Learning from demonstration with model-based Gaussian process.

摘要: In learning from demonstrations, it is often desirable to adapt the behavior of robot as a function variability retrieved human demonstrations and (un)certainty encoded in different parts task. this paper, we propose novel multi-output Gaussian process (MOGP) based on mixture regression (GMR). The proposed approach encapsulates covariance MOGP. Leveraging generative nature GP models, our can efficiently modulate trajectories towards new start-, via- or end-points defined by Our framework allows precisely track via-points while being compliant regions high variability. We illustrate simulated examples validate real-robot experiment.