As today’s industrial processes become more complex, controllers used in drivetrains for vehicles, machines, robots, process facilities, and other physical dynamic systems face increasing challenges with respect to e.g. efficiency and quality. In an industry 4.0 setting, a higher level of adaptivity and automation is required. Meanwhile, artificial intelligence (AI) is a promising enabling technology. However, examples wherein AI techniques such as reinforcement learning (RL) are directly in control of (high) power (up to kW) and (highly) dynamic (down to (m)s)) physical systems to improve energy efficiency and performance remain very limited.

Going beyond the fixed but safe structure of classical controllers and embracing the RL framework provides the ability to learn and adapt. While doing so, expensive trials and unsafe experimentation on real systems as is common in RL need to be avoided. We therefore propose a fundamentally new approach residing at the intersection of classical control and RL (CTRLxAI). Besides offering increased efficiency and performance (thrust) of the adaptive and autonomous controllers, we will strengthen the trustworthiness (trust) in terms of sample-efficiency, robustness, safety and explainability; critical capabilities for widespread industrial adoption. As such, we will realise our vision CTRLxAI=T(H)RUST.

Using dedicated infrastructure (+ a digital simulator), we will validate and showcase the added value of the CTRLxAI concepts for (i) electromechanical drivetrains, (ii) industrial robotics and (iii) multi-physical industrial processes.


Project Consortium

Within this project, the research is conducted by the academic partners. To ensure a strong connection to the needs and challenges from Industry, industry is represented in an Industrial Advisory Board (IAB).

Academic Partners

There are two academic partners in the project, the project leader is Lianne Doeswijk.

UGent logo Ghent University - Electromechanical, Systems & Metals Engineering

Principle Investigator: Prof. dr. ir. Guillaume Crevecoeur

Business Developer: Pieter Nguyen Phuc

Researchers: Tom Lefebvre, Jeroen Taets

VUB AI Lab Vrije Universiteit Brussel (VUB) - AI Lab

Principle Investigator: Prof. dr. ir. Ann Nowé

Business Developer: Leander Schietgat

Researchers: Kyriakos Efthymiadis, Denis Steckelmacher, Andrea Arcidiacono

Industrial Advisory Board (IAB)

The IAB involves 20 industrial key players which represent different stakeholders with R&D related activities in Flanders. 

  • Company A

Project objectives

Scientific objectives

In drivetrains, industrial robots, and industrial processes the dynamics are typically nonlinear and complex as multiple subsystems or physical domains are active. Experts face difficulties to fully apprehend and detail these dynamics. Also, during operation the systems change and interact with varying environments, further limiting the knowledge on the system dynamics. This hinders the design of controllers that optimize efficiency/performance while remaining trustworthy. Our goal is to go beyond the fixed but safe structure of classical controllers by embracing RL for the design of controllers that can learn and adapt. We propose a fundamentally new framework to design such adaptive controllers when only limited (approximate) prior knowledge is available and that can explore such that they learn upon observations while remaining trustworthy. By bridging research fields, CTRLxAI will open new possibilities for real control engineering problems helping to address i.a. the challenge to lower CO2 emissions and production footprint.

We will study novel cross fertilization between RL and classical control: leverage the advantages whilst overcoming their shortcomings. Theories will lead to new scientific insights; new hybrid control-RL algorithms and architectures will be furthermore derived to plan a trajectory and track setpoints. Such that non-repetitive tasks can be handled autonomously and this even under changing conditions. Our results will enable new controllers that

  1. Learn and explore safely. 
  2. Interact in a sample efficient way. 
  3. Remain stable and robust in operation. 
  4. Produce explainable policies and actions. 

We aim to go from TRL 2 up to TRL 4-5, validating our findings for fast-sampled (<10 Hz) feedback and slow-sampled (>10 sec) planning and this on several physical test setups, realistic in their scale (kW systems).

Impact objectives

In an industry 4.0 context, a higher level of efficiency, performance, automation and adaptability is required when controlling complex (high) power (up to kW), (high) dynamic (down to (m)s) physical systems as they are subjected to increasing challenges with respect to performance e.g. (energy) efficiency and quality (of movement/operation). The long term utilisation objectives of CTRLxAI are summarized by its vision ‘CTRLxAI=T(H)RUST’: to THRUST the use of TRUSTED AI in the form of CTRLxAI controllers for physical dynamic systems. Three types of applications are targeted:

  1. Electromechanical drivetrain systems in industrial machines. CTRLxAI aims at controllers that require less tuning of the control parameters along with a higher efficiency in operating conditions with larger uncertainties, while still remaining explainable.
  2. Industrial robotics. CTRLxAI aims at solutions to generate new, more natural trajectories, that are generated faster and transparently to allow for on-line adaptive trajectories; and a robust realization even in case of sudden events.
  3. Multi-physic industrial processes. The objective of CTRLxAI is to reduce the time in tuning the process conditions, while obtaining more optimal controllers learned from AI that are explainable even in the absence of detailed models and simulations. These controllers will be adaptive in order to cope with long term as well as sudden events.


Jeroen De Maeyer
+32 9 264 53 74
Technologiepark 131
9052 Zwijnaarde, Belgium


This Strategic Basic Research project is funded by FWO Research Foundation Flanders, number S007723N.

  • Starting date: 01/10/2022
  • Ending date: 31/12/2026

The project originates from the interactions in the framework of the Flanders AI Research. Program