Learning compliant assembly motions from demonstration

Abstract

We present a novel method for learning 6-D compliant motions from demonstration. The key advantage of our learning method compared to current Learning from Demonstration (LfD) methods is that we learn to take advantage of existing mechanical gradients (such as chamfers) with compliance to perform in-contact motions consisting of both translation and rotation, as in aligning workpieces or attaching hose couplers. We find the desired direction, which leads the robot's end-effector to the location shown in the demonstration either in free space or in contact, separately for translational and rotational motions. The key idea is to first compute a set of directions which would result in the observed motion at each timestep during a demonstration. By taking an intersection over all such sets from a demonstration we find a single desired direction which can reproduce the demonstrated motion. Finding the number of compliant axes and their directions in both rotation and translation is based on the assumption that in the presence of a desired direction of motion, all other observed motion is caused by the contact force of the environment, signalling the need for compliance. We evaluate the method on a KUKA LWR4+ robot with test setups imitating typical tasks where a human would use compliance to cope with positional uncertainty. Results show that the method can successfully learn and reproduce compliant motions by taking advantage of the geometry of the task, therefore reducing the need for initial accuracy. * This work was supported by Academy of Finland, decision 286580. † M. Suomalainen and V. Kyrki are with

Improving dual-arm assembly by master-slave compliance

“…As discussed in Section III-A.1, with a correctly chosen CoC, rotational compliance only is sufficient for completing the alignment. However, the results in [8] show that adding translational compliance increases the convergence region for a single-arm task. As our goal is to maximize the convergence region and consider methods that accomplish the alignment regardless of the choice of CoC, the option of single manipulator rotation-only compliant will not be considered further.…”

“…In Section III-A we observe how different compliance parameters can be set to enlarge the convergence region and decrease the required joint motions. In Section III-B we shortly present the method from [8] for learning compliant behaviour from human demonstration and how it is applied to a dual-arm scenario. Finally, we also show how to control the manipulators during reproduction of the motion.…”

“…Whereas this method can learn more complex trajectories in free space, the tight temporal coupling between pose and torque trajectories makes it susceptible to pose errors. As the scope of this paper is in-contact alignment tasks, we prefer error tolerance over complexity of applicable motions and thus base our work on the method from [8].…”

“…We also show how important the choice of Center of Compliance (CoC) is for a successful alignment. Then we present how this kind of motions for an impedance controller can be efficiently learned from human demonstrations with an LfD method presented in [7], [8]. Finally, we explain how compliances for both arms can be learned from a teleoperated demonstration where one arm is fixed, thus simplifying the demonstration.…”

“…In this work we present the following contributions: 1) we present an analysis on the importance of choosing the CoC and compliant axes in an assembly task; 2) we show that the compliance parameters learned by the method in [8] can be used in both arms performing a dual-arm assembly task, even from demonstration with one arm only; and 3) we show through experiments that having compliance on both manipulators eases an assembly task by decreasing the time and joint motions required to complete the task.…”