NVIDIA’s new Jetson T3000 and T2000 modules push Blackwell-class edge AI into smaller robotics systems. The real shift is cost and deployment: 865 FP4 TFLOPS, Cosmos 3 Edge, and memory-saving agent skills are being packaged for robots that cannot depend on cloud inference.
Robostral Navigate is an 8B embodied navigation model that uses a single RGB camera instead of LiDAR, depth sensors, or multiple cameras. Mistral says it reaches 76.6% success on unseen R2R-CE environments and beats multi-sensor systems by 4.5 points.
NVIDIA and Hugging Face are bringing Isaac GR00T 1.7, Isaac Teleop, datasets, and robotics workflows into LeRobot. The move connects NVIDIA’s 3 million robotics developers with Hugging Face’s 16 million AI builders through a shared open robotics stack.
AdaJEPA targets a practical weakness in robot and agent planning: frozen world models drift under distribution shift. The method updates during MPC with one gradient step per replan and a buffer of five recent transitions.
The sharper question was not only price. HN commenters focused on what the robot does autonomously and when remote human help enters the loop.
Figure’s robot fleet has crossed its human headcount for the first time, according to CEO Brett Adcock. The chart shared with the tweet points to a steep 2026 production ramp and a workforce line near the mid-hundreds.
Anthropic’s Project Fetch phase 2 moves agent evaluation from code-only tasks into physical robotics. Opus 4.7 completed the programming portion about 20x faster than last year’s best human team using Opus 4.1, while the robot still failed the final fetch.
Alibaba’s Qwen team packaged robotics into three foundation models: navigation, manipulation, and world modeling. Qwen-Robot Suite matters because it treats physical AI as an agent stack, not a vision demo.
NVIDIA Research is tying real-time animation and robotics motion to one generative backbone. MotionBricks reports 350,000-plus motion clips, 15,000 FPS, and 2 ms latency in a SIGGRAPH 2026 project release.
NVIDIA released open-source physical AI agent skills across Omniverse, Cosmos, Isaac, Metropolis, Alpamayo, and Jetson. The company points to manufacturing gains including 67% faster training and deployment at Pegatron and a 17% detection-rate improvement at Delta Electronics.
NVIDIA released Cosmos 3 as an open physical AI omnimodel with Super and Nano variants. Its technical post points to six synthetic datasets, Hugging Face checkpoints, and GitHub recipes for domain adaptation.
NVIDIA’s open humanoid reference design combines Unitree H2 Plus hardware, Sharpa five-finger hands, and Jetson AGX Thor T5000 compute. The 75-DoF system is aimed at making humanoid research more comparable across labs.