Jim Fan's team says the system can refine robot-control programs and reuse successful skills, with reported benchmark results reaching 99%.
By [Ryan Merket](https://runtimewire.com/author/ryan-merket)
· Published
Primary source: [X](https://x.com/DrJimFan/status/2111128797181567466)
Why it matters #
ASPIRE puts a reusable software layer between general-purpose AI models and robot hardware: successful control programs can accumulate as skills, while performance still depends on the model, benchmark conditions and safeguards around generated code.
NVIDIA researcher Jim Fan (@DrJimFan) announced ASPIRE v1.0 in an October 16th thread on X, releasing an agentic robotics framework built to improve robot behavior by revising programs and saving working skills. The ASPIRE project is part of NVIDIA Research's Generalist Embodied Agent Research group, not a standalone startup.
source & further reading
runtimewire.com — original article
Michael Denyer syncs 15 pstack commits into a seven-runtime agent port
Elorian launches Glimpse to give AI a second look at visual evidence
OpenAI adds parallel agents to GPT-6 Pro in ChatGPT