# PRIME-RL/SimpleVLA-RL

[ICLR 2026] SimpleVLA-RL: Scaling VLA Training via Reinforcement Learning

Repository: https://github.com/PRIME-RL/SimpleVLA-RL
Canonical: https://ross.abutalabs.com/products/simplevla-rl
Homepage: https://arxiv.org/abs/2509.09674
Language: Python
License: MIT
License Family: permissive
Topics: reasoning, rl, vla
Last push: 2026-01-06T15:01:06+00:00

## Health v2 (maintenance only)
Score: 46/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 61, release rhythm 35, longevity 33
- inputs: {"age_days": 465, "days_push": 239, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1834, forks 119 (observed 2026-08-28T04:05:42.478344+00:00)

## What it is
SimpleVLA-RL is an open-source reinforcement learning framework for training Vision-Language-Action (VLA) models for robotic manipulation, built on veRL. It improves long-horizon action planning under data scarcity, achieving SoTA results on LIBERO and RoboTwin benchmarks and supporting real-world RL training.

## Use cases
- train a VLA robot policy with reinforcement learning
- improve robotic manipulation beyond supervised fine-tuning
- run RL on LIBERO or RoboTwin simulation benchmarks
- reduce dependence on large-scale human demonstration data for robot learning
- apply RL to long-horizon dexterous manipulation tasks
- fine-tune OpenVLA-OFT with reinforcement learning

## When to choose
- you need an efficient, VLA-specific RL training pipeline built on veRL
- you have limited demonstration data and want RL to boost generalization
- you want to reproduce or extend published SoTA VLA RL results
- you need multi-environment rendering and scalable parallelization for robot RL

## When to avoid
- you only need simple supervised fine-tuning of a VLA model
- your robot stack is incompatible with supported VLA models like OpenVLA-OFT
- you lack GPU resources for large-scale RL training
- you need a general-purpose RL library unrelated to robotics

## Facets
- artifact type: framework
- maturity: active
- function: reinforcement-learning, machine-learning, llm-training, simulation
- domain: robotics, reinforcement-learning, machine-learning, deep-learning
- platform: python
- tags: vla, vision-language-action, robot-manipulation, verl, robot-learning, rlhf, openvla, gpu, linux

## Member repositories
- PRIME-RL/SimpleVLA-RL (main) score 46

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:42.478344+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:18:35.355744+00:00, confidence not recorded.
  - readme: https://github.com/PRIME-RL/SimpleVLA-RL (fetched 2026-08-28T04:05:42.478344+00:00, sha 79adef78f188)
  - homepage: https://arxiv.org/abs/2509.09674 (fetched 2026-08-29T10:58:00.427770+00:00, sha cca650a95fbd)
  - site_page: https://info.arxiv.org/about/donate.html (fetched 2026-08-29T10:58:00.437337+00:00, sha cca9c3a11c56)
  - site_page: https://info.arxiv.org/about/ourmembers.html (fetched 2026-08-29T10:58:00.441285+00:00, sha 47cbc55ff1de)
  - site_page: https://info.arxiv.org/about (fetched 2026-08-29T10:58:00.443313+00:00, sha a1f16f915a9a)
  - site_page: https://info.arxiv.org/labs/index.html (fetched 2026-08-29T10:58:00.439354+00:00, sha b14a8d05a0ec)
- Data as of 2026-08-30T08:39:29.467469+00:00.