Gather AI discusses drone inventory challenges and solutions

While mobile robots and stationary sensors can help track goods in the warehouse, drones are finding a key role. Editors Mike Oitzman and Eugene Demaitre recently spoke with Sankalp Arora, co-founder and CEO of Gather AI, to get an update on the use of drones for inventory management and to learn more about the Pittsburgh-based company’s product roadmap.

This interview is excerpted from Episode 153 The Robot Report Podcast — see the bottom of this article for the full episode.

Describe the inventory management problem and how important this is for warehouse owners.

Sankalp Arora, CEO of Gather AI

Arora: The inventory management problem has come to a head over the past five or six years and has become a major issue in warehouses and distribution centers. Especially through COVID-19, the acceleration of e-commerce adoption has led to warehouses shipping eaches instead of shipping out in bulk.

That’s coupled with an industry that is dealing with up to 70% attrition rates, and it’s hard to work on the warehouse floor all day. So you can’t rely on trained staff, and you have to rely on technology like barcodes that were not designed for the kind of volumes that warehouses and distribution centers have to handle today. 

As a result of all these combined things, warehouses are dealing with the increased cost of inventory accuracy in order to keep up with shipping rates and reduce shipping errors.

In addition, some operations just can’t invest so much in inventory accuracy and struggle to keep up with the necessary shipping rates while reducing shipping errors to keep their customers happy. So that’s the problem we are addressing through our solution.

What kind of information is your company gathering during an inventory scan? Are you just grabbing barcodes or are you counting individual items? How much information can the system capture?

It’s barcodes and way more in terms of what we do. Because barcodes took the supply chain so much forward, I think the next revolution in the supply chain is more going to be computer vision-powered.

As Gather AI, fortunately, we get to participate in that and be at the leading cutting edge of it. So our drones fly around and take pictures of inventory, and then pictures get converted to barcode reads.

It doesn’t matter how many barcodes are in a picture, Gather AI reads all the text on labels and on the boxes. We also calculate occupancy: how much of a pallet location is occupied or a case location is occupied.

We also do case counting, which is how many cases might be there in this specific pallet location and whether they are damaged. All of that dimensionality gets collected together and compared to the warehouse management system.

And wherever there is a mismatch, exceptions get raised so that people can act on it. You get all of this data from our AI, but you can also see the image and verify it if you feel inclined to do so. So there is full transparency on the inference that we are making.

gather ai flies inside of a warehouse.
Gather AI deploys commercial off-the-shelf drones to acquire product images in the warehouse. | Credit: Gather AI

What happens when you find an exception? How is this reported and how do you close the loop on the exception?

It depends on the specific customers’ operations, but I’ll describe a typical one. When the drone finds an exception, it flags an exception on the Gather system on the web dashboard flags.

The WMS [warehouse management system] might say X was supposed to be here, but the drone found Y. And Y was supposed to be in some other location. So in our dashboard, either you can do a walkthrough, and all the exceptions get popped up as red, or you can just go to the exception list where all the exceptions live. 

If you believe that the drone is right, then you just fix the WMS. Our accuracy numbers currently are all in the high 90s [percent], but sometimes the drone can be wrong. Since there is an image associated with the exception record, you can review the image, and if the drone was wrong, then ignore the exception.

And in some cases, maybe you can’t tell what’s happening by looking at the image. Then, someone needs to go there physically and resolve the exception.

As you’ve developed this system, what kind of technical challenges did you overcome to get this to work reliably and at scale? How did those challenges influence the choice to use off-the-shelf drones for the hardware?

We chose off-the-shelf hardware just because of reliability. For a lot of robotic solutions, when they try to make hardware, it can be so capital-intensive that you’re not able to deliver a finished solution to your customers.

With our choice came trade-offs. For example, if you use an off-the-shelf drone, there is no way to get time synchronization on any of your sensor data.

If you use an off-the-shelf drone, then you also need to be hardware-agnostic. We gave ourselves the challenge that we were not allowed to put any hardware on the drone. It has to be off the shelf because we don’t want to include any source of unreliability.

So all of our compute works on the system controller iPad. And our challenge then became, how do we make an autonomous system without any time synchronization?

That has never been done before, including all the autonomy systems that I had made before. Time synchronization was the key aspect. That is where this principle of making the drone actively curious about how to find the path so that it can stay localized and control itself and make it robust.


SITE AD for the 2024 RoboBusiness call for presentations.
Submit your presentation idea now
.


Do you use any Chinese-manufactured drones that potentially could be banned in the US?

I’m happy to answer that question. My philosophy has constantly been to give the best to our customers. And currently, the best drone on the market for this purpose is a DJI drone, which is a Chinese-based drone.

But we have done our pilots and proof of concepts with Parrot. So if a ban comes to pass, people just get switched to another drone vendor. It’s as simple as that.

But the best product in the market today is DJI. And we pass security with big Fortune 500s all the time because the drone never connects to the Internet. The data captured by the drone always goes directly to the iPad, and the drone connects directly to the network or internet.

Gather acquired the assets of Ware Robotics a year ago. It was using Skydio drones at the time of the acquisition. How has the customer integration gone? You’ve converted all those customers over to the latest Gather AI technology together now, but that’s an example of picking up a population of existing drones and deploying them with your solution.

Yeah, that’s exactly what happened. Those Ware customers are now Gather customers, and all of them are now transitioned to the same technology as the rest of our installed base.

The transition has been so joyful because these customers started with Ware when it was a young company, so they get the vision of where this is going and they are happy to be Gather customers now.

Listen to the full interview with Sankalp Arora in Episode 153 of The Robot Report Podcast.

mike oitzman headshot.
Written by

Mike Oitzman

Mike Oitzman is Senior Editor of WTWH's Robotics Group, cohost of The Robot Report Podcast, and founder of the Mobile Robot Guide. Oitzman is a robotics industry veteran with 25-plus years of experience at various high-tech companies in the roles of marketing, sales and product management. He can be reached at moitzman@wtwhmedia.com.