NeurIPS 2021 Competition "Learning By Doing": Controlling a Dynamical System using Control Theory, Reinforcement Learning, or Causality


The 35th NeurIPS conference (December 6-14, 2021) hosts a total of 23 competitions with challenges from different fields of machine learning. The Institute for Data Science in Mechanical Engineering (DSME) co-organizes the “Learning By Doing” competition, which brings together researchers from control theory, reinforcement learning, and causality.

Logo NeurIPS  

Since its foundation in 1987, the Conference on Advances in Neural Information Processing Systems (NeurIPS) has developed into the biggest machine learning conference worldwide. It is a multi-track, interdisciplinary meeting that is held annually to foster the exchange of research on neural information processing systems in their biological, technological, mathematical, and theoretical aspects.

The “Learning By Doing” competition is organized by an interdisciplinary team from five internationally renowned universities, including RWTH (see below). The challenge aims at bringing together three fields of research: control theory, reinforcement learning, and causality are all ways of mathematically describing how the world changes when we interact with it. They all consider the problem of learning how to control an unknown system to achieve a desired effect. However, their approaches towards this problem are complementary, and each perspective has its strengths and weaknesses. Combining the different views promises to provide intriguing insights for researchers from all fields. What is more, the results and accompanying discussions might create synergies that can help to further bridge the gap between the different fields.


Further Questions



The organizers of the challenge designed two tracks in which optimal control policies for dynamic systems have to be found. Both tracks are completely separate, and participants may submit entries to both or just one track. In Track CHEM, participants are required to design an open-loop control strategy for a chemical plant while the goal in Track ROBO is to design a feedback controller for a robotic system. The CHEM Track started on July 6, the ROBO Track on July 15. The competition deadline is September 26.

Both individuals and teams can participate in the competition, and there are no prerequisites. The prices of $3000/$2000/$1000 for 1st/2nd/3rd winner will be awarded separately for each track. Entries will be reviewed by the organization committee. For further information, please visit the competition’s homepage.

University Researcher
RWTH Aachen Dominik Baumann
Sebastian Trimpe
University of Kopenhagen Sebastian Weichwald
Niklas Pfister
Jonas Peters
Carnegie Mellon University Timothy Lee
Oliver Kroemer
University of Lund Søren Wengel Mogensen
Université Paris-Saclay Isabelle Guyon