ASPIRE explained: code-as-policy, evolving skill libraries and evolutionary search for robots
DrJimFan · x · 2026-10-17
The ASPIRE project page details a self-improving robotics system that writes and refines code-as-policy control programs from execution feedback.
Three key components:
- Closed-loop execution engine: logs observations, I/O and visual evidence for every perception/planning/grasping/control call, letting the agent localize failures and validate repairs via re-execution
- Expanding skill library: distills validated fixes into modular, transferable knowledge retrievable as in-context guidance
- Evolutionary search: generates diverse task sequences and programs beyond single-trajectory self-improvement
Real-robot demos include cross-embodiment sim2real table bussing, solving a Rubik's Cube online, unscrewing bottle caps, pill bottling and cup stacking, with baseline comparisons and fix code.
Related event: NVIDIA Open-Sources ASPIRE, a Self-Improving Robot Framework(3 posts)→
More from Embodied
- Princeton's LabOS, Featured in Nature: An AI Co-Scientist for Hypothesis and Experiment Design — MengdiWang10 · 2026-10-17
- Founder's China trip: robotics has no single hub, and robot firms work less hard — DominiqueCAPaul · 2026-10-17
- Standard Bots CEO: robots are at their '1973 moment'—and his plan to win US manufacturing — Rewkang · 2026-10-17
- Google Gemma hardware hackathon at AGI House offers $45k prizes, $15k for LiteRT track — agihouse_org · 2026-10-17
- P-BFM: unsupervised RL lets Unitree G1 humanoids adapt to rough terrain from flat-ground references — ChongZzZhang · 2026-10-17
- Researcher builds generative machine vision pipeline reconstructing scenes from monocular video — RexDouglass · 2026-10-17