Search

We have developed a new reinforcement learning (RL) technique called Bayesian-discrimination-function-based reinforcement learning (BRL). BRL is unique, in that it does not have state and action spaces designed by a human designer, but adaptively segments them through the learning process. Compared to other standard RL algorithms, BRL has been proven to be more effective in handling problems encountered by multi-robot systems (MRS), which operate in a learning environment that is naturally dynamic. Furthermore, we have developed an extended form of BRL in order to improve the learning efficiency. Instead of generating a random action when a robot functioning within the framework of the standard BRL encounters an unknown situation, the extended BRL generates an action determined by linear interpolation among the rules that have high similarity to the current sensory input. In this study, we investigate the robustness of the extended BRL through further experiments. In both physical experiments and computer simulations, the extended BRL shows higher robustness and relearning ability against an environmental change as compared to the standard BRL. [ABSTRACT FROM AUTHOR]

A novel approach to topology and weight evolving artificial neural networks (TWEANNs) is presented. Compared with previous TWEANNs, this method has two major characteristics. First, a set of genetic operations may be designed without recombination because it often generates an offspring whose fitness value is considerably worse than its parents. Instead, two topological mutations whose effect on fitness value is assumed to be nearly neutral are provided in the genetic operations set. Second, a new encoding technique is introduced to define a string as a set of substrings called operons. To examine our approach, computer simulations were conducted using the standard reinforcement learning problem known as the double pole balancing without velocity information. The results obtained were compared with NEAT results, which is recognised as one of the most powerful techniques in TWEANNs. It was found that our proposed approach yields competitive results, especially when the problem is difficult. [ABSTRACT FROM AUTHOR]

We have developed a new reinforcement learning technique called Bayesian-discrimination-function-based reinforcement learning (BRL). BRL is unique, in that it not only learns in the predefined state and action spaces, but also simultaneously changes their segmentation. BRL has proven to be more effective than other standard RL algorithms in dealing with multi-robot system (MRS) problems, where the learning environment is naturally dynamic. This paper introduces an extended form of BRL that improves its learning efficiency. Instead of generating a random action when a robot encounters an unknown situation, the extended BRL generates an action calculated by a linear interpolation among the rules with high similarity to the current sensory input. In both physical experiments and computer simulations, the extended BRL showed higher search efficiency than the standard BRL. [ABSTRACT FROM AUTHOR]

We have been developing a new reinforcement learning called BRL, which is especially effective to multi-robot systems (MRS). BRL has a unique feature that it not only learns in the learning space but also changes the segmentation of the learning space simultaneously. BRL has been proved to be clearly effective than the other standard RL algorithms to MRS problems where the learning environment is naturally dynamic. However, we have also noticed that MRS needs the more robustness for the learning mechanism as the complexity level increases. In this paper, BRL is extended to improve the robustness against the dynamics in a learning environment by showing a way of overcoming the unwanted feature of over-fitting. Computer simulations are conducted to illustrate the robust performance of the proposed technique. [ABSTRACT FROM AUTHOR]