Incremental Learning of Goal-Directed Actions in a Dynamic Environment by a Robot Using Active Inference

This study investigated how a physical robot can adapt goal-directed actions in dynamicallychanging environments, in real-time, using an active inference-based approach with incrementallearning from human tutoring examples. Using our active inference-based model, while goodgeneralization can be achieved with appropriate parameters, when faced with sudden, large changesin the environment, a human may have to intervene to correct actions of the robot in order to reachthe goal, as a caregiver might guide the hands of a child performing an unfamiliar task. In orderfor the robot to learn from the human tutor, we propose a new scheme to accomplish incrementallearning from these proprioceptive-exteroceptive experiences combined with mental rehearsal ofpast experiences. Our experimental results demonstrate that using only a few tutoring examples,the robot using our model was able to significantly improve its performance on new tasks withoutcatastrophic forgetting of previously learned tasks.