Towards a deep feature-action architecture for robot homing

This paper describes a model for robot navigation that uses an architecture similar to an actor-critic reinforcement learning architecture. Contrary to the abundance of models that use two neural networks one for the actor and one for the critic, this model sets up the actor as a layer seconded by another layer which deduce the value function. Therefore, the effect is to have similar to a critic outcome combined with the actor in one network. Hence, the model paves the way for a deep reinforcement learning architecture for future work The reward signal is back propagated through the critic then the actor. At the same time, the features layer have been deeply trained by applying a simple PCA on the whole set of images histograms acquired during the first running episode. The model is then able to shrink the whole architecture to fit a new reduced features dimension. Initial experimental result on real robot shows that the agent accomplished good level of accuracy and efficacy in reaching the goal.