Learning multiple collaborative tasks with a mixture of Interaction Primitives

Abstract

Abstract-Robots that interact with humans must learn to not only adapt to different human partners but also to new interactions. Such a form of learning can be achieved by demonstrations and imitation. A recently introduced method to learn interactions from demonstrations is the framework of Interaction Primitives. While this framework is limited to represent and generalize a single interaction pattern, in practice, interactions between a human and a robot can consist of many different patterns. To overcome this limitation this paper proposes a Mixture of Interaction Primitives to learn multiple interaction patterns from unlabeled demonstrations. Specifically the proposed method uses Gaussian Mixture Models of Interaction Primitives to model nonlinear correlations between the movements of the different agents. We validate our algorithm with two experiments involving interactive tasks between a human and a lightweight robotic arm. In the first, we compare our proposed method with conventional Interaction Primitives in a toy problem scenario where the robot and the human are not linearly correlated. In the second, we present a proof-of-concept experiment where the robot assists a human in assembling a box.

A Probabilistic Framework for Semi-autonomous Robots Based on Interaction Primitives with Phase Estimation

“…1(a), we applied our method on a multi-task scenario where the robot plays the role of a coworker that helps a human assembling a toolbox. This scenario was previously proposed in [8] where time-alignment was used on the training data. While in our previous work, conditioning could only be computed at the end of the movement, here, the robot can predict the collaborative trajectory before the human finishes moving, leading to a faster robot response.…”

“…Previous works [14,8] have only addressed spatial variability, but not temporal variability of movements. However, when demonstrating the same task multiple times, a human demonstrator will inevitably execute movements at different speeds, thus changing the phase at which events occur.…”

“…As a consequence, during execution, the conditioning (6) can only be used when the phase of the human demonstrator coincides with the phase encoded by the time-aligned model, which is unrealistic in practice. In [14,8] we avoided this problem by conditioning only on the last position of the human movement, since for this particular case, the corresponding basis function is known to be ψ T . For any other time step t, the association between y * t and the basis ψ t is unknown given that the human presents temporal variability and that the velocity is either unobserved or computation from derivatives impractical due to sparsity of position measurements 2 .…”

“…First, the training data must be time-aligned, for example by DTW; second, only one type of interaction pattern-or collaborative task-can be encoded within a single Gaussian (mixture of models were used to address the latter problem in an unsupervised fashion [8]). …”

“…It leverages on the representation of movements with ProMPs, our developments into the context of human-robot interaction [1,14], and the ability to address multiple tasks [14,8]. While our previous interaction models were explicitly time-dependent, here, we introduce a phase-dependent method.…”