Definition
Plain language
A robot-programming system that learns from its own mistakes by writing itself a readable notebook of debugging rules, without ever changing the underlying AI.
As stated in the literature
An agentic skills-discovery framework for code-as-policy robotics that pairs a detailed execution engine (per-call inputs, outputs, and before/after camera snapshots) with an induced text skill library and evolutionary program search over a frozen frontier model, accumulating reusable repairs as inspectable text rather than weight updates.
Why it matters: It lets a robot get more reliable over time through readable, checkable notes instead of costly and opaque retraining of the underlying model.
For example, after a robot keeps grabbing the wrong cup because two look alike, ASPIRE writes itself a note to sort objects by their position and applies that fix next time.
Heard on the show
“The paper is called ASPIRE, out of NVIDIA and a big academic collaboration, posted at the very end of June 2026 — so just days before we recorded this.”Episode 194 — How a Robot Builds a Debugging Notebook It Can Read, Edit, and Hand to Another Robot