Top AI Product

AGIBOT dropped GO-2 on April 9 — a next-gen embodied AI foundation model that claims to solve what the company calls the “last mile” between reasoning and physical execution. The predecessor GO-1 could see and plan, but stumbled on consistent execution. GO-2 unifies both into a single architecture: Action Chain-of-Thought for spatial reasoning, plus an Asynchronous Dual-System that decouples planning from real-time motor control. Trained on tens of thousands of hours of manipulation data, it reportedly beats state-of-the-art on multiple robotic benchmarks.

The hardware running it is the G2 — a 175 cm, 55 kg industrial wheeled humanoid with 49+ degrees of freedom. The brains: NVIDIA Jetson Thor delivering 2,070 TFLOPS at FP4, a 7.5x jump over the previous compute module. Force-controlled 7-DOF arms hit sub-millimeter precision with 0.5N force sensitivity — AGIBOT demoed live RAM insertion at CES 2026. Dual 3D LiDARs, stereo depth, fisheye cameras, and an 8-mic array feed the sensor stack. Hot-swappable batteries keep it running 24/7.

Nearly 1,000 units are reportedly ordered by a major electronics manufacturer.

For Developers: NVIDIA Isaac GR00T Stack

This is where it gets interesting for the API crowd. AGIBOT is a GR00T ecosystem partner — their Genie Sim 3.0 platform integrates directly with NVIDIA Isaac Sim on Omniverse. Developers can fine-tune and deploy models through the Isaac GR00T N1.6 SDK, run sim-to-real transfer training, and push policies onto Jetson Thor via the Isaac ROS stack. Genie Sim 3.0 itself is fully open-source (10,000+ hours of synthetic data, 100,000+ simulation scenarios), and supports LLM-driven natural language scene generation — describe an environment, get a structured sim automatically.

Why It Matters Now

National Robotics Week timing, NVIDIA co-sign, and a real deployment pipeline with industrial orders. Most humanoid companies are still doing lab demos. AGIBOT is shipping units and open-sourcing its simulation stack. If you’re building on the GR00T ecosystem, the G2 is one of the few platforms where the full NVIDIA robotics pipeline — from sim training to edge deployment on Jetson Thor — is already integrated end to end.

You Might Also Like

• Anvil Robotics Openarm the 5000 Dual arm Robot kit Already Shipping to Nvidia
• Nvidia Nemotron 3 Super 120b Parameters 12b Active the Math Behind the Fastest Open Source Reasoning Model
• Sunday Robotics Hits 1 15b Valuation a 200 Glove 10 Million Chore Episodes and a Robot Named Memo
• Ricursive Intelligence Just Raised 335m in 4 Months and i Think Nvidia Should be Paying Attention
• Nvidia Personaplex Just Made Every Voice ai Pipeline Feel Obsolete