Learning, planning, and representing knowledge at multiple levels of temporal abstraction are key, longstanding challenges for AI. In this paper we consider how these challenges can be addressed within the mathematical framework of reinforcement learning and Markov decision processes (MDPs). We extend the usual notion of action in this framework to include options-closed-loop policies for taking action over a period of time. Examples of options include picking up an object, going to lunch, and traveling to a distant city, as well as primitive actions such as muscle twitches and joint torques. Overall, we show that options enable temporally abstract knowledge and action to be included in the reinforcement learning framework in a natural and general way. In particular, we show that options may be used interchangeably with primitive actions in planning methods such as dynamic programming and in learning methods such as Q-learning. Formally, a set of options defined over an MDP constitutes a semi-Markov decision process (SMDP), and the theory of SMDPs provides the foundation for the theory of options. However, the most interesting issues concern the interplay between the underlying MDP and the SMDP and are thus beyond SMDP theory. We present results for three such cases: (1) we show that the results of planning with options can be used during execution to interrupt options and thereby perform even better than planned, (2) we introduce new intra-option methods that are able to learn about an option from fragments of its execution, and (3) we propose a notion of subgoal that can be used to improve the options themselves. All of these results have precursors in the existing literature; the contribution of this paper is to establish them in a simpler and more general setting with fewer changes to the existing reinforcement learning framework. In particular, we show that these results can be obtained without committing to (or ruling out) any particular approach to state abstraction, hierarchy, function approximation, or the macroutility problem. ! 1999 Published by Elsevier Science B.V. All rights reserved. * Corresponding author. 0004-3702/99/$ -see front matter ! 1999 Published by Elsevier Science B.V. All rights reserved. PII: S 0 0 0 4 -3 7 0 2 ( 9 9 ) 0 0 0 5 2 -1 182 R.S. Sutton et al. / Artificial Intelligence 112 (1999) 181-211

Planning and Learning are complementary approaches. Planning relies on deliberative reasoning about the current state and sequence of future reachable states to solve the problem. Learning, on the other hand, is focused on improving system performance based on experience or available data. Learning to improve the performance of planning based on experience in similar, previously solved problems, is ongoing research. One approach is to learn Value function (cost-togo) which can be used as heuristics for speeding up searchbased planning. Existing approaches in this direction use the results of the previous search for learning the heuristics. In this work, we present a search-inspired approach of systematic model exploration for the learning of the value function which does not stop when a plan is available but rather prolongs search such that not only resulting optimal path is used but also extended region around the optimal path. This, in turn, improves both the efficiency and robustness of successive planning. Additionally, the effect of losing admissibility by using ML heuristic is managed by bounding ML with other admissible heuristics.

Recent developments in the area of reinforcement learning have y i e l d e d a n umber of new algorithms for the prediction and control of Markovian environments. These algorithms, including the TD( ) algorithm of Sutton (1988) and the Q-learning algorithm of Watkins (1989), can be motivated heuristically as approximations to dynamic programming (DP). In this paper we p r o vide a rigorous proof of convergence of these DP-based learning algorithms by relating them to the powerful techniques of stochastic approximation theory via a new convergence theorem. The theorem establishes a general class of convergent algorithms to which both TD( ) and Q-learning belong.

Abstract-There is great interest in building intrinsic motivation into artificial systems using the reinforcement learning framework. Yet, what intrinsic motivation may mean computationally, and how it may differ from extrinsic motivation, remains a murky and controversial subject. In this article, we adopt an evolutionary perspective and define a new optimal reward framework that captures the pressure to design good primary reward functions that lead to evolutionary success across environments. The results of two computational experiments show that optimal primary reward signals may yield both emergent intrinsic and extrinsic motivation. The evolutionary perspective and the associated optimal reward framework thus lead to the conclusion that there are no hard and fast features distinguishing intrinsic and extrinsic reward computationally. Rather, the directness of the relationship between rewarding behavior and evolutionary success varies along a continuum.

Designing the dialogue policy of a spoken dialogue system involves many nontrivial choices. This paper presents a reinforcement learning approach for automatically optimizing a dialogue policy, which addresses the technical challenges in applying reinforcement learning to a working dialogue system with human users. We report on the design, construction and empirical evaluation of NJFun, an experimental spoken dialogue system that provides users with access to information about fun things to do in New Jersey. Our results show that by optimizing its performance via reinforcement learning, NJFun measurably improves system performance.