The BRIO Labyrinth Game - A Testbed for Reinforcement Learning and for Studies on Sensorimotor Learning
In Proceedings of the Multidisciplinary Symposium on Reinforcement Learning, (MSRL-09), 18.6.-19.6.2009, Montreal, o.A., 2009.
Applying Reinforcement Learning (RL) in the context of robot learning is a challenging problem mainly due to the continuous state and action spaces, inherent noise in sensors and actuators, and a lack of full observability due to hidden states. Furthermore, learning in the physical world is expensive because of material deterioration and the requirement of continuous human supervision to avoid that the robot gets damaged during learning. Thus, there is a high demand for robotic benchmark scenarios that allow to test existing and develop new RL algorithms suited for robotic applications.