Intelligent control of legged robots

Project members:prof.dr. R. Babuška (Robert), dr. G.A.D. Lopes (Gabriel)
Keywords:Learning and adaptive control, Robotics and mechatronics, Discrete-event systems, Hybrid and nonlinear systems, Intelligent control
Sponsored by:Delft University of Technology

This project aims to study various classes of machine learning tools applied to real legged robots.

Nature provides the best inspiration for the field of robotics. The complex neuromechanical tools inherited through DNA endows animals with a collection of sophisticated reactive and learning capabilities. In this project we take inspiration from nature to focus on the study of specific machine learning tools applied to concrete robotics applications. Reinforcement learning, thanks to its attractive bio-inspired mathematical framework, is our main tool of research.

The multi-disciplinary nature of this project encompasses the following topics:

- Low level control of nonlinear hybrid dynamical systems via traditional control methodology combined with machine learning tools. This includes the synthesis of motion gaits and energetic optimizations of motion.

- Supervisory control for navigation augmented with identification/fault detection and fault tolerant control.

- Hardware/software development of a modular high dynamical robotic platform.

- The development of a vision sensor for odometry and integration of exteroceptive sensors with inertial measurements for state estimation and fault detection.

© Copyright Delft Center for Systems and Control, Delft University of Technology, 2017.