Jan Schneider

(5/5) Many thanks to my amazing collaborators, @mridulmahajan_ , @clthegoat , @simonguist , @bschoelkopf , @IngmarPosner , and @dtrbchlr .

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 🧵👇

(4/5) Deploying the GeAN with a GPU-based simulator enables massive parallelization and efficient RL training. The learned policies transfer zero-shot to the real robot (4-DoF, tendon-driven, actuated by pneumatic artificial muscles), producing dynamic and precise motions.

(3/5) GeAN uses the known arm dynamics to learn the mapping from commands to joint torques from joint position trajectories alone. No torque sensors needed → the method is applicable to a wide range of robots (with all kinds of actuators).

(2/5) Our method first collects a dataset of randomized open-loop motions to explore the robot's actuator dynamics. This data is used to train the Generalized Actuator Network (GeAN).

Pushing for #icra but still missing real robot experiments? 😰 Skip the ROS headaches — get your Franka robot running in minutes with franky! 🦾 Super beginner-friendly, Pythonic, and fast to set up. 🔗 github.com/TimSchneider42… @ias_tudarmstadt @Jan_R_Peters 🧵👇

#ThrowbackThursday: We welcomed Federal Minister Hubertus Heil to Cyber Valley! After viewing projects from @MPI_IS, he spoke with the Cyber Valley Community about the challenges faced by qualified specialists from abroad 👉 cyber-valley.de/en/news/bundes… #BetterTogether #AI #AIFuture

Learning to play the piano with two robot hands is super challenging, even in simulation! It requires coping with bimanual coordination at high speed to achieve human-level dexterity. We introduce RP1M, a large-scale robot piano-playing motion dataset, featuring ~1M trajectories over 2k music pieces. Website: rp1m.github.io Paper: arxiv.org/abs/2408.11048

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…

If you are at #ICLR2024 and want to hear more about our work on gradient subspaces for RL, come to poster #167 tomorrow from 10:45 to 12:45. Looking forward to chatting with you!

Many thanks to my amazing collaborators @rlfromlux, @simonguist, @clthegoat, @HaeufleLab, @bschoelkopf, @dtrbchlr.

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