Project: ERLARS - International Workshop on Evolutionary and Reinforcement Learning for Autonomous Robot Systems
Researchers: Siebel N. , Pauli J.
Evolutionary and Reinforcement Learning methods are important learning approaches for neural networks and other knowledge representations. They are inspired by nature and known to be used extensively in biological systems. However, so far their use in artificial cognitive systems, e.g. autonomous robot systems, is limited. This is mainly due to the large number of necessary robot actions and/or learning cycles before an acceptable mapping from perceptions to actions is found. Autonomous robots are becoming more and more common even in nonindustrial settings, an example area being toys like Sony's Aibo robotic dog and the new Pleo toy dinosaur. However, they tend to have very limited learning capabilities. These are usually restricted to adjusting a few parameters in an otherwise fixed control strategy that determines how the robot interacts with the environment.In recent years, fast computers have made evolutionary and reinforcement learning more feasible from a computational point of view. Therefore research in these areas has attracted more attention. A number of new and efficient algorithms have shown promising results, albeit many of these still rely on training in simulated environments or in combinations of offline and online learning.
The main goal of this workshop is to bring together researchers and promote work on evolutionary and reinforcement learning methods with a focus on their (future) application in autonomous robot systems. We believe that in order to achieve this a great deal of fundamental research, e.g. on the efficiency of algorithms, is just as important as their practical applications. Therefore contributions are invited both on theoretical and practical results in this area.
More information can be found on the home page of the ERLARS workshop.