direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Logo der TU Berlin

Inhalt des Dokuments

Collective Learning Phenomena in Swarms of Robotic Agents

From Fachgebiet Regelungssysteme TU Berlin

Jump to: navigation, search


Fig. 1: CAD rendering of multiple TWIPRs

The goal of this project is to study collective learning phenomena in a swarm of two-wheeled inverted pendulum robots (TWIPR) that use a combination of learning control methods and distributed control methods to accomplish a complex task. These research activities are carried out within the Cluster of Excellence "Science of Intelligence".

Each robot consists of a chassis that houses the central computing unit, motor drivers, batteries and inertial measurement units as well as two DC motors that drive the two wheels. We use feedback control algorithms to stabilize the inverted pendulum that is formed by the chassis and wheels in its upright position and allow the robot to move and navigate in the horizontal plane on different paths and terrains.

On top of the chassis there is a magnetic mechanism that enables the robot to assimilate objects. However, assimilation requires the robot to perform a rapid and complex maneuver that cannot be accomplished by the aforementioned feedback control algorithms. Instead, the robot must perform the task repeatedly and gradually improve its performance by learning from previous trials. To this end, we implement algorithms that combine methods from feedback control theory and machine learning.

All robots that form the swarm are trying to assimilate objects simultaneously. During the learning process, the robots can exchange information and help each other to accomplish the task faster than by purely individual learning. The resulting collective learning dynamics are investigated both theoretically and experimentally. In particular, we study (i) the interplay between prior knowledge, individual learning and knowledge transfer between agents, (ii) the relevance of the communication topology, and (iii) the effect of agent heterogeneity with respect to prior knowledge and learning strategies, dynamics and distortion of shared knowledge.

Fig. 2: Progression of the error norm of three different ILC systems and their cooperative learning progress.

People involved

Acknowledgments

http://www.control.tu-berlin.de/wiki/images/9/96/Logo_maxon_motor.jpg

The drive solution used in this project is sponsored by Maxon Motor AG.

Related Publications

Music, Zenit, Molinari, Fabio, Gallenmüller, Sebastian, Ayan, Onur, Zoppi, Samuele, Kellerer, Wolfgang, Carle, Georg, Seel, Thomas, Raisch, Jörg. Design of a Networked Controller for a Two-Wheeled Inverted Pendulum Robot. 11 2018.

T. Seel, T. Schauer, J. Raisch. Monotonic Convergence of Iterative Learning Control with Variable Pass Length. International Journal of Control, 90 (3):409–422, 2016.
M. Guth, T. Seel, J. Raisch. Iterative Learning Control with Variable Pass Length Applied to Trajectory Tracking on a Crane with Output Constraints. In Proceedings of the 52nd IEEE Conference on Decision and Control, pages 6676–6681, Firenze, Italy, December 2013.
T. Seel, T. Schauer, J. Raisch. Iterative Learning Control for Variable Pass Length Systems. In Proceedings of the 18th IFAC World Congress, pages 4880–85, Milan, Italy, 2011.
J. Beuchert, J. Raisch, T. Seel. Design of an iterative learning control with a selective learning strategy for swinging up a pendulum (accepted). In European Control Conference (ECC), 2018.

Recommend this page