In order to strengthen our dedicated team in the Robotics Innovation Center (RIC) research department in Bremen we are looking for a
Master Thesis/Internship
(full-time/part-time, 3 – 9 months)
“Hybrid State Estimation using Machine Learning for the Humanoid Robot RH5”
The Robotics Innovation Center research department, headed by Prof. Dr. Dr. h.c. Kirchner, develop robot systems that are used for complex tasks on land, under water, in the air, and in space. The recently established underactuated lab at DFKI-RIC is looking for outstanding students to join us in pushing the boundaries of highly dynamic and agile robots.
Topic:
Mobile legged robots rely on state estimation algorithms to determine the robot position and orientation in the world reference frame. Typically, proprioceptive sensors such as inertial measurement unit, joint encoders, and foot contact sensors are used to retrieve the robot state by means of sensor fusion approaches such as Kalman filter variations. However, the robot model is subject to uncertainties that affect the quality of model-based state estimation algorithms. Machine learning models [1] can be employed to cope with model uncertainties and learn the unmodelled dynamics [2]. The goal of this thesis is to design a neural network model which learns the unmodelled dynamics of the RH5 humanoid robot and compensates for the model uncertainties in the Kalman filter framework. The performance of hybrid state estimation will be evaluated using a marker-based motion capture system as ground truth and compared to a state of the art method for state estimation [3]. For experimental evaluation, the humanoid robot RH5 will perform dynamic legged locomotion.
Prior Knowledge:
lMathematical: linear algebra, basic control theory.
lProgramming: C/C++, Python, Git, ROS / RoCK, ideally experience with robotic simulation software (e.g. Raisim, Pybullet etc).
Related Work:
1)Jin, X.-B.; Robert Jeremiah, R.J.; Su, T.-L.; Bai, Y.-T.; Kong, J.-L. The New Trend of State Estimation: From Model-Driven to Hybrid-Driven Methods. Sensors 2021, 21, 2085. doi.org/10.3390/s21062085
2)Buchanan, R., Camurri, M., Dellaert, F., & Fallon, M. (2022). Learning inertial odometry for dynamic legged robot state estimation. 1575–1584.
3)Hartley R, Ghaffari M, Eustice RM, Grizzle JW. Contact-aided invariant extended Kalman filtering for robot state estimation. The International Journal of Robotics Research. 2020;39(4):402-430. doi:10.1177/0278364919894385
Please contact Mihaela Popescu (Phone: +49 421 17845 4141) for further information and send your application via E-Mail to Mihaela.Popescu@dfki.de.
