Finding and transferring policies using stored behaviors

Finding and transferring policies using stored behaviors

We present several algorithms that aim to advance the state-of-the-art in reinforcement learning and planning algorithms. One key idea is to transfer knowledge across problems by representing it using local features. This idea is used to speed up a dynamic programming based generalized policy iteration. We then present a control approach that uses a library of trajectories to establish a contro...

Finding and transferring policies

SOS++: Finding Smart Behaviors Using Learning and Evolution

We present SOS++, a bioinspired method combining evolution and learning, allowing the automatic design of the controller of autonomous agents, described as a finite-state machine. The application of this method to well-known problems, for example the follow-up of a trail or the resolution of a maze, led to the emergence of some behaviors we could qualify as intelligent. Moreover, it is possible...

Finding Best k Policies

An optimal probabilistic-planning algorithm solves a problem, usually modeled by a Markov decision process, by finding its optimal policy. In this paper, we study the k best policies problem. The problem is to find the k best policies. The k best policies, k > 1, cannot be found directly using dynamic programming. Näıvely, finding the k-th best policy can be Turing reduced to the optimal planni...

On Finding Optimal Policies for Markovian Decision Processes Using Simulation

A simulation method is developed, to find an optimal policy for the expected average reward of a Markovian Decision Process. It is shown that the method is consistent, in the sense that it produces solutions arbitrarily close to the optimal. Various types of estimation errors are examined, and bounds are developed.