One more thing: In a 100% empirical side-quest, we found a surprising alternative to target nets: Bounded activations permit stable, divergence-free training! 😂 Feature normalizers like LayerNorm, BatchNorm, etc. also permit this, *with* unbounded activations like ReLU! 😆

Looks like scaling BC is hitting limits in robot learning. We need serious algorithmic improvements to unblock progress.

Absolute superstar panel at #ICRA40 - what a treat!

A new lab member just arrived! DFKI Intelligent Autonomous Systems Group hessian.AI TU Darmstadt

Introducing Torso, a bimanual android actuated with artificial muscles.

Honored to give an Early Career Keynote at #CoRL2024! My talk, "On the Quest for Robotic Embodied Intelligence," will explore how structure in robot learning enables faster, safer, and adaptive robot behaviors. Excited to share this vision! 🚀 corl.org/program/speake… #robotics

Oleg Kaidanov With properly tuned compliance, low-inertia arms allow the human to move faster💨and not worry too much about collisions💥

Another demo of our teleoperation exosuit, this time with contact-rich manipulation. 💪🦾 More capabilities coming soon!🦿 Intelligent Autonomous Systems Group DFKI Oleg Kaidanov Puze LIU Jan Peters x.com/aditya_bhatt/s…

Check out our latest work, where we train an omnidirectional locomotion policy directly on a real quadruped robot in just a few minutes based on our CrossQ RL algorithm 🚀 Shoutout to Nico Bohlinger, Jonathan Kinzel and MAB Robotics. Intelligent Autonomous Systems Group Computer Science, TU Darmstadt hessian.AI

CrossQ can learn quadrupedal walking in minutes! Use it as a lightweight replacement for SAC, and train faster in fewer gradient steps!

Operate fast ⚡️record more data ▶️ All videos at 1x speed

🚀 We’re releasing #LocoMuJoCo2 — An extensive library for whole-body imitation & RL. 🔥 Highlights: - Supports MuJoCo & MJX - Clean JAX impl of PPO, GAIL, AMP, DeepMimic - 22K+ trajectories per humanoid - Add your humanoid easily 🔗 github.com/robfiras/loco-… 🧵Details 👇