Saildrone puts ML at the edge to work with uncrewed ocean vessels checking weather, ocean floor

In an attempt to lower costs for ocean intelligence, Saildrone is using uncrewed vessels along with Nvidia Jetson modules for AI at the edge and Nvidia’s Deepstream software development kit for video analytics.

The startup’s approach can rely on cloud intelligence to change a vehicle’s course, but AI at the edge is mostly used because bandwidth to the cloud can be limited or costly when porting high-resolution images from sensors that range from deep sonar mapping to collecting underwater acoustics.

The data Saildrone collects is intended to help scientists, fisheries, weather forecasters, ocean mapping and maritime security. Saildrone makes use of three models of uncrewed surface vehicles in its work and a control center in the San Francisco area.    Saildrone, founded in 2012, has raised $190 million and is a member of Nvidia’s Inception program.

“We’ve sailed into three major hurricanes, and right through the eye of Hurricane Sam, and all the vehicles came out the other side,” Blythe Towal, vice president of software engineering at Saildrone, told Nvidia.  “They are pretty robust platforms.”

University of Hawaii at Manoa is using three 23-foot Saildrone vessels to study ocean acidification’s impact on climate change in a six-month mission around Hawaii.  The vessels are powered by sun and wind.

Also, Saildron has partnered with Seabed 2030 to map the world’s oceans in a collaboration with Nippon Foundation and the General Bathymetric Chart of the Oceans. A complete map of the ocean floor is needed to promote a healthy ocean for a more sustainable planet, according to Saildrone founder and CEO Richard Jenkins.

Nvidia is keen on using AI to perform climate studies and renewable energy research.  The company is building Earth-2, an AI supercomputer, to create a digital twin of Earth in its Omniverse tool.

Saildrone uses Nvidia’s JetPack SDK to run machine learning on the Jetson module. The ML runs mostly on the Jetson module locally, but can run in the cloud as well. Bandwidth can be limited and costly to port data from vessel sensors collecting high-resolution images, Nvidia said.

Sensors aboard are used to measure wind, temperature, salinity and dissolved carbon. Bathymetric sensors are used to research ocean and lake floors with deep snoar mapping.  Radar and underwater acoustic sensors are also used.

Saildrone relies on Nvidia’s DeepStream SDK for vision AI apps and services. It offers 10x faster throughput and can be used in edge to cloud to handle video, image and audio streams.  Teams used Deepstream to do image preprocessing and model inference for ML at the edge.

RELATED: Perceive looks to tain edge in busy edge AI market