Since 1995, numerous ActorCritic architectures for reinforcement learning have been proposed as models of dopaminelike reinforcement learning mechanisms in the rat's basal ganglia. However, these models were usually tested in different tasks, and it is then difficult to compare their efficiency for an autonomous animat. We present here the comparison of four architectures in an animat as it performs the same rewardseeking task. This will illustrate the consequences of different hypotheses about the management of different Actor sub modules and Critic units, and their more or less autonomously determined coordination. We show that the classical method of coordinations of modules by mixture of experts, depending on each module's performance, did not allow solving our task. Then we address the question of which principle should be applied to efficiently combine these units. Improvements for Critic modeling and accuracy of Actorcritic models for a natural task are finally discussed in the perspective of our Psikharpax project -an artificial rat having to survive autonomously in unpredictable environments.

No abstract

This paper investigates the visual Localization of a mobile platform using a new sampling adaptive bag-of-features patches techniques. The method is specifically developed for the navigation of robots. It is based on the idea of an adaptive dense sampling of images using an optimal multilayer Quadtree decomposition of the image driven by the quantity and homogeneity of the information contained within subpatches. Extracted patches will be of different sizes according to the covered zones in the image. Experimental results carried out on real images in the case of a navigation of a mobile robot using an omnidirectional camera are presented. The method of generating maps will be briefly introduced. Large amount of measures in different cases of noise and occlusion will be presented showing the robustness of the method.

No abstract

No abstract