Learning for contact-rich tasks with cobots


Loading...

Author / Producer

Date

2024-09-20

Publication Type

Master Thesis

ETH Bibliography

yes

Citations

Altmetric

Data

Abstract

A common approach to perform contact-rich tasks with robots is using force impedance control. The traditional method relies on commanding torque to the robot actuators. However, in modern cobots such interface is restricted due to safety reasons. To address this limitation, we implement an outer control loop that uses information from an external force sensor mounted in the robot’s end-effector to command joint velocities in a way that effectively emulates an impedance behavior. Motivated by the challenging tuning of this controller, we explore a data-driven tuning method based on Bayesian Optimization. Once robots operate in impedance mode, the next step is designing trajectories that achieve the desired tasks. We propose a strategy that uses Proximal Policy Optimization (PPO) to learn a motion pattern to perform a complex assembly without relying on a strong trajectory parametrization. Overall, this work demonstrates a pipeline towards more autonomous procedures to enable cobots for contact-risk manipulations.

Publication status

published

External links

Editor

Contributors

Examiner: Lygeros, John
Examiner : Balta, Efe
Examiner : Eberhard, Enrico
Examiner : Reber, Dominic
Examiner : Papaspyros, Vaios

Book title

Journal / series

Volume

Pages / Article No.

Publisher

ETH Zurich

Event

Edition / version

Methods

Software

Geographic location

Date collected

Date created

Subject

Organisational unit

03751 - Lygeros, John / Lygeros, John check_circle

Notes

Funding

Related publications and datasets