NVlabs/alpamayo

NVIDIA Alpamayo 1 Nano is an open 10B reasoning VLA model for autonomous vehicles that pairs driving trajectories with Chain-of-Causation reasoning.

What it solves

Alpamayo 1 is designed to improve autonomous driving by bridging the gap between high-level reasoning and action prediction. It specifically targets the "long tail" of rare or complex driving scenarios where generalizable autonomous driving is difficult to achieve.

How it works

The project implements a Vision-Language-Action (VLA) architecture that combines a Cosmos-Reason backbone with an action expert. It uses a technique called Chain-of-Causation (CoC) reasoning to generate reasoning traces alongside trajectory predictions. The model takes multi-camera video and egomotion history as inputs and predicts a 6.4-second horizon of 64 waypoints at 10 Hz.

Who it’s for

This tool is intended for researchers and developers in the autonomous vehicle (AV) domain who want to use it as a foundation or building block for developing customized AV applications or reasoning-based auto-labeling tools.

Highlights

  • Chain-of-Causation Reasoning: Generates reasoning traces to explain the predicted driving actions.
  • VLA Architecture: Integrates vision, language, and action prediction into a single model.
  • Trajectory Prediction: Provides high-resolution trajectory output (64 waypoints over 6.4 seconds).
  • CUDA Graph Acceleration: Supports optional acceleration for repeated trajectory inference to reduce launch overhead.

Related

  • Project
    NVlabs/alpamayo1.5

    Alpamayo 1.5 is a Vision-Language-Action model for autonomous driving that combines chain-of-causation reasoning with a diffusion expert to predict driving trajectories.

  • Project
    OpenGalaxea/GalaxeaVLA

    GalaxeaVLA is a vision-language-action model for general-purpose robot control that integrates reasoning and action into a single autoregressive stream for complex manipulation tasks.

  • Project
    dexmal/opendm

    OpenDM is a Vision-Language-Action (VLA) model framework for open-world robot control, providing tools for training, fine-tuning, and deploying models across various robot embodiments.

  • Project
    allenai/molmoact2

    MolmoAct2 is a family of action reasoning models that connect vision-language backbones to flow-matching action experts for real-world robot control and manipulation.

  • Project
    InternRobotics/VLAC

    A Vision-Language-Action-Critic model that provides dense reward feedback and trajectory quality screening for real-world robotic reinforcement learning.