Add as a preferred source on GoogleNvidia Makes Alpamayo 2 Edge Available to Developers on Hugging Face
The two-billion-parameter model turns camera feeds, navigation data and vehicle motion into driving paths and a running explanation.
Nvidia has made Alpamayo 2 Edge available to developers on Hugging Face, adding an in-vehicle AI model that plans driving paths and explains its decisions as it processes vehicle data.
Nvidia announced the model on Oct. 21 at GTC Berlin. It takes in camera feeds, navigation data and the vehicle’s recent movement, then produces a planned trajectory and an adaptive reasoning trace explaining the plan.
Alpamayo 2 Edge has two billion parameters and was developed from the 34-billion-parameter Alpamayo 2 Super through distillation. On NVIDIA DRIVE AGX Thor, it runs at about 10 Hz. With TensorRT optimization, the time from a camera image to a planned path is 160 to 190 milliseconds.
The model is part of Nvidia’s toolkit for Level 4 autonomous vehicles and robotaxis, which are designed to handle driving within defined conditions. TokenPost previously covered Alpamayo 2 Edge’s size and processing rate in its report on NVIDIA’s Hyperion and Bolt robotaxi plans.
Nvidia’s Alpamayo Agentic Factory, a blueprint for integrating AI agents, is expected in early 2027.
Simon Yoon
ReporterSimon Yoon reports on blockchain technology for TokenPost. Send corrections or tips to info@tokenpost.com.