Collect AI, a startup that provides an AI platform for warehouse cameras and drones, has raised a $40 million Sequence B funding spherical led by Smith Level Capital. That’s the VC agency based by former Salesforce co-CEO Keith Block.
The Collect group first met Smith Level a 12 months in the past at a logistics convention, and “it took Keith and his group 5 minutes to get what we’re doing,” co-founder and CEO Sankalp Arora informed TechCrunch.
What Gather AI is doing is uncommon. The 4 founders met as PhD college students at Carnegie Mellon College, the place they constructed one of many first autonomous helicopters and examined it on the FBI coaching grounds in Quantico. (Block is a trustee for CMU.)
In 2017, the founders took what they realized about instructing helicopters to fly and land safely and launched Collect AI. Utilizing off-the-shelf cameras positioned on strategic transferring gear like forklifts, in addition to off-the-shelf drones flying across the warehouse, the cameras watch on-the-floor operations and log what they discover into the warehouse administration programs.
However the catch is, the AI isn’t being random about what it scans. It’s being “curious,” as Arora described it.
“My PhD work targeted on how one can make totally different sorts of flying robots curious,” he stated. “In order that they’re interested in bins and bar codes and workflows.”
Along with barcodes, they search for lot codes, textual content, expiration dates, case counts, damages, occupancy, and different objects. The concept is that they’ll uncover and predict points like low stock, misplaced inventory, and workflows that will trigger issues of safety.
Techcrunch occasion
Boston, MA
|
June 23, 2026
Additionally they work in environments unfriendly to individuals, like freezers and chilly storage.
As a result of Collect’s underlying tech was constructed years earlier than the age of enormous language fashions, this isn’t the sort of AI that an LLM makes use of.
“They’re not end-to-end neural networks,” Arora explains. “They’re classical Bayesian methods, mixed with neural networks.”
AI imaginative and prescient Bayesian methods use probability-base strategies to show computer systems how one can interpret visible information. These programs permit the expertise to study by utilizing information and prior information to make selections — which means they don’t endure the hallucination issues of LLMs.
As a substitute they “get curious,” as Arora put it, to assemble info (therefore the startup’s title) and decide on the subsequent motion primarily based on what they’ve realized.
As old-school as that sounds, Collect AI is sitting on the fringe of the subsequent huge factor in AI, typically referred to as “embodied AI.” These are robots that work together with the actual world, versus an LLM interacting through pc chat or net app.
To that finish, in December, the startup gained the 2025 Nebius Robotics award for Imaginative and prescient AI and Streaming Video Analytics. (Nebius is a Netherlands firm that gives AI infrastructure.)
Collect at present employs about 60 individuals, Arora stated, and prospects embody Kwik Journey, Axon, GEODIS, and NFI Industries. With this contemporary funding, the startup has now raised $74 million complete. Different traders embody Bain Capital Ventures, XRC Ventures, and Hillman Investments.
