Motor learning involves a widespread brain network including the basal ganglia, cerebellum, motor cortex, and brainstem. Despite its importance, little is known about how this network learns motor tasks and which role different parts of this network take. We designed a systems-level computational model of motor learning, including a cortex-basal ganglia motor loop and the cerebellum that both determine the response of central pattern generators in the brainstem. First, we demonstrate its ability to learn arm movements toward different motor goals. Second, we test the model in a motor adaptation task with cognitive control, where the model replicates human data. We conclude that the cortex-basal ganglia loop learns via a novelty-based motor prediction error to determine concrete actions given a desired outcome, and that the cerebellum minimizes the remaining aiming error.

[This corrects the article DOI: 10.1371/journal.pcbi.1011024.].

We present a distributed and dynamic model of visual attention based on the continuum neural field theory that allows to sequentially focusing salient locations in an image. A working memory system ensures that the corresponding objects are only focused once, even if they are moving around, such that the visual search is efficient. The model has been implemented on a robotic platform in order to search for natural objects such as fruits.