home ” news ” Munich AI Lecture: Ludovic Righetti on synergies between reinforcement learning and MPC

Munich AI Lecture: Ludovic Righetti on synergies between reinforcement learning and MPC

As part of the Munich AI Lectures – our lecture series that attracts top-class AI experts to Bavaria through the collaboration of several Munich research institutions – Ludovic Righetti, Professor at New York University (NYU), gave an insightful lecture on the integration of Model-Predictive Control (MPC) and Reinforcement Learning (RL) in robotics. He directs the Machines in Motion Laboratory at NYU, which focuses on the algorithmic foundations of robotic motion that enable complex locomotion and manipulation tasks.

MPC is a robust framework for generating a variety of robot behaviors, which solves an optimization problem at each time step. Future states are predicted on the basis of a system model. Although MPC is powerful, it has problems with real-time optimization in complex scenarios, such as multi-contact behavior during manipulation and locomotion. On the other hand, RL is characterized by the learning of complex behaviours from interaction with the environment. RL can handle multimodal sensor technology and adapt to dynamic changes, but usually requires extensive offline calculations and large amounts of data for training.

A combined approach

Prof. Righetti’s approach shows that synergies between the two methods effectively combine their strengths. The use of numerical optimization methods to develop reliable solvers as part of MPC, together with machine learning techniques, allows complex, multimodal inputs to be processed and behavior generation to be accelerated. This hybrid approach aims to overcome the limitations of the respective approaches. In this context, he also calls for openness in finding solutions and the inclusion of approaches from other disciplines.

Call for conscious research

“Technology is not value-neutral; it shapes possibilities,” Ludovic Righetti said at the end of his lecture. He calls for the next generation not to view their research in a vacuum – researchers need to be aware of what their work can be used for. His appeal to the audience is: Stand up for your values together and create goals and boundary conditions for robotics and AI!

About Ludovic Righetti

Professor Righetti has a rich academic background, having obtained his engineering degree and doctorate in natural sciences from the Ecole Polytechnique Fédérale de Lausanne. His research career includes a postdoctoral fellowship at the University of Southern California and heading the Movement Generation and Control Group at the Max Planck Institute for Intelligent Systems. His awards include the Georges Giralt PhD Award 2010, the IEEE Robotics and Automation Society Early Career Award 2016 and the Heinz Maier-Leibnitz Prize 2016. For more information about Ludovic Righetti’s research and contributions, visit his Machines in Motion laboratory.

An overview of upcoming AI Lectures and a review of previous ones can be found on the official website.