Representation Learning in Partially Observable Environments using Sensorimotor Prediction
This work addresses the challenge of autonomous exploration and action for agents in noisy, ambiguous environments, though it appears incremental as it builds on existing sensorimotor prediction methods.
The paper tackles the problem of learning compact sensory representations from noisy, high-dimensional observations in partially observable environments by using sensorimotor prediction. The result is a model that integrates sensorimotor information over time to project it into a representation useful for prediction, demonstrated through a simple example highlighting the roles of motor and memory.
In order to explore and act autonomously in an environment, an agent needs to learn from the sensorimotor information that is captured while acting. By extracting the regularities in this sensorimotor stream, it can learn a model of the world, which in turn can be used as a basis for action and exploration. This requires the acquisition of compact representations from a possibly high dimensional raw observation, which is noisy and ambiguous. In this paper, we learn sensory representations from sensorimotor prediction. We propose a model which integrates sensorimotor information over time, and project it in a sensory representation which is useful for prediction. We emphasize on a simple example the role of motor and memory for learning sensory representations.