From Tokens to Embodied Minds — a study path Reports Lab Chapters Reference My study
From Tokens to Embodied Minds  ·  Drill cards · Chapter 33
Drills

NVIDIA GR00T N1 and N1.5 with Isaac Lab

10 atomic recall cards. Export to Anki and let spaced repetition do its slow work.

10 cards due for review
Download Anki TSV ← Back to chapter

In Anki: File → Import, choose this TSV, set field separator to Tab, deck = Tokens to Embodied Minds · Ch 33, note type = Basic.

FrontBack
What is GR00T's System 1?A Diffusion Transformer (DiT) action model that executes motor commands at 50Hz, conditioned on System 2's context embedding and current proprioceptive state.
What is GR00T's System 2?A Vision-Language Model (frozen Eagle 2.5 in N1.5) that processes camera observations and language instructions, generating a context embedding for System 1 at a slower deliberative rate.
What is the FLARE loss?A trajectory-aware training objective for GR00T N1.5 that penalizes temporal inconsistency in the action sequence, encouraging smoother and more executable trajectories.
What is Eagle 2.5?The VLM backbone used in GR00T N1.5 as the frozen System 2. Provides better vision-language grounding than N1's VLM — RefCOCOg IoU 89.6, above Qwen2.5-VL-3B's 85.2.
What benchmark measures GR00T's improvement from N1 to N1.5?RoboCasa 30-demo task success: 17.4% (N1) → 47.5% (N1.5). A 2.7x improvement from three architectural changes.
What is DreamGen / GR00T-Dreams?NVIDIA's synthetic data generation pipeline: retargets real demonstrations to Isaac Lab simulation using Newton physics, generates thousands of synthetic hours from small real-seed datasets. Produced 6,500 hours from 11 hours of compute.
When was GR00T N1 announced?NVIDIA GTC, March 18, 2025.
When was GR00T N1.5 released?NVIDIA Research, June 11, 2025. research.nvidia.com/labs/gear/gr00t-n1_5/
What GPU is required for GR00T N1.5 inference?40GB+ GPU (A100 or H100 class). Not feasible on consumer 12GB GPUs — use cloud for inference or evaluation.
What is the closest reference deployment architecture to the JHU humanoid capstone using GR00T?1X Technologies NEO Gamma deployment of GR00T N1 (NVIDIA-1X partnership, March 18, 2025). Full Isaac Lab stack: sim training → GR00T post-training → on-robot deployment.