Reinforcement learning (RL) enables an agent to learn behavior by acquiring experience through trial-and-error interactions with a dynamic environment. However, knowledge is usually built from scratch and learning to behave may take a long time. Here, we improve the learning performance by leveraging prior knowledge; that is, the learner shows proper behavior from the beginning of a target task, using the knowledge from a set of known, previously solved, source tasks. In this paper, we argue that building stochastic abstract policies that generalize over past experiences is an effective way to provide such improvement and this generalization outperforms the current practice of using a library of policies. We achieve that contributing with a new algorithm, AbsProb-PI-multiple and a framework for transferring knowledge represented as a stochastic abstract policy in new RL tasks. Stochastic abstract policies offer an effective way to encode knowledge because the abstraction they provide not only generalizes solutions but also facilitates extracting the similarities among tasks. We perform experiments in a robotic navigation environment and analyze the agent's behavior throughout the learning process and also assess the transfer ratio for different amounts of source tasks. We compare our method with the transfer of a library of policies, and experiments show that the use of a generalized policy produces better results by more effectively guiding the agent when learning a target task.

In this paper we improve learning performance of a risk-aware robot facing navigation tasks by employing transfer learning; that is, we use information from a previously solved task to accelerate learning in a new task. To do so, we transfer risk-aware memoryless stochastic abstract policies into a new task. We show how to incorporate risk-awareness into robotic navigation tasks, in particular when tasks are modeled as stochastic shortest path problems. We then show how to use a modified policy iteration algorithm, called AbsProb-PI, to obtain risk-neutral and risk-prone memoryless stochastic abstract policies. Finally, we propose a method that combines abstract policies, and show how to use the combined policy in a new navigation task. Experiments validate our proposals and show that one can find effective abstract policies that can improve robot behavior in navigation problems.

Relational transfer across reinforcement learning tasks via abstract policies / M.L. Koga.versão corr.-São Paulo, 2014. 119 p. Dissertação (Mestrado)