BeNeRL Reinforcement Learning Seminar: Elle Miller from University of Edinburgh (March 12, this Thursday)
Zhao Yang
Dear colleagues,
Our next BeNeRL Reinforcement Learning Seminar (March 12) is coming:
Speaker: Elle Miller (https://elle-miller.github.io/), PhD student at the University of Edinburgh.
Title: Is RL enough? Achieving blind superhuman dexterity with the help of World Models
Date: March 12, 16.00-17.00 (Amsterdam time zone)
Please find full details about the talk below this email and on the website of the seminar series: https://www.benerl.org/seminar-series
The goal of the online BeNeRL seminar series is to invite RL researchers (mostly advanced PhD or early postgraduate) to share their work. In addition, we invite the speakers to briefly share their experience with large-scale deep RL experiments, and their style/approach to get these to work.
We would be very glad if you forward this invitation within your group and to other colleagues that would be interested (also outside the BeNeRL region). Hope to see you on March 12!
Kind regards,
Zhao Yang & Thomas Moerland
VU Amsterdam & Leiden University
Upcoming talk:
Date: March 12, 16.00-17.00 (Amsterdam time zone)
Speaker: Elle Miller (https://elle-miller.github.io/)
Title: Is RL enough? Achieving blind superhuman dexterity with the help of World Models
Zoom: https://universiteitleiden.zoom.us/j/65411016557?pwd=MzlqcVhzVzUyZlJKTEE0Nk5uQkpEUT09
Abstract: While RL has mastered the art of robotic locomotion, high-precision dexterity remains a stubborn challenge. Despite on-board sensors providing rich tactile and proprioceptive data, standard RL frameworks struggle to integrate this complex and partially observable sensory history into useful latent states for continuous control. In this talk, I demonstrate how augmenting RL with self-supervised World Models enables agents to bypass this bottleneck, achieving blind superhuman dexterity with up to a 3x performance increase over the PPO baseline (NeurIPS 2025). Whether you work in robotics or more abstract control, this is a generalisable approach for handling the representation problem in complex POMDPs. To conclude, I’ll share my key tips for conducting robust robotics RL research and - most elusively - how to get robots to do what you want!
Bio: Elle Miller is a third-year PhD student at the University of Edinburgh researching Reinforcement Learning for contact-rich manipulation. Her work leverages self-supervised objectives, such as learning world models, to bridge the representation gap in robotic agents. This research has enabled new capabilities in superhuman dexterity and humanoid physical assistance for healthcare. Elle has previously conducted research at NASA’s Jet Propulsion Laboratory, the DLR Institute for Mechatronics and Robotics, Waseda University, and a Max Planck Institute, with experience spanning robotic hands and humanoids to robotic wheelchairs and off-road vehicles. She graduated with First Class Honours in Mechatronic Engineering and Physics from the University of Sydney in 2023.
Link to personal page: https://elle-miller.github.io/