{"id":"9487375c-385f-4914-a8f7-83c638f8331d","slug":"alphanist","name":"AlphaNIST","description":"AlphaNIST is a 2026 NVIDIA and University of Washington research framework that trains a single contact-rich robotic assembly policy in simulation across all 20 NIST Assembly Task Board #1 tasks using only sparse binary completion rewards, without demonstrations or per-task tuning.","brainType":"research-model","isOpen":false,"maturityStage":"research","architecture":"End-to-end RL (PPO-style) with sparse binary task rewards, beta sampling over start states, and multi-task joint training; also trains multi-pair episodes rewarded only when all pairs are complete.","reviewStatus":"unreviewed","sources":[{"url":"https://alpha-nist.github.io/","title":"AlphaNIST project site","sourceName":"alpha-nist.github.io","publishedAt":"2026-10-09"},{"url":"https://arxiv.org/html/2610.13354","title":"AlphaNIST: Emergent Assembly Skills from Binary Task Rewards","sourceName":"arXiv HTML","publishedAt":"2026-10-09"}],"keyFacts":[{"label":"NIST board sim success (paper, Oct 2026)","value":"One policy trained from scratch across all 20 NIST Assembly Task Board #1 sim tasks with sparse binary rewards only; 98.2% overall training success, no demos or per-task tuning."},{"label":"Multi-pair chaining (paper, Oct 2026)","value":"Same framework trains policies to finish ten assembly pairs in one episode; reward only when all pairs complete; no prescribed order."}],"aliases":["AlphaNIST","Alpha-NIST"],"collisionRisk":"low","reviewNote":null,"builtOnBrainId":null,"createdAt":"2026-10-14T01:59:46.835Z","updatedAt":"2026-10-14T01:59:46.835Z","jsonLd":{"@context":"https://schema.org","@type":"SoftwareApplication","@id":"https://registry.deploy.report/brains/alphanist","url":"https://registry.deploy.report/brains/alphanist","name":"AlphaNIST","alternateName":["AlphaNIST","Alpha-NIST"],"description":"AlphaNIST is a 2026 NVIDIA and University of Washington research framework that trains a single contact-rich robotic assembly policy in simulation across all 20 NIST Assembly Task Board #1 tasks using only sparse binary completion rewards, without demonstrations or per-task tuning.","identifier":"9487375c-385f-4914-a8f7-83c638f8331d","applicationCategory":"research-model","publisher":{"@id":"https://deploy.report/#organization"}},"framework_metadata":{"framework_schema_version":"0.1.0","verification_status":"unreviewed","maturity_stage":"research","lifecycle_state":null,"architectural_position":{"cohort":null,"sub_cohorts":[]},"within_cohort_verified_vs_claimed_pair":null,"cap_flags":[],"verification_depth":{"sources_count":2,"primary_source_types":["preprint"]}}}