AWS and NVIDIA have collaborated to create a Physical AI model factory using NVIDIA Cosmos 3 on SageMaker HyperPod. This system is designed to continuously generate synthetic data, train perception and policy models, and evaluate them in simulation, enabling robots and autonomous vehicles to act in the real world. The NVIDIA Cosmos 3 model, a Mixture-of-Transformers architecture, integrates various modalities like video, image, action, and sound into a single token stream, allowing it to handle data generation, post-training, and evaluation within a unified framework on AWS infrastructure. AI
IMPACT Enables continuous improvement of embodied AI systems by streamlining the pipeline for data generation, training, and evaluation.
RANK_REASON This is a product/infrastructure announcement detailing how to use existing tools (NVIDIA Cosmos 3, SageMaker HyperPod) to build a specific type of AI system, rather than a novel model release or research breakthrough.
Read on Mastodon — mastodon.social →
- AWS
- NVIDIA Cosmos 3
- SageMaker HyperPod
- Amazon EKS
- Amazon SageMaker HyperPod
- DROID dataset
- embodied artificial intelligence
- GitHub
- Mixture-of-Transformers
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →