Achieving Cooperative Behavior Based on Intention Estimation by Learning Combinations of Modules

Abstract

A robot needs to process information appropriately depending on the environment or context. However, some of the abilities required by a robot are often common irrespective of the environment or context. In such situations, the learning agent should not learn the abilities again but use the learning results of previous tasks. In the field of the study of intellectual systems, models have been proposed that solve complex problems by combining modules, each of which serve a specific function such as recognition, planning, or action selection. The models can use the learning results of previous tasks in different environments or contexts by combining modules it has learnt. In this paper, we focus on achieving cooperative behavior based on intention estimation, and propose a model for a learning agent that can acquire combinations of modules using which the agent can achieve cooperative behavior based on intention estimation. The experimental results indicate that a desirable combination of the modules was acquired and the learning process suitably progressed.