing number of new levels unfold on closer inspection and the different levels also interact with each other. This is one of the implicit reasons for the popularity of Embodied Robotics as opposed to simulated autonomous agent environments.

It turns out that the use of competitive multiagent scenarios allows to vigorously reintroduce this aspect into simulation studies while at the same time taking additional advantage of the specific possibilities offered in simulated environments, e.g. reproducibility, controllability' and monitoring capability ('observability').

The talk will introduce paradigmatic examples of such scenarios e.g. ant colonies and robotic soccer), discuss selected AI challenges posed by them and present approaches to tackle these challenges.