_1722956449156001fz1R.png "Anyscale Logo")
Robotics ML presents unique challenges to ML training due to reliance on multimodal data and the necessity for extensive robot simulations to be integrated in the training process. Our training platform, built with Ray, addresses these complexities by providing tools to manage and query metadata, preprocess data offline and online, and scale up training infrastructure with simple configuration changes.
In this presentation, we will demonstrate an end-to-end machine learning training platform tailored specifically for robotics. Using RayClusters, this platform integrates physical simulation tools directly into training workflows. The platform supports multi-tenancy within the Kubernetes infrastructure, allowing many users to share compute resources.
With a live demo of the ML training process, we will showcase how this platform accelerates research and enhances the scalability of training robot policy learning models. These models can then be deployed to actual robots.
In this presentation, we will demonstrate an end-to-end machine learning training platform tailored specifically for robotics. Using RayClusters, this platform integrates physical simulation tools directly into training workflows. The platform supports multi-tenancy within the Kubernetes infrastructure, allowing many users to share compute resources.
With a live demo of the ML training process, we will showcase how this platform accelerates research and enhances the scalability of training robot policy learning models. These models can then be deployed to actual robots.
