NVIDIA Launches GR00T 2 and Isaac Lab 3.0 at CoRL
Reported by Crypto Briefing, Unite.AI, TipRanks and 4 more outlets7 sources ↓
NVIDIA introduced GR00T 2 and Isaac Lab 3.0 at CoRL. NVIDIA reports 95% average success on selected tasks after five demonstrations; the task suite and evaluation details limit comparisons.
Key points
- GR00T 2 predicts future camera views alongside robot actions, aiming to transfer skills from human video
- NVIDIA reports 71% average success with one demonstration per task and 95% with five on selected tasks
- Isaac Lab 3.0 adds support for Newton and PhysX physics solvers and a reinforcement-learning post-training library
- NVIDIA says Isaac Lab 3.0 with Newton achieved up to 2.5 times the training throughput of version 2.3 with PhysX
- NVIDIA has not stated GR00T 2’s release terms or a weights release date
Sources
- Crypto BriefingRead ↗Nvidia unveils GR00T 2 and Isaac Lab 3.0 at CoRL conference
- Unite.AIRead ↗NVIDIA Introduces GR00T 2 for Robots Learning From Human Video
- TipRanksRead ↗Nvidia Reveals New Robot AI with 95% Success Rate in Tests
- The Robot ReportRead ↗NVIDIA launches GR00T 2 and Isaac Lab 3.0 at CoRL
- Blockchain NewsRead ↗NVIDIA Showcases GR00T 2 and Isaac Lab 3.0 at CoRL for Advanced Robot Training
Details
NVIDIA announced GR00T 2 and Isaac Lab 3.0 at the Conference on Robot Learning in Austin on Nov. 9. GR00T 2 is a robot model that predicts both a future camera view and the action expected to produce it, rather than mapping an image and instruction directly to movement. NVIDIA says this approach can transfer movements demonstrated by people to different robot bodies.
NVIDIA reports that GR00T 2 completed at least one trial on 40 of 50 tasks without task-specific demonstrations. Average success was 71% with one demonstration per task and 95% with five. The figures are company-reported: Tech Times notes that the task identities, robot types and rollout counts were not disclosed. Unite.AI says the five-demonstration average covers tasks selected for achieving at least 80% success after five episodes, a qualification that affects how the result should be interpreted.
Unite.AI reports that GR00T 2’s pretraining included 402,730 hours of first-person human video, 10,184 hours of handheld-gripper demonstrations and 23,500 hours of robot teleoperation. NVIDIA’s 100,000-hour figure refers to the portion of human data it is releasing as Human-1M-100K, according to that outlet. The articles describe the system being tested by Nexuni for laundry handling, Foxconn for server manufacturing and Siemens for industrial applications; no production rollout is reported.
Isaac Lab 3.0, generally available since Oct. 20 according to Unite.AI, lets researchers switch between PhysX and Newton physics solvers and adds a reinforcement-learning post-training library. NVIDIA reports up to 2.5 times the training throughput of Isaac Lab 2.3 with PhysX in a vision-based dexterous-manipulation benchmark. NVIDIA calls GR00T 2 open, but Tech Times reports that its announcement did not specify a license or weights release date; Unite.AI says code is expected in December.
Written from coverage as of
Images
- The Decoder
- Unite.AI
- The Robot Report
- Tech Times
Stocks in this story
Change since open: 20 Nov to 21 Nov 2026
Last refresh