Analysis updated 2026-05-18
Compare vision-language-action approaches like OpenVLA, OpenVLA-OFT, and OpenPI in simulation.
Fine-tune and evaluate VLA models on the LIBERO and LIBERO-Plus robotics benchmarks.
Reuse tested environment setup scripts as a starting point for your own VLA research.
| johnjaejunlee95/vla-finetuning-workspace | 0xhassaan/nn-from-scratch | 3ks/embedoc | |
|---|---|---|---|
| Stars | 0 | 0 | — |
| Language | Python | Python | Python |
| Last pushed | — | — | 2023-06-08 |
| Maintenance | — | — | Dormant |
| Setup difficulty | hard | moderate | hard |
| Complexity | 5/5 | 4/5 | 1/5 |
| Audience | researcher | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires separate Conda or uv environments per subproject and likely a GPU for training.
This repository is a personal workspace for research into vision-language-action (VLA) models, which are AI systems that let robots understand instructions and turn them into physical actions. The author does not have access to real robot hardware, so the project focuses on testing these models in simulation, particularly through a benchmark called LIBERO, along with a related benchmark called LIBERO-Plus that tests performance under more difficult conditions. Inside the repository are local copies of several existing open source projects: openpi (from Physical Intelligence, covering models called pi_0, pi_0-FAST, and pi_0.5), openvla (a codebase for training and evaluating VLA models), and openvla-oft (a version of OpenVLA built for more efficient fine-tuning). A copy of the ALOHA robot codebase is also included for reference, though the author has not yet tested it in their own setup. The stated goal is to compare these different approaches and refine their settings over time, so the configuration files and setup scripts included here should be seen as starting points rather than final answers. The author plans to update them as better settings are found. To use any of the included projects, you would move into its specific subfolder and follow the instructions in that project's own README or setup files, since each one has its own dependencies and is meant to be installed separately from the others. Setup scripts such as installation.sh are provided for some of the included codebases, and openpi uses a uv-based setup instead. This is best understood as a research archive rather than a polished tool. It was not built with a general audience or onboarding materials for newcomers in mind, but it collects a related family of robotics research code in one place for comparison. Questions about code from the original upstream projects should go to those projects directly, while questions about this specific archive or its local setup can be raised through this repository's own issue tracker.
A personal research workspace that collects and configures several open source vision-language-action robotics models for simulation testing and comparison.
Mainly Python. The stack also includes Python, PyTorch, Conda.
Setup difficulty is rated hard, with roughly 1day+ to a first successful run.
Mainly researcher.
This repo across BitVibe Labs
Verify against the repo before relying on details.