# Success-Guided Sampling (SGS): robot AI research model

Success-Guided Sampling is a 2026 CoRL research method from the [University of Washington](/companies/university-of-washington.md) and [NVIDIA](/companies/nvidia.md) (Zhang, Guaman Castro, Yin, and co-authors) that changes only which task configurations PPO resets to, weighting resets toward moderate-success states so mega-scale sim RL unlocks contact-rich assembly and hard locomotion without demonstrations.

- **Slug:** sgs
- **Type:** Research model
- **Maturity (DEPLOY ladder):** Research
- **Open source:** no

## Architecture

Standard PPO plus an outer success-history sampler over a fixed set of task configurations (initial state, goal, environment). Resets are weighted by recent success so neither trivial nor impossible configs dominate; same reward across NIST-board assembly tasks.


## Key facts

- **Contact-rich sim2real zero-shot (research-stated):** CoRL 2026 [SGS](/companies/sgs-sa.md) (UW + [NVIDIA](/companies/nvidia.md)) trains PPO with success-weighted resets; UR5e meshes gears, threads nuts, inserts pegs from RGB with no demos, zero-shot from simulation.
- **Scales to 1M parallel envs (research-stated):** Authors report [SGS](/companies/sgs-sa.md) keeps improving as parallel environments grow to one million, where uniform sampling and other baselines stall on locomotion and assembly.


## Sources (2)

1. **Success-Guided Sampling project page** · https://sgs-rl.github.io · 2026-10-09
2. **SGS release thread** · https://x.com/octi_zhang/status/2108583315544682914 · 2026-10-09


## Common questions

### What is Success-Guided Sampling (SGS)?

Success-Guided Sampling is a 2026 CoRL research method from the [University of Washington](/companies/university-of-washington.md) and [NVIDIA](/companies/nvidia.md) (Zhang, Guaman Castro, Yin, and co-authors) that changes only which task configurations PPO resets to, weighting resets toward moderate-success states so mega-scale sim RL unlocks contact-rich assembly and hard locomotion without demonstrations.

### Is Success-Guided Sampling (SGS) open source?

No. Success-Guided Sampling ([SGS](/companies/sgs-sa.md)) is recorded as proprietary on the DEPLOY registry. Model weights and source are not publicly available.

### What type of AI is Success-Guided Sampling (SGS)?

Success-Guided Sampling ([SGS](/companies/sgs-sa.md)) is a research model, built on a Standard PPO plus an outer success-history sampler over a fixed set of task configurations (initial state, goal, environment). Resets are weighted by recent success so neither trivial nor impossible configs dominate; same reward across NIST-board assembly tasks. architecture on the DEPLOY registry.

### What is Success-Guided Sampling (SGS)'s maturity stage?

Success-Guided Sampling ([SGS](/companies/sgs-sa.md)) is at the research stage on the DEPLOY maturity ladder. Research stage means active development without commercial deployments on file.

### Which robots run on Success-Guided Sampling (SGS)?

No robot models on the DEPLOY registry are recorded as running Success-Guided Sampling ([SGS](/companies/sgs-sa.md)). DEPLOY wires brain-to-model connections only when the wiring is verifiable from primary sources; absence may reflect pre-deployment or unverified manufacturer claims.


_API: GET /v1/brains/bb715c80-3a69-4fdd-8f0b-b4ad445ed9bb · canonical URL: /brains/sgs_