sneak peek of what Xander Hinrichsen and I will show next week at RSS 2025 for the maniskill demo session: Zero shot visual sim2real (basic) manipulation, one camera and <1 hour of RL in sim with SO100 ~3 seconds to pick cubes, could be faster but goal is accessibility and low cost!

Excited to present SimbaV2 at ICML 2025 (Spotlight)! We’ll be sharing how a simple change in network architecture can significantly improve sample efficiency in RL. Come visit our poster at 4:30 p.m.- 7:00 p.m., Tuesday (07.15)!

Robots still jitter and pause even after action chunking. We introduce ACG, a test-time guidance that smooths actions of flow-based VLAs. +7% on RoboCasa +3% on DexMG with GR00T-N1-2B. Try it: arxiv.org/abs/2510.22201

People love using AMASS for humanoid motion learning — it’s great but kinda small. We built PHUMA, a large-scale, high-quality motion dataset from human videos with physical constraints. Project page: davian-robotics.github.io/PHUMA/

Kyungmin Lee did an excellent work with PHUMA, and has now provided a conversion script for ProtoMotions 🤝! Over 70 hours of high quality data for your G1! nvlabs.github.io/ProtoMotions/g… PHUMA sequence trained in IsaacLab with domain randomization --> tested in Newton.

Check out our latest work, "Actor-Critic Model Predictive Control: Differentiable Optimization meets Reinforcement Learning for Agile Flight," published in the IEEE Transactions on Robotics, where we reconcile #OptimalControl and #ReinforcementLearning, achieving the same

Our work has been accepted at #ICRA2026 ! Huge thanks to my co-fisrt author (Kinam Kim), co-authors (junhahyung , hyojin, hoiyoung, jooyeol, Hojoon Lee) and Prof. Jaegul Choo. See you in Vienna!

🎉 Really excited, our paper "XQC: Well-conditioned Optimization Accelerates Deep Reinforcement Learning" has been accepted at #ICLR2026 . If you are interested in reinforcement learning, sample-efficiency, compute-efficiency go check it out. See you in Rio!