### Abstract

This paper presents a method by which a robot can learn through observation to perform a collaborative manipulation task, namely lifting an object. The task is first demonstrated by a user controlling the robot's hand via a haptic interface. Learning extracts statistical redundancies in the examples provided during training by using Gaussian Mixture Regression and Hidden Markov Model. Haptic communication reflects more than pure dynamic information on the task, and includes communication patterns, which result from the two users constantly adapting their hand motion to coordinate in time and space their respective motions. We show that the proposed statistical model can efficiently encapsulate typical communication patterns across different dyads of users, that are stereotypical of collaborative behaviours between humans and robots. The proposed learning approach is generative and can be used to drive the robot's retrieval of the task by ensuring a faithful reproduction of the overall dynamics of the task, namely by reproducing the force patterns for both lift the object and adapt to the human user's hand motion. This work shows the potential that teleoperation holds for transmitting both dynamic and communicative information on the task, which classical methods for programming by demonstration have traditionally overlooked.

### Bibtex reference

@inproceedings{Calinon09ICAR,
author = "S. Calinon and P. Evrard and E. Gribovskaya and A. Billard and A. Kheddar",
title = "Learning collaborative manipulation tasks by demonstration using a haptic interface",
booktitle = "Proc. Intl Conf. on Advanced Robotics ({ICAR})",
year = "2009",
month="June",
location="Munich, Germany",
pages="1--6"
}

### Video

Learning of a collaborative manipulation skill with HRP-2 by showing multiple demonstrations of the skill in slightly different situations (different initial position and orientation of the object). The robot is standing in an half-sitting posture during the experiment, where the 7 DOFs of the right arm and torso are used in the experiment. A built-in stereoscopic vision is used to track colored patches, and a 6-axes force sensor at the level of the right wrist is used to track the interaction forces with the environment during task execution. The robot is teleoperated through a Phantom Desktop haptic device from Sensable Technologies. An impedance controller is used to control the robot.
This work is in collaboration with the Joint Japanese-French Robotics Laboratory (JRL) at AIST, Tsukuba, Japan.