Simon Guist

First sim-to-real transfer for muscle-actuated robots 🦾 Muscle-actuated robots are powerful but hard to model. GeAN addresses this longstanding challenge by learning muscle and tendon dynamics, enabling sim-to-real transfer. arxiv.org/abs/2604.09487 youtu.be/EOgdJxghYaw 🧵👇

Excited to welcome Dieter Buchler @dtrbchlr to the University of Alberta! We are bringing robotics and reinforcement learning together for an awesome, interactive, and experiential AI.

@simonguist I’m very proud that we can finally present this project at #RSS2024. We iterated on the hardware over the past 3 years to deliver a truly robust *and* fast open-source tendon-driven arm. Robot bodies play an essential role in learning control!

This was truly a collaborative effort, many thanks to my brilliant colleagues @JanS1854, Hao Ma, @clthegoat, Vincent Berenz, Julian Martus, Heiko Ott, Felix Grüninger, Michael Muehlebach, Jonathan Fiene, @bschoelkopf and @dtrbchlr

Experiments include a high-speed table tennis task (ball speeds up to 20 m/s). Open-source hardware, software, and a motion dataset: webdav.tuebingen.mpg.de/pamy2

We presented our work on a novel open-source (mostly) 3D-printable tendon-driven robot arm at #RSS2024 today. Our design enables safety through reduced inertia and passive compliance, while addressing challenges regarding friction and robustness youtube.com/watch?v=5vH2_T…

Gradient subspace optimization unlocked for RL 🔒➡️🔓 Used only for supervised learning so far, our #ICLR2024 paper illustrates that policy gradients evolve in a small, slowly-changing subspace, opening up many opportunities for more efficient RL. arxiv.org/abs/2401.06604

Title: ‘Hindsight States (HiS): Blending Sim and Real Task Elements for Efficient Reinforcement Learning' arxiv.org/abs/2303.02234 Work done at @MPI_IS with my brilliant colleagues Jan Schneider, @alexdttrch, Vincent Berenz, @bschoelkopf and @dtrbchlr (4/4)

We show how this approach can accelerate learning both on its own and when combined with Hindsight Experience Replay (HER). Check out how we learned to play table tennis on a real muscular robot with multiple virtual balls. (3/4)

Robots are limited by reality, but everything is possible in virtual environments. Our #RSS2023 paper presents Hindsight States (HiS), a method that leverages imbalances in complexity between parts of the task for efficient learning. (1/4)