# dexmal/dexbotic

Dexbotic: Open-Source Vision-Language-Action Toolbox

Repository: https://github.com/dexmal/dexbotic
Canonical: https://ross.abutalabs.com/products/dexbotic
Homepage: https://dexbotic.com
Language: Python
License: MIT
License Family: permissive
Topics: codebase, pi0, real-world, robotics, simulation, toolbox, vla, vln, cogact, memoryvla, openvla-oft
Last push: 2026-08-06T10:51:53+00:00

## Health v2 (maintenance only)
Score: 72/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 96, release rhythm 69, longevity 22
- inputs: {"age_days": 320, "days_push": 27, "days_rel": 204, "gap_med": 25, "n_releases_24m": 2}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1403, forks 179 (observed 2026-08-28T04:04:37.893532+00:00)

## What it is
Dexbotic is an open-source PyTorch-based toolbox for developing Vision-Language-Action (VLA) models for embodied intelligence. It unifies pretraining, fine-tuning, inference, and evaluation of mainstream VLA policies such as π0, CogACT, OpenVLA-OFT, and MemVLA for robot manipulation and navigation.

## Use cases
- fine-tune a VLA model like pi0 on my own robot manipulation data
- train vision-language-action policies for a Franka or ALOHA robot
- run inference with pretrained VLA models on consumer GPUs
- evaluate robot policies on LIBERO simulation benchmarks
- apply LoRA fine-tuning to a CogACT or pi0 policy
- unify training data formats across different robot arms
- reproduce state-of-the-art embodied AI research results

## When to choose
- you need a unified framework to train, fine-tune, and deploy multiple VLA policies
- you work with mainstream robots (UR5, Franka, ALOHA) and want standardized data formats and deployment scripts
- you want to reproduce or extend published VLA algorithms like π0, CogACT, or OpenVLA-OFT
- you need both cloud and local GPU training support for embodied AI experiments

## When to avoid
- you need general-purpose robot middleware or hardware control rather than VLA model development
- you only need classical robotics (planning, SLAM, control) without learning-based policies
- you require a lightweight inference-only runtime with minimal dependencies
- your project is not Python/PyTorch based

## Facets
- artifact type: framework
- maturity: active
- function: machine-learning, deep-learning, llm-training, llm-inference, robotics, simulation, benchmarking
- domain: robotics, machine-learning, deep-learning, artificial-intelligence, simulation
- platform: python, cloud
- tags: vla, vision-language-action, embodied-ai, robot-manipulation, robot-navigation, pi0, cogact, openvla-oft, memoryvla, pytorch, fine-tuning, lora, libero, linux, gpu, docker

## Member repositories
- dexmal/dexbotic (main) score 72

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:37.893532+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 2026-08-30T04:38:54.900713+00:00, confidence not recorded.
  - readme: https://github.com/dexmal/dexbotic (fetched 2026-08-28T04:04:37.893532+00:00, sha 4fddcebaa28e)
  - homepage: https://dexbotic.com (fetched 2026-08-29T11:53:02.072024+00:00, sha d2de08f85a6b)
- Data as of 2026-08-30T08:39:29.467469+00:00.