RoboSkills agent-as-policy is now in the real world! 🤖
- Hard coordination: twisting the lid off a bottle.
- Precise action: picking a slim pen and stacking, with no crashes.
- Cross embodiment: one workflow for both grippers and hands!
code: https://t.co/tmWLyxmGmC
Come and try ORA v0.1 for your own harness. We show that without Astra/Fable, other models (like Qwen/Deepseek) can still achieve fairly good results on robotics benchmarks!
Stay tuned for more future works!
Been playing around with GPT-6.2 Astra today, and the improvement in 3D character animation over GPT-6 Astra is pretty noticeable! (The character is also generated by the model🤯)
Open source sharing ➡️ Codex Sculpted Policy.
In addition to GPT that can edit LLM weights, we have also explored how codex could directly edit robot policy weights to have a higher success rate
And yes ‼️ it works well :)
Github: https://t.co/kCaHCUX4pq
This is another project from Guanxing 🪼
We gave a coding agent a robot’s neural network. Then asked it to fix the bimanual handoff.
6 parameters edited. Zero gradient steps. 18/28 → 25/28 successes📈.
Meet PolicySculpt. Codex & Claude Code. Inspired by @rikiyatakehi et al.
👉 Code here: https://t.co/uN2vZ9nQrT
NEW OPEN SOURCE PACKAGE: Five agent workflows for robotics, one pip install!
RoboSkills is a pip-installable Python package from OpenRoboKit, a six-repository organization whose only institutional marker on the site is a UCSD-named favicon.
It packages five agent workflows for robotics, each held to under 1,000 lines of code by stated policy:
1. Auto Research, where an agent runs repeated simulation experiments to improve a policy and then freezes it
2. Agent as Policy, where an LLM observes an episode and chooses the next action turn by turn
3. Agentic Simulation, where an agent builds and solves a task from a natural-language instruction with privileged simulator access
4. Real2Sim, which rebuilds an interactive physics scene from a recorded demonstration video
5. Motion Generation, which turns text, an image or a video into a rendered motion with no physics at all
Underneath sit 13 tools wired into one interface: two first-party ones, RoboKit for kinematics, motion planning and hands, and GraspKit for learned grasps, plus SAM 3 and SAM 3D from Meta, MoGe from Microsoft, Kimodo from NVIDIA, and GeoCalib, HaWoR, WiLoR, GVHMR, CoACD, MuJoCo and Blender.
Everything runs in simulation, over RoboDojo, RoboCasa365, DexJoCo, LIBERO and reconstructions of DROID, DexYCB, ARCTIC, EgoDex and OakInk2.
Caveat: there is no paper, no author list and no hardware.
Excited to share that FINGR won Best Paper at the CoRL 2026 Dexterous Manipulation Workshop! 🏆
Huge thanks to my collaborators @QuanquanPeng03@mattkiim6@xiaolonw and the workshop organizers! It’s been so rewarding to bring my speedcubing hobby into robotics!
Come chat about RoboKit at #CoRL2026 today! 🤖
🌟 Spotlight S6: 8:30–9:00 AM at Hilton Austin
🪧 Poster #105: 10:30 AM–12:00 PM at JW Marriott
Check out the code: https://t.co/FdTOdu5L8d
🤖 HuG is at #CoRL2026 today!
📍 Poster #190 | JW Marriott, Griffin (2nd floor) Today 5:00–6:30 PM
Come say hi and chat about humanoid whole-body grasping!
🎮 Try our interactive web demo:
https://t.co/tBCQNS9JJD
Find it. Reach it. Grasp it. Easy for us, hard for a humanoid.
Introducing HuG (#CoRL2026): Learning Humanoid Whole-Body Affordance-Aware Grasping with Active Perception.
The robot looks around for a half-hidden object, squats or stretches its whole body to reach it, and grasps. Zero real demos, zero-shot to unseen objects and everyday scenes, from low tables to high shelves.
🌐 https://t.co/XQ8SDmZf97
Find it. Reach it. Grasp it. Easy for us, hard for a humanoid.
Introducing HuG (#CoRL2026): Learning Humanoid Whole-Body Affordance-Aware Grasping with Active Perception.
The robot looks around for a half-hidden object, squats or stretches its whole body to reach it, and grasps. Zero real demos, zero-shot to unseen objects and everyday scenes, from low tables to high shelves.
🌐 https://t.co/XQ8SDmZf97
This project kicked off last summer, and the first version was already running early this year. All the kinematics & optimization in my #CoRL2026 paper are powered by it. @jianglong_ye pushed it to the absolute limit 🔥
Honored to be part of the team building it, and one of its earliest “beta testers.” Today it’s finally open-source! We hope more people get to use this fast, scalable, and comprehensive robotics toolkit. Every repo I build from now on will be built on RoboKit, no question 🤩
Thrilled to finally share our OpenRoboKit family! 🤖
Building OpenRoboKit with @jianglong_ye and the team has been incredibly rewarding. Amazing to see it all come together, from scalable optimization to real-world applications.
All open source! If you’re working on robot learning or real-world robotics, don’t miss it!