Drones and robots are getting computer vision with higher-resolution cameras and artificial intelligence to recognize objects and images.
Many are made with developer boards like Nvidia’s Jetson TX1, which provides the smarts for auto-navigation and collision avoidance. TX1 has the horsepower to process live image feeds, and software tools to instantly analyze and provide context to visuals.
The TX1 is now a lot faster and better equipped to handle AI and image processing. Nvidia’s new Jetpack 2.3 software tools for TX1, announced on Tuesday, are a major update that doubles the deep-learning performance of the board.
The first TX1 went on sale earlier this year, but the new software tools will make drones and robots more reliable. The tools will also make TX1 more relevant in medical imaging and surveillance systems. A surveillance camera made with TX1 could recognize faces in a crowd, identify objects or count the number of people in a frame.
The TX1 has a quad-core CPU and 256-core GPU, which is enough to run algorithms and on-board analytics. The board relies on a new machine-learning engine called TensorRT to analyze pixels and provide the right context to images.
It’s possible to make robots with developer boards like Raspberry Pi, but most don’t have the AI software tools to support machine learning.
Computers can be trained to recognize images and speech through deep-learning systems, as proven by companies like Microsoft, Facebook and Google. Deep learning is a type of machine learning in which algorithms aid in cross-referencing and classification of data.
Training models and rich data sets for TX1 can be created on larger servers with teraflops of processing power, and then exported to the developer board. For example, image recognition involves classification and labeling of pixels. The TX1 processing payload isn’t as high as servers, but it is powerful enough to approximate answers by cross-referencing the trained models.
Algorithms run two times faster compared to the original software tools that shipped with TX1 earlier this year, said Deepu Talla, general manager and vice president for Tegra products at Nvidia.
The improved software stacks do the tuning and recompiling to ensure more efficient use of the GPU when running deep-learning tasks, Talla said. Nvidia’s deep-learning tools are called CUDNN (CUDA Deep Neural Network), and TensorRT sits on top of it.
Deep-learning relies heavily on parallel processing for quick computations. Nvidia has added support for CUDA 8, which helps programmers write applications that harness the computing power of all GPU and CPU cores on the TX1. CUDA 8 is also used on most Nvidia GPUs, so programs written on TX1 could be exported to other computing devices.
The new software stack has new multimedia APIs (application programming interfaces), which will allow more cameras to be attached to the board. For example, if a developer wants to attach a specific 4K camera, that can be done, Talla said.
The Jetson TX1 developer kit has a quad-core 64-bit ARM CPU and GPU based on the Maxwell architecture. It sells for US$599.99.