This paper presents a neural conditioning model for on-line learning of behaviors on mobile robots. The model is based on Grossberg's neural model of conditioning as recently implemented by Chang and Gaudiano. It attempts to tackle some of the limitations of the original model by (1) using a temporal difference of the reinforcement to drive learning, (2) adding eligibility trace mechanisms to dissociate behavior generation from learning, (3) automatically categorizing sensor readings and (4) bootstrapping the learning process through the use of unconditioned responses. Preliminary results of the model that learn simple behaviors on a mobile robot simulator are presented.