Thesis (m/f/d) Learning-based control allocation for morphing UAVs

Deutsches Zentrum für Luft- und Raumfahrt e.V.

ThesisGerman needed: not said

Still open on 1 Nov

You apply on the company's own website.

Key facts

Pay: Paid by a fixed public sector pay scale
Subject you study: Computer Science, Control Engineering, Robotics, Aerospace Engineering

Literature review on learning-based methods for safe and feasible control allocation under model uncertainty
Comparison of approaches to handling uncertainty (adaptive, observer-based, set-based, statistical learning) regarding guarantees, assumptions and real-time capability
Development of a learning-based control allocation concept for a UAV with a trailing-edge morphing wing under model uncertainty
Integration of safety mechanisms (e.g. CBF-based safety filters) and strategies to ensure feasibility (fallback strategies, slack variables, feasibility certificates)
Training and benchmarking of the developed approaches under varying operating conditions and model uncertainties

Enrolled in a master's program in computer science, control engineering, robotics, aerospace engineering or a related field
Practical experience with reinforcement learning is mandatory, ideally with policy-gradient methods such as PPO or SAC
Practical experience with Gymnasium or comparable RL environments is mandatory
Good knowledge of control engineering, especially flight controller design, is an advantage
Very good programming skills in Python
Independent, structured way of working and strong interest in combining RL, control engineering and safety-critical autonomous systems
Willingness to undergo a reliability check under § 7 of the Aviation Security Act (Luftsicherheitsgesetz)
A requirement for working at DLR in Cochstedt is eligibility for a security clearance under the Security Clearance Act (SÜG) and willingness to undergo a security check under §8 ff. SÜG.

DLR stands for diversity, appreciation and equality for all people.
We promote independent work and the individual development of our employees, both personally and professionally.
For this, our many training and development opportunities are available to you.

Literature review on learning-based methods for safe and feasible control allocation under model uncertainty
Comparison of approaches to handling uncertainty (adaptive, observer-based, set-based, statistical learning) regarding guarantees, assumptions and real-time capability
Development of a learning-based control allocation concept for a UAV with a trailing-edge morphing wing under model uncertainty
Integration of safety mechanisms (e.g. CBF-based safety filters) and strategies to ensure feasibility (fallback strategies, slack variables, feasibility certificates)
Training and benchmarking of the developed approaches under varying operating conditions and model uncertainties

Enrolled in a master's program in computer science, control engineering, robotics, aerospace engineering or a related field
Practical experience with reinforcement learning is mandatory, ideally with policy-gradient methods such as PPO or SAC
Practical experience with Gymnasium or comparable RL environments is mandatory
Good knowledge of control engineering, especially flight controller design, is an advantage
Very good programming skills in Python
Independent, structured way of working and strong interest in combining RL, control engineering and safety-critical autonomous systems
Willingness to undergo a reliability check under § 7 of the Aviation Security Act (Luftsicherheitsgesetz)
A requirement for working at DLR in Cochstedt is eligibility for a security clearance under the Security Clearance Act (SÜG) and willingness to undergo a security check under §8 ff. SÜG.

DLR stands for diversity, appreciation and equality for all people.
We promote independent work and the individual development of our employees, both personally and professionally.
For this, our many training and development opportunities are available to you.