In this paper, we propose a method for controlling an active camera in tracking a moving target highly accurately using PaLM-tree, an easy-to-use function approximator we developed. Our feedback error learning controller enhanced by PaLM-tree improves accuracy in tracking a moving target. PaLM-tree correctly learns the inverse model in the feedback error learning. Despite unknown mechanical friction and relatively large closed-loop control latency, our active camera head can successfully track eye movement using the feedback error learning controller. Through experimental results, we confirmed that our method achieved better performance control than a tuned feedback controller.
展开▼
