Learning Through Imitation for Mobile Robots

Imitative learning facilitates skill acquisition in a social environment where one agent learns how to act by perceiving and, in some sense, imitating the actions of another. Most of the research work is concentrated in action–level imitation learning, where the low level motor actions are learned. We are interested in program–level imitation learn-ing, where the learner learns through imitation what behaviour should execute under a perceptual state. Furthermore, most of the research work in imitation learning in-volves behaviours that do not involve object interaction. There are a few attempts with behaviours that involve object interaction, but these are studied in computer simulated environments. We are also interested in behaviours that involve object interaction, but we wish to study them in real robots. A foraging scenario was chosen as the experi-mental platform. By robot foraging, we mean that the robot searches for food objects and brings them to its home. There are two ways that the robot could bring the food; it can either pick it up and carry it, or it can push it. In addition, an avoidance behaviour is executed when the food is poisoned. There is a teacher robot, who is an expert in for-aging, and intends to teach a learner robot. The learner robot follows the teacher robot and imitates its demonstrated behaviours. The learner then associates the behaviours, it just imitated, with its perception. With this way, the learner will be able to perform the behaviours in a similar situation, without needing the teacher to show it what to do. The experimental results we obtained, show that the learner improves as more training is acquired. However, these results also show that the rate of the learner’s improvement drops beyond a number of training episodes. By observations of the experiments, we also came to the conclusion that some design issues in the recall phase of the learner as well as the variety of the training episodes affect the performance of the learner.