Learning adaptive target reaching for robots with recurrent neural networks Theory of Neural Systems

Description

BACKGROUND

We would like to address the problem of fast adaptation for motor-control through end-to-end learning, rather than hand-designed mechanisms that perform adaptation for specific use-cases [1]. Furthermore, we are interested in doing fast adaptation for torque-controlled robots for two reasons: (a) They provide a promising technology for performing more complex and delicate tasks that emulate human dexterity and (b) Human motor control is torque driven and we wish to understand the mechanisms of fast adaptation in humans.

YOUR TASK

In this thesis, you will develop multiple RNN architectures to learn target reaching. The arm used for reaching will be simulated in the PyBullet framework and will be torque controlled. You will also test the network’s ability to adapt to different arm properties (e.g. by varying masses and lengths of the arm’s links). The robotics aspects of the project will be supervised in collaboration with TU Munich.

Required Skills
* Good Knowledge of Python and Linux
* Basic Knowledge in Robotics
* Experience with Deep Learning frameworks and is preferred

References
[1] Cheah, C. C., et al. “Adaptive Tracking Control for Robots with Unknown Kinematic and Dynamic Properties.” The International Journal of Robotics Research, vol. 25, no. 3,
Mar. 2006, pp. 283–296

[2] Hitzler, Kevin, et al. ‘Learning and Adaptation of Inverse Dynamics Models: A Comparison’. 2019 IEEE-RAS 19th International Conference on Humanoid Robots (Humanoids),
2019, pp. 491–98.

This thesis will be co-supervised by Mahmoud Akl (mahmoud.akl@tum.de) from Technische Universität München

The Institut für Neuroinformatik (INI) is a central research unit of the Ruhr-Universität Bochum. We aim to understand the fundamental principles through which organisms generate behavior and cognition while linked to their environments through sensory systems and while acting in those environments through effector systems. Inspired by our insights into such natural cognitive systems, we seek new solutions to problems of information processing in artificial cognitive systems. We draw from a variety of disciplines that include experimental approaches from psychology and neurophysiology as well as theoretical approaches from physics, mathematics, electrical engineering and applied computer science, in particular machine learning, artificial intelligence, and computer vision.

Universitätsstr. 150, Building NB, Room 3/32
D-44801 Bochum, Germany

Tel: (+49) 234 32-28967
Fax: (+49) 234 32-14210