TYSONS, Va. — Planet Labs, the company that made daily Earth photos a thing thanks to their smallsat constellation, is aggressively exploiting AI to turn its trove of pictures into actionable intelligence. But a senior executive cautioned against getting too caught up in the AI hype, emphasizing that substantive solutions go way beyond flashy demos.
“We have evolved in our thinking,” said Troy Toman, Planet’s senior vice president of product and software engineering.
Toman and other Planet executives spoke at an event June 12 held near Washington, D.C., aimed at U.S. government customers
While capturing the entire planet every 24 hours is undeniably cool, Planet years ago learned that customers don’t necessarily want to look at pictures, Toman said, but crave actionable intelligence and targeted insights that solve specific problems.
He recalled that in 2017, Planet started to make investments in machine learning projects “where we thought we were gonna go build these solutions to do things like ship detection, roads and buildings protection, and this was going to immediately sort of give people insights and answers to everything that they wanted to know.”
As it turned out, “we really quickly learned that while there is power in these technologies and a great deal of potential, it only really works when you have lots of other pieces that fit around that deep domain expertise, lots of situational knowledge about a region or a particular customer problem and application,” Toman said.
With AI tools today, he said, “you can do some really powerful use cases in seconds … but when you get down to providing more than the superficial sort of ‘wow’ moment, for the things that people need every day, you need much more expertise and knowledge.”
Planet in April launched an “insights platform” that injects data from the company’s large constellation of satellites and from other sources, and uses AI tools to analyze the data and produce insights.
Toman noted that U.S. intelligence agencies with significant in-house expertise analyzing imagery only care about getting raw data, which the National Reconnaissance Office buys under a multi-year contract. But for most other customers, Planet is trying to figure out how to use AI and other technologies to fix specific problems, such as tracking illegal vessels, identifying drug dealers in the Amazon, and determining where in Ukraine crops can be grown.
Staggering amounts of data
Toman oversees a staff of more than 300 software engineers and scientists. The company downloads a staggering 30 terabytes of data from its satellites every day.
“We probably have about 80,000 servers, crunching away,” he said. “The digital age has given us lots of data, but it’s a double edged sword. It allows you to see more. It’s also way more data for analysts to process, for people to try to figure out how to sort through, and creates this sort of needle in a haystack.”
AI makes it possible to quickly find “where the data is that you care about,” he said. “AI gives us a way to take all of this data and begin to turn it into insights more quickly.” However, “it is not magic.”
An example of this challenge is the maritime surveillance work Planet does with SynMax, a satellite data analytics company that specializes in worldwide tracking of maritime vessels and turns it into intelligence for customers.
“We take their data, map it to our imagery to create a history of ship traffic in a way that will stand up in court,” which means it has to be accurate, said Toman. “It takes a lot of extra work. But without AI this wouldn’t even be possible.”
Military worries about China
Commercial satellite imagery and products derived from that data are of growing interest to the U.S. military, said Maj. Gen. Gregory Gagnon, the Space Force’s deputy chief of space operations for intelligence
During a fireside chat with Robert Cardillo, chief strategist and chairman of the board of Planet Federal, Gagnon said the Pentagon is concerned about China’s rapid deployment of remote-sensing satellites.
“Unlike our remote sensing that is designed to share sunlight, their remote sensing is designed to erode privacy and freedom so that they can monitor, track and engage an adversary at ever longer ranges so that they can achieve their military objectives,” said Gagnon.
This sets the stage for a race between the U.S. and China, he said, that will require the Space Force to invest in its own sensors but also on commercial services. “We need to move fast for a very important reason, because our adversary is moving very fast.”
The next phase of AI
Planet’s phase of AI in Earth observation will be on-orbit computing, said Roman. The goal is to run AI models directly on satellites, reducing the time required to make critical identifications from hours to just seconds. Instead of beaming back loads of raw data that then gets processed on the ground, the AI could make sense of it in real-time overhead.
Later this year, the company plans to launch one of its new, high-resolution Pelican-2 satellites with Nvidia’s Jetson edge AI platform onboard. It would make Planet one of the first companies to space-qualify and fly Nvidia’s AI chips on an Earth observation satellite.
“We are constantly updating our satellite designs,” he said.
The focus on actionable insights aligns with Planet’s push towards profitability. According to its latest financials, the company reported growth primarily fueled by the government sector, especially defense and intelligence agencies.
Dave Gauthier, chief strategy officer at logistics company GXO Inc. and a former senior official at the National Geospatial-Intelligence Agency, has followed Planet’s trajectory since its early days. He sees the company’s current AI analytics push as a noticeable shift in its positioning.
“It’s interesting to hear the language used to market their capability,” Gauthier said during a panel discussion at the conference. “In the beginning, we heard ‘skim of the Earth every day.’ Then we heard about planetary variables, measuring things about the Earth and getting data. And today it’s about insights.”
Related
Source link
lol
