Bachelor’s/Master’s thesis in the field of automatic, data- and model-based calibration of mobile manipulators

Automatic, data- and model-based calibration of mobile manipulators
ContactName:  Dr.-Ing. Christoph Henke (IfU) Email: christoph.henke@ifu.rwth-aachen.de Type of Thesis:Bachelor’s & Master’s thesis

The precise positioning of mobile manipulators is a key challenge in modern robotics. Particularly in dead reckoning motion planning, where no external localization is used, model uncertainties, measurement noise, or insufficiently tuned state estimators can lead to significant target-actual deviations. To minimize these, optimal parameterization of the kinematic robot model and the state estimators is required.

The aim of this work is to develop a data- and model-based optimization method that determines the optimal parameterization of the robot kinematics and estimator models.

The optimization problem is formulated mathematically as a multi-criteria black-box problem. Based on a Bayesian optimization approach with a surrogate model (e.g., Gaussian process regression), parameter explorations are performed to efficiently search the parameter space. The goal is a data-efficient optimization process that delivers a near-optimal set of parameters with as few calibration attempts as possible.

Possible research questions/topics:

• How can the parameterization of robot kinematics and state estimators be formulated as a multi-criteria optimization problem?
• Which surrogate model and optimization methods (e.g., Gaussian processes, Bayesian optimization) are suitable for data-efficient parameter determination?
• How can the quality of probabilistic state estimators be evaluated using NIS and NEES and integrated into the optimization?
• How can experimental calibration trajectories be used efficiently for model validation?
• To what extent can the developed approach be transferred to different mobile manipulators?

Prerequisites:

• Degree in robotics, mechatronics, mechanical engineering, CES, computer science, data science, or related fields
• Interest in robot modeling, probabilistic state estimation, and optimization methods
• Knowledge of Python or C++
• Initial experience with ROS and machine learning methods (e.g., Gaussian processes, Bayesian optimization) is an advantage

We offer:

• Collaboration in a cutting-edge research project with practical relevance to industrial automation
• Opportunity for practical implementation and experimental validation on mobile manipulators
• Close supervision in German or English
• Modern research environment with ROS-based simulation and hardware platforms
• Opportunity to combine theoretical optimization methods with real-world robotics