A Collaborative Mobile Robot Arm that can Learn Impedance Critical Tasks from Humans

Informazioni relative al progetto

CoMRAde

ID dell’accordo di sovvenzione: 753219

DOI

10.3030/753219

Progetto chiuso

Data della firma CE 6 Marzo 2017

Data di avvio 1 Aprile 2017

Data di completamento 31 Marzo 2019

Finanziato da

• EXCELLENT SCIENCE - Marie Skłodowska-Curie Actions

Costo totale

€ 157 845,60

Contributo UE

€ 157 845,60

Coordinato da

OZYEGIN UNIVERSITESI
Türkiye

Periodic Reporting for period 1 - CoMRAde (A Collaborative Mobile Robot Arm that can Learn Impedance Critical Tasks from Humans)

Periodo di rendicontazione: 2017-04-01 al 2019-03-31

Sintesi del contesto e degli obiettivi generali del progetto

This project CoMRAde aimed to initiate a change in the current view of robotic automation, which is dominated by industrial robot arms that are incapable of working side-by-side with humans. To change this view, the project mainly focused on the impedance-critical tasks that cannot be easily accomplished by the conventional robot arms. Our industrial partner Arcelik A.S. Turkey’s largest white goods company, donated a mobile manipulator system to us and specifically asked us to focus on the polishing task. The main problem in a polishing task is to adjust the applied force from the polisher. Unlike classical position control-based tasks, e.g. pick and place, this task enforces the finely tuned adjustments of both force and position constraints. Furthermore, there is no explicit definition of force constraints, i.e. there is no systematic way to generate force trajectories. Therefore, automating this task would be a game-changer implementation in increasing the efficiency of the manufacturing lines.

The importance for society lies on the fact that the project includes active involvement of human interaction. Rather than taking up humans’ job, it keeps them in the loop and elevates their position within the factory hierarchy. In a possible scenario, a human may train multiple robots on different workpieces and once trained, he can manage the multiple robot-aided manufacturing tasks. This is in line with the Industry 4.0 criteria, therefore, it definitely helps to keep manufacturing jobs in the EU with no particular need to increase manpower. With the accumulation of similar projects, the EU may be the focal point in smart factories that effectively play a crucial role in mass customization driven manufacturing market. This also should have positive economic implications on society as well.

The main research goal in CoMRAde project is to devise necessary technologies to accelerate the robot-centered automation processes of industrial manufacturing in Europe, in accordance with the Industry 4.0 criteria. The objectives are as follows:

* A force sensor-based haptic interface with proven stability is built to allow the physical interaction between human workers and the robot. This haptic interface should exert approximately zero impedance when a human is interacting.

* A human-to-robot skill transfer framework is constructed to acquire task-specific force and position constraints to achieve the trained tasks. Per our industry partner Arcelik' s request, the metal sheet polishing task is determined as the primary focus.

* Upon successful laboratory experiments, the whole system is integrated into the real manufacturing task in Arcelik’ s factories. The robot is expected to work adaptively in the human working environment with no robot-specific arrangements and is reconfigurable for distinct settings.

Lavoro eseguito dall’inizio del progetto fino alla fine del periodo coperto dalla relazione e principali risultati finora ottenuti

While the detailed technical results are provided in the Final Report, the results obtained during the project could be summarized as follow:

Test Setup and System Integration: Our industry partner Arcelik A.S. donated a mobile manipulator that includes a MiR mobile robot base and a Universal Robot UR5 robot arm. Per Arcelik’s request, the metal sheet polishing was the primary focus. To that end, an apparatus to attach a commercially available polisher to the robot end-effector was built and tested for mechanical loading. Furthermore, a 6-axis Force/Torque sensor was mounted on the robot end-effector as well to measure interaction forces. In order to ensure flexible and real-time operable control, a ROS-based operating system was installed and the fastest possible real-time operation for UR5 has been activated: 8 ms.

Haptic Interface Development: In order to allow safe and dependable human-robot physical interaction, an admittance-based haptic interface was developed and integrated into the system. The interface enables the robot to exert near-zero impedance when a human user is training the robot for learning by demonstration purposes. Once the robot is trained, the same interface is activated to satisfy both force and position constraints in a way so as to address the autonomous polishing task.

A Human-to-Robot Skill Transfer Scheme: With the help of the haptic interface, it is possible to collect data concerning force and position constraints which are equally vital for the success of polishing tasks. We proposed a method called force-induced motion generation in which the task forces are parametrized in terms of joint states, i.e. joint angles and angular velocities via a deep neural network approach.

Integration to Real Factory Scenario: Currently, we are in the process of integrating the system to a real factory setting in which real human users will train the robot for the actual meta sheet polishing task. Therefore, the proposed method will be exploited in a real-life scenario.

Regarding the results and outcomes explained above, 1 workshop presentation was given. Two conference proceedings were published and one journal paper is in preparation.

Progressi oltre lo stato dell’arte e potenziale impatto previsto (incluso l’impatto socioeconomico e le implicazioni sociali più ampie del progetto fino ad ora)

When looking at the related literature, the state-of-the-art systems extensively utilize machine learning algorithms. While useful in its own functionality domain, such algorithms rely on long and exhausting trial-and-correction periods, in other words, episodes. Although a skill transfer can be acquired using machine learning, it may not be completely realistic to implement it to a real-life factory setting. In such a setting, the robots must be easily reconfigurable. Furthermore, human workers may not be willing to train the robot repeatedly for multiple trials.

The progress beyond the state-of-the-art lies on the implementation of our proposed method, namely, force-induced motion. This method requires human workers to teach the task only once or twice. Afterward, a deep neural network is trained using the collected data within a couple of minutes and then the robot can execute the motion based on the force constraints and robot states that are obtained from the trained neural network. This scheme is novel in the sense that it provides a feasible solution for a realistic scenario since it does not require exhaustively long training procedures. Once the robot is trained, it can also fine-tune its motion via machine learning. Yet, note that it is able to execute the task reliably from the initial execution.

The socio-economic impact of the project is vital in the sense that it may pave the way for the new generation collaborative robots which can be easily trained by human workers with no engineering background. There will be no need to hire specialized robotics engineers or such; the robots will take their place side-by-side with human workers. In contrast to popular belief that suggests robots taking up humans’ jobs, our approach keeps the human workers in the loop and gives them a crucial role. It will provide human workers with the ability to “program” robots for monotonous tasks. In other word, the robots will take the bottom level in the workflow and allow human workers to focus on more complicated tasks, such as higher-level planning of the tasks. This will improve the work efficiency and the factories in the EU can take larger steps for the transformation towards mass customization, positively impacting the socio-economic status of the EU.

Experimental setup: UR5 Robot arm with a MiR mobile base

A polisher is attached to the robot end-effector