NVIDIA AI Launches ASPIRE: Self-Improving Robot Framework with 31% Zero-Shot on LIBERO-Pro Long Tasks

NVIDIA AI unveiled ASPIRE, a self-improving robot framework that autonomously writes and iteratively refines control programs, distilling validated fixes into a reusable skill library. On the LIBERO-Pro benchmark, ASPIRE achieved up to a 77-point improvement and 31% zero-shot generalization on unseen long-horizon tasks. This means robots can accumulate skills via self-improvement without retraining for each new task, significantly reducing development costs. For AI builders, ASPIRE offers a practical shift from manual programming to automated self-improvement, especially for task-diverse scenarios. Its zero-shot capability enables direct deployment in unknown environments, minimizing human intervention. The open-source framework can integrate into existing robot stacks, making it worth watching for long-horizon task automation impact.