Speeding-Up Action Learning in a Social Robot With Dyna-Q+: A Bioinspired Probabilistic Model Approach

Robotic systems that are developed for social and dynamic environments require adaptive mechanisms to successfully operate. Consequently, learning from rewards has provided meaningful results in applications involving human-robot interaction. In those cases where the robot's state space number of actions is extensive, dimensionality becomes intractable this drastically slows down process. This effect specially notorious one-step temporal difference methods because just one update performed per robot-environment paper, we prove how action-based a robot can be improved by combining classical reinforcement methods, such as Q-learning or Q( λ), with probabilistic model environment. architecture, which have called Dyna, allows simultaneously act plan using experience obtained during real interactions. Principally, Dyna improves algorithms terms convergence speed stability, strengthens Hence, work embedded architecture our robot, Mini, endow it ability autonomously maintain an optimal internal while living