Learning Physical Collaborative Robot Behaviors From Human Demonstrations

Abstract

Abstract-Robots are becoming safe and smart enough to work alongside people not only on manufacturing production lines, but also in spaces such as houses, museums or hospitals. This can be significantly exploited in situations where a human needs the help of another person to perform a task, because a robot may take the role of the helper. In this sense, a human and the robotic assistant may cooperatively carry out a variety of tasks, therefore requiring the robot to communicate with the person, understand his/her needs and behave accordingly. To achieve this, we propose a framework for a user to teach a robot collaborative skills from demonstrations. We mainly focus on tasks involving physical contact with the user, where not only position, but also force sensing and compliance become highly relevant. Specifically, we present an approach that combines probabilistic learning, dynamical systems and stiffness estimation, to encode the robot behavior along the task. Our method allows a robot to learn not only trajectory following skills, but also impedance behaviors. To show the functionality and flexibility of our approach, two different testbeds are used: a transportation task and a collaborative table assembly.

Flexible Automation Driven by Demonstration: Leveraging Strategies that Simplify Robotics

“…Certainly, an interesting extension of the motion controller would be the inclusion of automatic generation of an impedance control scheme starting from the automatically generated robot description from [7], [8], which could potentially be combined with the demonstration-driven encoding of the impedance target parameters as in [17], [18]. In addition, approaches to optimize the redundancy resolution, instead of simply adding damping for the null-space motions, could be also considered.…”

A learning-based shared control architecture for interactive task execution

“…The learned distribution of demonstrations can then be used for designing control schemes. For instance, in [12] the robot learns collaborative skills and adapts its impedance behavior from demonstrations. However, to the best of our knowledge, no prior work has been proposed for leveraging the distribution of the demonstrated trajectories for controlling the balance between the controllers autonomy and the human inputs.…”

“…In LfD methods, the distribution of the demonstrated trajectories is often modeled, and the learned distribution is leveraged for generalizing the demonstrated behaviors [10], [11]. Recent work on LfD showed that the variance of the demonstrated trajectories can be used to adaptively control the robot behaviors [12]. Leveraging the distribution of the demonstrated trajectories into the design of shared control frameworks seems, therefore, a meaningful/promising approach for obtaining a shared control framework able to adjust online the balance between the robot autonomy and the human preference.…”

Co-manipulation with a library of virtual guiding fixtures

“…In this respect, our work can be compared to (Amor et al, 2014;Medina et al, 2012;Rozo et al, 2016;Wrede et al, 2013) where the user and the robot have to execute a learned task together. Regarding the definition of the virtual guides through PbD, our work can be compared to the work done by (Vakanski et al, 2012;Mollard et al, 2015;Boy et al, 2007;Ewerton et al, 2016;Lee and Ott, 2011;Sanchez Restrepo et al, 2017) where the concept of Task refinement is exploited, as we will see in Section 6 this is possible due to the incremental training of GMM .…”