ENPIRE lets AI agents auto‑improve real‑world robot policies to 99% success
NVIDIA's ENPIRE framework lets coding agents close the loop on real‑world robot learning: reset, execute, verify, and refine policies autonomously. Using this pipeline, agents achieved 99% success on dexterous tasks like pin insertion and zip‑tie cutting, while introducing metrics to track fleet efficiency.