Rotate back to portrait for the best experience.

02 — Data · AI Model Performance

GR00T N1.7

Open weightsNVIDIA

3 of 6 axes published · values checked against their primary sources on 23 Oct 2026

  1. Success

    Does it do the trained task? (% of trials, trained setup)

    97.9%

    SimulationGrade B · Paper or report with method (rubric or error bars)LIBERO clean

    Clean LIBERO suite, simulation. No published real-robot success rate for N1.7 itself.

    Source: IndustrialVLA-Bench, arXiv 2609.25562 · Table 2

  2. Robustness

    Does it survive the cell drifting? (% under the perturbation set)

    75.1%

    SimulationGrade B · Paper or report with method (rubric or error bars)LIBERO-Plus

    Non-language perturbations, simulation.

    Source: IndustrialVLA-Bench, arXiv 2609.25562 · Table 3

  3. Instruction

    Does rewording the instruction break it? (% under paraphrase)

    74.3%

    SimulationGrade B · Paper or report with method (rubric or error bars)LIBERO-Para

    Paraphrased instructions, 23.6 points below clean.

    Source: IndustrialVLA-Bench, arXiv 2609.25562 · Table 4

  4. Recovery

    Does it fix its own mistakes? (% of injected failures recovered)

    Owed

    Not reported.

    Checked: Isaac GR00T repository (N1.7)

  5. Speed

    Is it as fast as a trained person? (human cycle time ÷ robot cycle time)

    Owed

    No cycle time against a person. Related, not equivalent: 19.1 ms per policy call (Table 5).

    Checked: IndustrialVLA-Bench, arXiv 2609.25562

  6. Interventions

    How often does a person step in? (per 1,000 cycles)

    Owed

    Not reported.

    Checked: Isaac GR00T repository (N1.7)

Built on GR00T N1.7

Published work that starts from GR00T N1.7: fine-tunes, add-ons and models trained from it. Their results are listed here, apart from GR00T N1.7's own values above, with the source each one comes from.

  1. R2A

    Add-onOpen code

    VinUniversity · VinRobotics · DFKI · TU Darmstadt · 27 Oct 2026

    Looks up similar moments in past successful demonstrations and blends those motions into the frozen model’s actions to keep it on track when conditions change.

    • Success82.5%+25 pts vs GR00T N1.7 in the same paper (57.5%)Real robot · Mobile ALOHA, 2 tasks · Grade BMean of Banana2Pot and Cube2Drawer (80/85 vs 60/55). Trial count not stated. With shifted start poses: 80/45 vs 30/10.
    • Robustness89%+11.9 pts vs GR00T N1.7 in the same paper (77.1%)Simulation · LIBERO-Plus, 7 perturbation types · Grade B
    • Instruction86.6%+2.7 pts vs GR00T N1.7 in the same paper (83.9%)Simulation · LIBERO-Plus, changed instructions · Grade B

    Source: R2A paper, arXiv 2610.37215

  2. Horizon–depth coupling

    Inference setting

    CASIA · FiveAges · NUS · 23 Oct 2026

    No retraining: when the arm runs longer before re-planning, the model takes fewer sampling steps per move.

    • Success95.1%−0.6 pts vs GR00T N1.7 in the same paper (95.7%)Simulation · LIBERO average · Grade BSlightly below the released setting; the gain here is speed (55.6 → 9.9 ms per call).

    Source: Execution horizon and sampling depth, arXiv 2610.32102

Know of one we missed? Send it with its source.

Compare in the scorecard →Submit a result or correctionHow values are graded

Get notified

New results on AI Model Performance, and news from B2A. Nothing else.