Robots and drones for inventory were among the trending technologies seen at this past spring’s trade shows, from Manifest to ProMat. Gather AI was among the exhibitors and has continued to make strategic moves.
Founded in 2017, it recently expanded into cold storage environments. Early this year, Gather AI announced new heads of sales, finance, and marketing. In February 2025, the Pittsburgh-based company launched a partner program and said it will use U.S.-made ModalAI Starling 2 Logis drones in addition to systems from China’s DJI.
Automated Warehouse spoke with Sankalp Arora, Ph.D., co-founder and CEO of Gather AI, about its new drones, big data, and warehouse intelligence offerings.
Gather AI adds U.S. inventory platform
Geopolitics and trade concerns have affected the drone industry. What are some advantages of Gather AI using a new drone for inventory?
Source: Linkedin
Arora: We launched Starling 2 at Manifest. We’ve always been drone-agnostic. We were on DJI’s platform, and now, we’re on ModalAI’s platform as well.
The main reason is that it’s twice as fast in terms of inventory data gathering. Starling gives us 3D data for better case counting and depth estimation, and it does obstacle avoidance, which means it can operate in live facilities.
In addition, up until now, people had to swap batteries for our drones. Now, we have a battery swapper that the drone lands into. The battery is swapped, and it takes off again. We never liked the idea of a drone sitting for an hour to charge, so we wanted to release a fully autonomous version where the drone only needs to sit for two minutes, and then it can go back out.
Did you develop this specific charging technology in house, or did you work with partners?
Aroroa: We partnered with Hextronics, which makes swappers. It used to make batter swappers for DJI, and we asked it to develop a swapper for the new drone for us. It’s their IP, as we strictly adhere to our “no-hardware” policy.
What have the reactions been so far to the Starling 2?
Arora: Our existing customers saw how fast it was compared with DJI, and we have a list of people to serve with the new platform.
The new drone lets operate in aisles as narrow as 4 ft., 5 in. [134 cm], and that has opened up a few conversations that people were waiting for. At ProMat, we tripled our pipeline from the past five years.
Data is more important than drones
How do improvements in perception help with both navigation and inventory data collection?
Arora: First of all, the system is able to sense obstacles that are moving around. For instance, if you left a forklift parked in the aisle, it will just move around it.
At the other end, with depth sensors, we released a case-counting capability in October or November of last year. The 3D interface enables to move to partial and mixed pallets.
Usually, we focus on software so that we can provide the best insights from vision to inventory. The new sensors augment that, and there’s so much opportunity in partial pallets.
It’s not just about the amount of data you collect, but also the accuracy, right?
Arora: At ProMat, Sean Mitchell [vice president of customer success at Gather AI] and Andy Johnston [senior director of design engineering and innovation] at GEODIS spoke about the downstream impacts of improving inventory accuracy. What changes when a facility goes from 99% to 99.99% accuracy?
It’s great that GEODIS could cite real-world examples. We constantly tell our customers that drones don’t matter. The data you get matters. Customers don’t care if it’s collected by a drone or by a person with a stick, except you’d need somebody to hold that stick.
What do you think of the proliferation of drones and mobile robots for inventory such as systems from Corvus, Dexory, or Verity?
Arora: It’s a natural response to demand for throughput in e-commerce and because the space is struggling for labor.
People now understand that this is the right modality to solve this problem and get their warehouses [to be] proactive. I see this as validation — there is so much white space right now and so many businesses that can benefit.
The only challenge is for customers to find a partner they can trust. We’re fortunate to have a track record with known names they can reference.
MHE Vision looks across the warehouse
Speaking of partners and modalities, Gather AI has also launched MHE Vision, which uses a camera and artificial intelligence for visibility into material handling equipment (MHE). How did you expand from inventory drones to pallet tracking?
Arora: We exist to provide value to our customers and their operations. We are now putting camera suites on other equipment so that each MHE knows what it is carrying, what is its state, how many cases, the dimensions, whether it’s damaged or not, from the truck to the rack and from the rack back to the truck.
In true Gather AI style, it’s MHE-agnostic and real-time location system [RTLS]-agnostic. MHE Vision works with Slamcore’s RTLS, and we launched it with Noblelift’s order picker.
It’s the same technology that powers our drones in a multiple-camera forklift so that we can monitor not just how the inventory is moving during unloading, putaway, replant, and then loading. We can also monitor the productivity of each of those workflows, the productivity of workers, and whether they are following SOPs [standard operating procedures] or not.
What are some of the potential benefits of getting even more data?
Arora: If the rate of misplaced or lost inventory plummets, then warehouse throughput increases. That’s what everyone wants. For per unit dollar spent, they want their best warehouse throughput because e-commerce keeps growing.
The benefit of something like this is not just putaway. When you’re putting something on a truck, it says what truck you’re putting it in and what state was it in when it was loaded onto the truck. So OS&D [over, short, and damaged] claims also get fully digitized.
How can Inventory Intelligence lead to smarter warehouses?
Arora: In this case, you don’t have a digital twin from mapping your facility once; it’s a live digital twin that shows what’s happening in a facility and is based on the ground truth.
Our customers are constantly surprised by the number of errors in inventory placement. Once we digitize the warehouse, they realize there are gaps in how much people are reporting.
The exciting thing is that we now get enough data to power our dashboards to provide proactive suggestions, like “This operator is messing up,” or “This zone is slower than it needs to be.” “This specific vendor ships more damaged goods.”
You can now talk to our database and ask it those questions in natural language. At ProMat, people from large logistics companies would come by our booth, start using the product, and keep using it for the next 10 to 15 minutes.
What sorts of queries did visitors pose?
Arora: One of the most common questions asked was, “How is my inventory spread out, and how can I optimize it?” There’s no simple answer, and people were testing the algorithm, but our system came up with nuanced answers.
The next most common question was, “Which employees need the most help with training?” We capture every forklift movement, transaction, and inventory and can help identify where they need help.
Food distributors want to know what products are expiring. They can ask the WMS [warehouse management systems], or they can ask the copilot, which can also say what should be done by category of expiring products and how they’re distributed. They can then worry about the bananas before the ice cream and ask follow-up questions.
Customers that haven’t connected our system to the order management system [OMS] can look at all the items ordered, but it won’t be able to predict. It’s very important with LLMs [large language models] to know their limits. They can lie with confidence, which we tried very hard to avoid.
How much process re-engineering should warehouse operators expect?
Arora: I think warehouse operators should always be doing process re-engineering, but to optimize their workflows, not for automation. Once they see how their warehouses are operating, then they can bring in automation to improve them.
Gather AI building a trusted copilot
What are you fine-tuning now?
Arora: First, we’re increasing our software’s capability to integrate with more systems — the diversity is very high.
Second, we take pride in how much data we can extract from drone images, so we want to bring forklift images to that same point, from bar codes to dimensioning and pallet and case counting.
Third, we made our system conservative. It doesn’t say more than it needs to — we’re still dialing it to be trustworthy and express itself more. For example, if you have SKU description data, what is the consumption rate of this specific SKU in the market?
How do you make sure that the AI recommendations are trustworthy?
Arora: Our system can search the Internet but tries to protect the user from false information. We’re gradually gaining more confidence based on WMS, OMS, and TMS [transportation management systems]. It’s inevitable — three years from now, people will be mostly talking to the data.
If everything and every movement in your warehouse is digitized, you can optimize better. If you can query data, you can optimize it as well. You could make your warehouse like an Amazon Go store, with the dream of scanner-less workflows.