Learning from Demonstration in Static Environment and Generalizing to Dynamic Environments

Robot learning from demonstration has been successfully used, in industrial environments, to increase the speed and reduce the complexity of the programming phase. A major challenge, however, consists in enabling a robot to generalize task demonstrations to a complex dynamic environment. To tackle this problem, an approach has been proposed in [1], combining GMM/GMR and DMPs with optimal control. This approach allows a robot to generalize noisy task demonstrations to different goal points and environments with static obstacles. Considering that in practical applications, a robot must perform a task in dynamic environments, here we extend the approach in such a way that a robot can learn a task model from demonstrations in an environment with stationary obstacles and reproduce the task in an environment with moving obstacles.