[NeurIPS 2025 (Spotlight)] The implementation for the paper "4DGT Learning a 4D Gaussian Transformer Using Real-World Monocular Videos"
https://t.co/x16oJxdhxv
Of course mjlab supports the native MuJoCo viewer. Makes it a breeze to pause, slow down, inspect contacts, perturb the robot, etc. There's also a brand new pane for reward visualization :)
First foundation model that zero-shot runs on any robot arm.
[📍 Fully open source]
What if robot arms could learn new tasks with zero data collection or fine-tuning?
That is what the team behind RDT2 just demonstrated.
RDT2 is the first foundation model that zero-shot deploys across robot arms, unseen scenes, objects, and instructions. Plug and play robotics…
No extra training.
✅ 10,000 hours of human manipulation data from 100 real homes with redesigned UMI hardware
✅ 7B one-step diffusion policy running at 23 Hz for real-time closed-loop control
✅ Zero-shot: pick, place, press, wipe, block arrows at 30 m/s, play ping-pong, extinguish burning incense
The result is a clear milestone:
models that generalize across embodiments and environments, moving toward embodied superintelligence.
Amazing work by the team, thanks for sharing, @songming_liu!
📍Project: https://t.co/KloV8tjPck
Code: https://t.co/Vf3HLDMRbP
this is beyond mindblowing for me.
somebody built a 5 million param language model inside minecraft, trained it, equipped it with basic conversational ability.
probably the best thing i have seen entire month.