Master Thesis Hierarchical Reinforcement Learning Approach for Computational Resource Allocation of Concurrent Control Applications

In modern control systems, multiple control applications often operate simultaneously on a shared computational platform. The available computational resources, however, may be limited or costly, necessitating the need for an efficient resource allocation. By strategically degrading some control applications, the resource constraints can be managed at the cost of some accuracy. Recently, the problem has been tackled with various optimization and heuristic techniques aimed at maximizing the control performance within the available resource constraints. The aim of this thesis would be learning a more efficient resource allocation policy by developing a Reinforcement Learning (RL) agent.

• During your thesis you will develop and implement a Reinforcement Learning (RL) agent for optimal resource allocation in advanced control applications deployed on a shared platform with limited computational resources.
• You will conduct a state-of-the-art analysis to frame the topic in the existing control systems literature.
• Furthermore, you will translate the elements of the control framework into state space, action space, and reward function of the RL agent.
• Finally, you will test and demonstrate the approach by simulation in a relevant industrial usecase.

• Education: Master studies in the field of Control Engineering, Robotics, Mathematics, Computer Science or comparable
• Experience and Knowledge: solid knowledge of control systems and optimization; coding experience in Python; familiarity with machine learning; experience with frameworks such as PyTorch, TensorFlow
• Personality and Working Practice: you collaborate effectively, work autonomously and systematically, meticulously document your processes, and clearly present your results
• Languages: very good in English