AI Center x Sony AI Impact Talk Mireille El Gheche - Learning to Play at Human Speed: From Simulation to Expert-Level Robot Table Tennis
Event details
| Date | 15.10.2026 |
| Hour | 10:15 › 11:15 |
| Speaker | Dr. Mireille El Gheche |
| Location | |
| Category | Conferences - Seminars |
| Event Language | English |
Registration with an EPFL email is required: HERE.
! This event is restricted to the EPFL community !
Title
Learning to Play at Human Speed: From Simulation to Expert-Level Robot Table Tennis
Abstract
What does it take for a robot to compete with a human in one of the fastest and most dynamic sports?
Table tennis compresses many of the fundamental challenges of physical AI into just a few hundred milliseconds. The robot must perceive a ball travelling at high speed, estimate its position and spin, predict its trajectory, decide how to respond, and execute a precise motion, all while interacting with an intelligent and unpredictable opponent.
In this talk, I will present Project ACE, an autonomous table tennis robot developed at Sony AI that progressed from learning basic racket skills to competing with elite human players. I will discuss the complete system behind this progression: high-speed visual perception, reinforcement learning, simulation, robot control, and the iterative sim-to-real process that enabled policies trained in simulation to perform on a real robot. Beyond table tennis, ACE provides a case study in what it takes to move from impressive robotic demonstrations to robust physical intelligence. I will share some of the lessons we learned about combining learning with control, using real-world failures to improve simulation, and designing AI systems that can adapt and act under the speed, uncertainty, and constraints of the physical world.
Bio
Mireille El Gheche is a Lead Research Scientist at SoftBank Robotics, working at the intersection of artificial intelligence and robotics. Her expertise spans reinforcement learning, computer vision and perception, simulation, optimization, and real-world robotic systems, with a strong focus on translating research advances into robust physical systems.
Prior to joining SoftBank Robotics, Mireille was a Staff Research Scientist at Sony AI in Zurich, where she led research on Project ACE, an autonomous table tennis robot developed to compete with elite and professional human players. Before joining Sony AI, Mireille conducted academic research at EPFL and the University of Bordeaux, working across computer vision, and applied AI, including applications in medical AI. Her broader research interests lie in physical AI and intelligent robotic systems that can perceive, learn, adapt, and act in complex real-world environments.
! This event is restricted to the EPFL community !
Title
Learning to Play at Human Speed: From Simulation to Expert-Level Robot Table Tennis
Abstract
What does it take for a robot to compete with a human in one of the fastest and most dynamic sports?
Table tennis compresses many of the fundamental challenges of physical AI into just a few hundred milliseconds. The robot must perceive a ball travelling at high speed, estimate its position and spin, predict its trajectory, decide how to respond, and execute a precise motion, all while interacting with an intelligent and unpredictable opponent.
In this talk, I will present Project ACE, an autonomous table tennis robot developed at Sony AI that progressed from learning basic racket skills to competing with elite human players. I will discuss the complete system behind this progression: high-speed visual perception, reinforcement learning, simulation, robot control, and the iterative sim-to-real process that enabled policies trained in simulation to perform on a real robot. Beyond table tennis, ACE provides a case study in what it takes to move from impressive robotic demonstrations to robust physical intelligence. I will share some of the lessons we learned about combining learning with control, using real-world failures to improve simulation, and designing AI systems that can adapt and act under the speed, uncertainty, and constraints of the physical world.
Bio
Mireille El Gheche is a Lead Research Scientist at SoftBank Robotics, working at the intersection of artificial intelligence and robotics. Her expertise spans reinforcement learning, computer vision and perception, simulation, optimization, and real-world robotic systems, with a strong focus on translating research advances into robust physical systems.
Prior to joining SoftBank Robotics, Mireille was a Staff Research Scientist at Sony AI in Zurich, where she led research on Project ACE, an autonomous table tennis robot developed to compete with elite and professional human players. Before joining Sony AI, Mireille conducted academic research at EPFL and the University of Bordeaux, working across computer vision, and applied AI, including applications in medical AI. Her broader research interests lie in physical AI and intelligent robotic systems that can perceive, learn, adapt, and act in complex real-world environments.
Links
Practical information
- General public
- Registration required
Organizer
- EPFL AI Center