NVIDIA officially launched its AI platform earlier at the NeurIPS AI conference in San Diego, California.A new open-source AI model named "Alpamayo-R1"This is a "Reasoning Vision Language Model" designed specifically for autonomous driving research. It will become a backbone technology for physical AI, enabling vehicles to possess "driving common sense" like humans and handle subtle driving decisions.
The first visual language motion model focused on autonomous driving, possessing "thinking" capabilities.
NVIDIA claims that Alpamayo-R1 is the industry's first Vision Language Action Model focused on autonomous driving, and that this model is based on NVIDIA's Cosmos-Reason model.
Unlike traditional models that react directly, Cosmos-Reason possesses "reasoning" capabilities, enabling it to think and analyze decisions before responding. This means that Alpamayo-R1 can process both textual and image information simultaneously, allowing the vehicle to not only "see" its surroundings but also further understand the situation and make judgments more like those of a human.
NVIDIABlog articleThe article points out that this type of reasoning-enabled technology is crucial for companies moving towards Level 4 autonomous driving (i.e., fully autonomous driving in limited areas and under specific conditions). NVIDIA hopes to use this to give self-driving vehicles more "common sense" to cope with complex road conditions.
Alpamayo-R1 is currently available as open source on GitHub and the Hugging Face platform.
"Cosmos Cookbook" Developer Guide Released
In addition to the model itself, NVIDIA also released the "Cosmos Cookbook" development guide on GitHub. This is a complete toolkit that includes step-by-step instructions, inference resources, and a post-training workflow.
The Cosmos Cookbook helps developers train and deploy Cosmos models more effectively for specific use cases, covering key aspects such as data curation, synthetic data generation, and model evaluation.
Locking in the next wave of physical AI
This new announcement also echoes the viewpoint that NVIDIA CEO Jensen Huang has been emphasizing recently: the next wave of AI will be "physical AI".
NVIDIA's chief scientist, Bill Dally, has previously stated that robots will ultimately play a significant role in the world, and NVIDIA's goal is to "create the brain for all robots." Through the Alpamayo-R1 and Cosmos ecosystem, NVIDIA is actively extending its advanced GPU computing power to physical applications such as robotics and self-driving cars.
