In Proceedings of the 6th European Workshop on Learning Robots, pages 22–32, 1997. Paper abstract bibtex
The behaviour of reinforcement learning (RL) algorithms is best understood in completely observable, finite state- and action-space, discrete-time controlled Markov-chains. Robot-learning domains, on the other hand, are inherently infinite both in time and space, and moreover they are only partially observable. In this article we suggest a systematic method whose motivation comes from the desire to transform the task-to-be-solved into a finite-state, discrete-time, ``approximately'' Markovian task, which is completely observable too. The key idea is to break up the problem into subtasks and design controllers for each of the subtasks. Then operating conditions are attached to the controllers (together the controllers and their operating conditions which are called modules) and possible additional features are designed to facilitate observability. A new discrete time-counter is introduced at the ``module-level'' that clicks only when a change in the value of one of the features is observed. The approach was tried out on a real-life robot. Several RL algorithms were compared and it was found that a model-based approach worked best. The learnt switching strategy performed equally well as a handcrafted version. Moreover, the learnt strategy seemed to exploit certain properties of the environment which could not have been seen in advance, which predicted the promising possibility that a learnt controller might outperform a handcrafted switching strategy in the future.