Reinforcement Learning using Sigmoid Networks and Griddy-Gibbs Sampler for Mobile Robots