Ambarella showcases new robot platform and AWS AI programming protocol.

2020-01-05

Ambarella, headquartered in Santa Clara, California, is renowned for its chips. Recently, the company announced a new robot platform based on its CVflow architecture for artificial intelligence processing. Additionally, it has signed an agreement with Amazon Web Services to simplify the process of designing products with its chip, which aids in training machine learning models. Ambarella will showcase the highest-level version of this platform in the form of a single CV2 chip at CES 2020. This chip will perform stereo processing (up to 4Kp30 or multiple 1080p30 pairs), object detection, keypoint tracking, occupancy grid, and visual odometry. The company states that this high level of computer vision performance, combined with Ambarella's advanced image processing technology, can make robot design simpler and more powerful compared to traditional robot architectures. Jerome Giget, Senior Director of Marketing at the company, stated that this technology combines advanced imaging capabilities with the high-performance computer vision CVflow architecture, resulting in smarter and more efficient consumer and industrial robots. The platform supports the Linux operating system and also ThreadX real-time operating system, providing support for products that require functional safety. It offers a complete toolkit for image tuning, neural network porting, and computer vision algorithm development. Additionally, it supports the Robot Operating System (ROS) for ease of development and visualization. Ambarella and AWS simplify the process by integrating the Ambarella toolchain and Amazon SageMaker Neo cloud service. Developers can now simply bring their trained models to Amazon SageMaker Neo and automatically optimize them for Ambarella cvflow chips. The expected outcome is that customers can build ML models using MXNet, TensorFlow, PyTorch, or XGBoost and train them in the cloud or on local machines using Amazon SageMaker. They can then upload the models to their AWS account and optimize them for Ambarella SoCs using Amazon SageMaker Neo. They can choose CV25, CV22, or CV2 as the compilation target. Amazon SageMaker Neo compiles the trained models into executable files optimized for Ambarella's CVflow neural network accelerator. The compiler applies a series of optimizations that can make the models run up to 2 times faster on Ambarella SoCs. Customers can download the compiled models and deploy them on their Ambarella-equipped devices. The optimized models run in the Amazon SageMaker Neo runtime, specifically built for Ambarella SoCs and available in the Ambarella SDK. The Amazon SageMaker Neo runtime occupies less disk and memory compared to TensorFlow, MXNet, or PyTorch, resulting in significantly improved efficiency when deploying ML models on connected cameras.

RECENT AI NEWS

RECENT AI TOOLS