Аңдатпа
Robot programming by demonstration (PbD) in the unstructured environment is usually a challenging task, which requires to take into account different parameters. One of the main difficulties in an unstructured environment is that the location and orientation of the objects may change dynamically, requiring the learning algorithm to posses acceptable generalization and extrapolation capabilities. There are several category of PbD approaches proposed to tackle such issues, some of which look at the objects in the environment as external parameters (task parameters, or TPs) and assume that the movement or trajectory is modulated by such objects. While, some of those TPs might not be completely observable all the time, introducing additional difficulties on the task learning. On the other hand, in specific situations two or more objects may contain similar information for the task execution. In this paper, an approach based on task-parameterized Gaussian mixture model (TP-GMM) for PbD is proposed that exploits the redundancy in the environment to deal with the partial observability of the task parameters and provide a fault tolerant approach in the sense of availability of the task parameters. The proposed approach is tested using some simulation experiments.