Learning to understand tasks for mobile robots

We propose a way to represent the environment by storing observations taken in that environment, together with their task-related ´values´. This representation allows for robots being taught by human instructors based on rewards and punishments. We show that the robot is able to learn to execute different tasks. The results of training can be interpreted to gain understanding about the environment in which the robot has to operate. So instead of first modeling the environment and using this to execute the task, we first start by learning to execute the task and use this to obtain knowledge about the environment.