NVIDIA recently announced a new workflow that leverages artificial intelligence and Omniverse to accelerate humanoid robot development. The workflow dramatically reduces the amount of data needed to train a humanoid robot by combining Apple's Vision Pro with NVIDIA's AI and simulation technology.
According to IT House, traditionally, training humanoid robots requires a large amount of manual presentation data, which is often an expensive and time-consuming process. NVIDIA's new approach collects small amounts of human demo data remotely and then generates large amounts of synthetic data using its Isaac Sim simulator and MimicGen NIM microservices. This synthetic data, along with real-world data, is used to train NVIDIA's Project GR00T humanoid base model, saving time and money.
In addition, NVIDIA's Robocasa NIM microservices and Isaac Lab platform can be used to generate robotic learning experiences to further optimize robot models. The NVIDIA OSMO Cloud Platform is responsible for efficiently managing the allocation of computing resources throughout the development process, further improving development efficiency.