Nvidia Releases ASPIRE for Autonomous Robot Control

The open-source system enables robots to rewrite their own Python code to complete complex tasks.

Updated on Oct. 16, 2026 in Robotics

Bold flat-color editorial illustration of a robotic gripper holding a cube, representing autonomous software control. — Nvidia released ASPIRE, an open-source system that enables autonomous robots to rewrite their own control code for improved task execution. AI Illustration. Upload story photo >

Live Poll

Do you believe rapid advancements in AI robotics will improve your daily life in the future?

Nvidia’s GEAR lab has released version 1.0 of ASPIRE, an open-source system designed to enable generalist robotics through large language models. The software, now available on GitHub under an Apache 2.0 license, allows robots to write, execute, and iterate on their own Python control code.

Why it matters

By providing an open-source operating layer, ASPIRE aims to accelerate the development of general-purpose robots in external labs and startups. The system allows robots to store validated code fixes in a skill library, reducing the need to restart tasks from scratch when encountering errors.

ASPIRE v1.0 achieved a 99% success rate on 80 LIBERO-Pro benchmark tasks and a 72% success rate on the Robosuite bimanual handover test. Performance depends heavily on the underlying language model, with success rates jumping from 15% to 99% on specific tasks when switching from GPT-6 Luna to Claude Opus 5.5.

The players

Nvidia GEAR Lab

The research division at Nvidia focusing on general-purpose embodied AI and robot autonomy.

The details

The system functions by using a large language model (LLM) as a reasoning engine that writes and executes Python control code. When an action fails, the system observes the error, rewrites the code to compensate, and saves successful scripts into a skill library for future reuse. This iterative process allows the robot to handle complex tasks, such as solving a Rubik's Cube, without needing task-specific training for every possible permutation.

June 30, 2026: The ASPIRE research paper was posted to arXiv.
October 2026: Nvidia released ASPIRE version 1.0 on GitHub.

The Tech Race

ASPIRE attempts to standardize the control stack for generalist robotics, positioning itself as a foundation for modular AI hardware development. By moving beyond task-specific training, it seeks to challenge the current reliance on bespoke software pipelines that dominate contemporary robotics research.

Robotics researchers and software developers can now access the full codebase on GitHub to integrate ASPIRE into their own hardware stacks. While current performance is validated through benchmarks like LIBERO-Pro, the immediate impact is limited to developers seeking an open-source framework for autonomous code execution.

The takeaway

The trajectory of generalist robotics is shifting from rigid, pre-programmed behaviors toward adaptive systems that can rewrite their own logic. Researchers should track future performance metrics for the system on long-horizon tasks, where zero-shot success currently sits at 31%.