Skip to content

AlphaBrain Documentation

AlphaBrain is the world’s first all-in-one, open-source community for embodied intelligence, built to be ready out of the box. We unifies multiple VLA architectures, world model backbones, biologically-inspired learning algorithms, and reinforcement learning paradigms under a single, extensible framework. AlphaBrain brings embodied AI within everyone’s reach.

AlphaBrain Architecture Overview

Where to go next

  • Installation

    Set up conda envs, Flash Attention, pretrained weights, dataset, and .env — everything you need before the first run.

  • Quickstart

    Start with Baseline VLA for the default finetune + eval, then pick a capability: NeuroVLA, RL-Token, World Model, or Continual Learning.

  • API Reference

    API reference for every public class and function in AlphaBrain/.

Key capabilities

Capability Summary Quickstart
Baseline VLA PaliGemmaOFT / PaliGemmaPi05 / LlamaOFT finetune + LIBERO eval. Baseline VLA
NeuroVLA Brain-inspired VLA with Spiking Neural Network action head and R-STDP learning. NeuroVLA
RL-Token Off-policy TD3 online RL fine-tuning with an information-bottleneck encoder. RL-Token
World Model V-JEPA / Cosmos / Wan backbones with pluggable GR00T / OFT / PI decoders. World Model
Continual Learning Experience-replay fine-tuning across 10 LIBERO tasks (LoRA or full-param). Continual Learning