Inspired by self-driving cars, Ati Motors’ autonomous mobile robots not only break free of the magnetic tape of older-generation AGVs but also break free of the straight lines and predictable grids of warehouse work. Combining tires and suspensions that can handle mechanical challenges with 3D lidar, high-precision mapping, and artificial intelligence to provide autonomy, these AMRs are venturing into real-world manufacturing environments. They’re dealing with challenging curves, uneven floors, bad lighting, metal gratings, oil slicks — you name it.

While showing off several of the company’s Sherpa AMRs at PACK EXPO Las Vegas, Ati Motors’ founder and CEO Saurabh Chandra sat down with TECH B2B Marketing’s Winn Hardin to explain how new technologies are tackling old factories with mobile robots and humanoids built from the ground up.

Winn Hardin: [00:01:44] Everybody, I’m Winn Hardin, and we’re back with Manufacturing Matters, our latest episode, where we look at the technology and trends that are shaping the global manufacturing industry. Today I’m lucky enough to be here with Saurabh Chandra, CEO of Ati Motors. Saurabh, how are you doing, sir?

Saurabh Chandra: [00:02:17] Doing good. Thanks for having me.

Winn Hardin: [00:02:18] It’s my pleasure. We’re shooting today from the Pack Expo show floor, actually. So hopefully you’re not getting a lot of ambient noise, but it’s a busy day, so you might get a little bit of something. But today we’re going to find out a little bit more about one of the more experienced and more versatile, diverse AMR portfolios that are coming to North America. So again, Saurabh, thanks for joining us today.

Saurabh Chandra: [00:02:38] Thank you for having me.

Winn Hardin: [00:02:39] All right. Let’s kick off a little bit. Tell us a little bit about, for those who don’t know, tell us a little bit about Ati Motors.

Saurabh Chandra: [00:02:45] So, as you said, we make AMRs, mainly targeted towards the manufacturing industry, have a wide portfolio. We started with tugging robots, which was quite unique to us. We now tug anywhere from 1,000 pounds to 10,000 pounds. Outdoor, indoor – I mean, that was another specialty of ours. We started being quite inspired by self-driving cars. So our approach was a bit different. Instead of making the old-generation magnetic-tape-based AGV smarter, we kind of made these self-driving cars dumber to work reliably and cost-effectively in factories. So that’s kind of how we did it. So we were quite a pioneer in using 3D lidars, using 3D navigation. And that allowed us to do all these kind of things, which was go on ramps, go outdoors, handle bad flooring. We don’t need any markers. So, as I said, we started with tugging, but now we have lifting robots. We have pallet-moving robots. Recently launched a humanoid robot with wheels and dual arms. So, yeah, we are doing a lot.

Winn Hardin: [00:03:50] Gotcha. Absolutely. And it’s interesting you draw the connection between self-driving cars and your AMR line. Because while most have a standard AMR platform, hidden wheels underneath, really made for warehouse environments, you’re making a specific point to help manufacturers outside of the warehouse environment, including, as you mentioned, being in outside environments, which is sort of a unique approach, I think, for a lot of AMR platforms. So that was kind of the genesis? Was there any other major reason that you decided, I want to make this Sherpa line of AMR robots that are very versatile, from the pallet lifters . . . this is an AGV-almost unit we’re looking at here. And then we’ve got the outdoor bots, and we’ve got standard warehouse. Was that versatility, that flexibility the big genesis for you?

Saurabh Chandra: [00:04:35] Yeah, once we decided that we want to focus on making the autonomous technology based on that kind of a base and that kind of inspiration, then we could give various embodiments to it. And that’s how I think the various embodiments have kind of come out from the different use cases that we are seeing from customers. And the thing is, the technology is now pretty generic that we can do various kinds of products on top of that technology. And that’s kind of how it slowly developed. And we now have so many products.

Winn Hardin: [00:05:03] So let’s talk about the differences between a manufacturing deployment environment and a warehouse deployment. How are they different?

Saurabh Chandra: [00:05:10] So in a warehouse you’ve got grids. You’re going in straight lines. Racks. So you can really optimize to a warehouse. And I mean, frankly, if I drop you into the middle of a warehouse, you won’t be able to tell one from the other. But that’s not true for factories, right? Every factory is very unique. They are often very challenging. Floors are dirty. Sometimes they have oil slicks. And they have very heavy-duty movement. That’s another thing which is pretty unique. The turns. Sometimes you have S-curves. Sometimes you have bad lighting. You have these small outdoor sections. You have metal gratings that you have to go over in manufacturing. So a warehouse is a good first place where a lot of AMRs went. I think it’s simpler, and it was the right thing for the industry to do, but we wanted to do something more challenging.

Winn Hardin: [00:06:12] It’s funny because when you think about it, warehouses I think represent, unless we’re going for the monster distribution centers, but if you were to say a square footage manufacturing plant versus a square footage warehouse/storage facility/distribution, you’re going to have lower CapEx on the warehouse side than you are in manufacturing. You know, the manufacturing is just wall to wall with R&D, advanced manufacturing, metrology systems, whatever it is. With most of these deployments not being greenfield. I mean, most manufacturing environments that you’re going to be servicing already have buildings. They already have facilities that might have evolved over decades or possibly longer.

Saurabh Chandra: [00:06:47] The oldest plant we’ve been to was around 90 years old.

Winn Hardin: [00:06:49] Wow.

Saurabh Chandra: [00:06:50] Yeah.

Winn Hardin: [00:06:51] So 90 years ago, they definitely weren’t building it with a perfect eye to completely laser-level floors, you know? I mean, you’re not going to have magnetic strips. You’re not going to have artifacts for dead reckoning, calibration on the fly with an AMR environment. So what are some of the biggest challenges? Obviously, we know the diversity and the uniqueness of the application. What technology is Ati using to overcome those manufacturing-specific challenges.

Saurabh Chandra: [00:07:16] So typically we make the robots really ground up, from first principles. So we’re not trying to buy a kit from somebody and kind of put it together. So we really think: What do our customers need? So physically the robots are made very robust. They’re heavy.

Winn Hardin: [00:07:34] I’m seeing a lot of steel builds around here.

Saurabh Chandra: [00:07:35] That’s right. They have the right ground clearance, the right kind of tires being used. So that’s on the mechanical side. The right suspension. But apart from that, then on the technology side, the 3D lidars help us. You brought up the right thing. They’re all brownfield deployments typically. So the facilities are already there. And that was kind of our mindset. Usually robot makers would go to customers and say, “Here’s the robot and here’s this long list of things you need to do to make the robot work.” And our approach has always been that we should not make the customer change anything for us. So our robot should be able to operate in whatever the environment we find ourselves in. So the the 3D sensing technology allows us to map the entire factory in three dimensions with a very high-precision point cloud, which allows us to do very good mapping. So it’s our own algorithm that we developed. The trick is not just the algorithm but to make sure that the algorithm can work in real time on the [00:08:34] robot, without [00:08:36] breaking your bank on compute. So there’s a lot of tough engineering challenges to make that work. And very recently we were in an automotive plant, automotive tier 1, where the floor has a lot of oil slicks. And they told us that no one else has been able to come and do a successful demo like what we did.

Winn Hardin: [00:08:58] No kidding. The oil slick is, obviously, it’s possible occlusions on your sensing elements. There’s tire and traction issues. But what else do oil slicks do to an AMR, which is usually a pretty robust unit?

Saurabh Chandra: [00:09:13] That’s right.

Winn Hardin: [00:09:15] It’s not like they’re taking corners at 75.

Saurabh Chandra: [00:09:16] No, no, we are not, we are not. But you do get slips. So what happens typically when AMRs use the 2D lidars, typically they rely a lot on counting wheel rotations to correct their localization.

Winn Hardin: [00:09:29] Dead reckoning with the lidar.

Saurabh Chandra: [00:09:31] That’s right. And what we do is, because we have a very accurate lidar-based localization, we can ignore the wheel slips. And the oil on the floor or in general a bad floor or even grime can make the slip be a very bad problem. So we are very robust against wheel slip if you have to get to the technology of it.

Winn Hardin: [00:09:50] Right on, right on. So we’ve already talked about how this approach, this 3D lidar approach, allows you to not have to worry about magnetic strips, a lot of other retrofit expensive elements as part of the facility before they can even bring the AMRs in for an initial deployment. Is lidar always based on the vehicles or are we doing digital twinning at the beginning? And is that a part of your algorithm? Tell me about the process.

Saurabh Chandra: [00:10:15] No, no. So, as I said, we don’t want the customer to deploy anything. So there’s no infrastructure at all. All we do is just run the robot, the same robot around, using a remote control initially, and capture the data. So we do the data capture for sure, because there’s some post-processing of the data initially that you need to do to get the map quality to be high. And we have, again, very good algorithms which do that. So you do loop closures, you correct the map, and you make sure your maps are very good quality. But then because we do maps which capture the whole facility, we are very robust against changes. So things keep changing. But typically the changes happen at the ground level. And we make sure that we ignore those changes, and we use features in the environment which don’t change over time.

Winn Hardin: [00:11:05] Gotcha. Is there a tagging step after you do that?

Saurabh Chandra: [00:11:09] We don’t do any tagging. That happens automatically with the algorithms. The only tagging we do – we call that annotation – is to mark stations as to where you need to go. That’s about it.

Winn Hardin: [00:11:20] For charging and docking primarily?

Saurabh Chandra: [00:11:22] And also where do you need to pick stuff up. Where do you drop stuff off. So those stations. Apart from that we don’t tag anything in the environment manually at all. Everything is done algorithmically.

Winn Hardin: [00:11:31] Okay. When you’re talking with major integrators or engineering service providers who are then contracting with the bigs to bring these in, what are some of the challenges and questions or the things that make them so excited when they first see the Sherpa group?

Saurabh Chandra: [00:11:45] We are in 70 factories all over the world. Not a single marker anywhere. That it actually does what it advertises is a surprise.

Winn Hardin: [00:12:02] We’ve got references to share with the audience. When you all write in.

Saurabh Chandra: [00:12:07] So that’s something that especially the integrators really understand because they have been through these steps with so many different providers. So that’s something that really surprises them. Also, how fast are we able to map and deploy these things. So we do demos typically on the same day. So we land in a factory, and by evening or afternoon we are actually running an actual route, with the customer not usually expecting that to happen. So there’s a lot of positive, pleasant surprises that people get when they first demo this kind of technology.

Winn Hardin: [00:12:37] And how long have you been fielding these systems? I know you’re building a new facility in Detroit right? I think you’ve already got some assembly work here in North America.

Saurabh Chandra: [00:12:46] That’s right.

Winn Hardin: [00:12:47] Right. So price competitive and everything to the U.S. market. Good for domestic production. But how many years altogether have you been building out this platform?

Saurabh Chandra: [00:12:56] Overall we are eight years old, but we’ve been selling around four. So we took four years really to build a lot of the base technology before we started selling. So because we chose to do a lot of stuff ground up rather than just pick systems off the shelf, because there are a lot of existing assumptions in the open source systems which are out there. A lot of people use these open source systems, like ROS, etc., and we don’t actually. We just build the whole system ground up ourselves. So it’s very optimized. We are able to pack in a lot of algorithms that can operate in real time on not the most expensive compute. So we use a mid-range Nvidia compute rather than the most expensive top-end one, which allows us two things, which is keep the costs under control and also keep our energy budget under control. That’s very important as an AMR that your batteries are not huge. They last a whole shift, and you can use the robot longer than it’s out for charging.

Winn Hardin: [00:13:51] So talk to us real quick. I mean, that’s a data point everyone wants to know. So what’s an average shift for these people before recharge?

Saurabh Chandra: [00:13:58] Yeah. So eight hours typically. If it’s a light load, we even go up to 10 in many places. But typically we try to go for a normal/heavy shift: eight hours. And after that, many of our smaller units you can actually just swap the battery, so that the robots get used around the clock. Some of the really heavy ones, like the 10,000-pound tugger, the battery is too heavy to swap out, but it’s just a one-hour charge after that.

Winn Hardin: [00:14:23] And how long does it take on the smaller units to swap out the battery?

Saurabh Chandra: [00:14:25] Two minutes.

Winn Hardin: [00:14:27] No kidding.

Winn Hardin: [00:14:28] All right, all right, we’re gonna hold ’em to that. We’re gonna do a video of that one later.

Saurabh Chandra: [00:14:31] Of course.

Winn Hardin: [00:14:31] That’d be cool. So we talked about not being open source ROS, ground up, which of course brings reliability when you’ve got good engineers and coders who are behind it. But what does that mean in terms of APIs? It’s very common for large manufacturers or warehouse operations to use a WMS. They’re using multiple different AMR models from different vendors, who are based on unique value propositions. But talk to us about integration, fleet management, because that’s always a critical component.

Saurabh Chandra: [00:14:57] Oh yeah. So the fleet management. We’re very open. I think there’s going to be millions of robots out there, and I don’t think we are making all of them. And nobody is. So we need to coexist with other robots. And I think we want to be a good citizen in the robot world, where we play ball with everyone else. So we have very open APIs, very open to working with any other robot maker to make sure that we can intersect paths, talk to each other.  Our fleet manager itself is VDA 5050 compliant. So that means that we can manage other robots when required — not a problem at all. And if they are VDA 5050 compliant, we can enable our APIs to let them manage us. Absolutely.

Winn Hardin: [00:15:38] Okay. Cool. Cool. So we’ve mentioned Detroit. Where is that facility at right now in terms of process, operations, what kind of time frame? Tell us about what’s the next year or two.

Saurabh Chandra: [00:15:50] So I mean so right now we are already in Rochester Hills, but we are expanding that facility to be much bigger. So we are taking a new facility, which is much larger.

Winn Hardin: [00:16:00] Awesome. Congratulations.

Saurabh Chandra: [00:16:00] Yeah. Thank you so much. We love being in Detroit. It feels like home for our kind of company. And I think over time, bringing more and more of our manufacturing here itself in the U.S. That’s the plan.

Winn Hardin: [00:16:19] Okay. So being in Detroit, obviously, I’m going to guess that you’re manufacturing focused, just not warehouse focused.

Saurabh Chandra: [00:16:26] Mostly manufacturing focused. Actually, we as of now actually don’t have any warehousing customers at all.

Winn Hardin: [00:16:32] Before we jump into some other things, what are some common manufacturing applications that you guys are solving? Is it fetching parts to assembly areas or unloading? You know, you’ve got depalletizers and AGVs.

Saurabh Chandra: [00:16:46] With the pallet mover now actually we are starting to get more and more into warehousing too. But so far most of our work has been moving stuff between warehouse and production area, and work in progress within the production area itself. So you’re going from one process to another process. You may be going from heat treatment to deburring or from the stamping area.

Winn Hardin: [00:17:09] So a known staging area. So we’re going from staging area to staging area.

Saurabh Chandra: [00:17:12] That’s right.

Winn Hardin: [00:17:12] What does that do in terms of productivity gains? I’m not sure you’ve been able to collect enough data from your customers. Obviously you’ve got enough around the world – 70 to 80 deployments.

Saurabh Chandra: [00:17:22] We do actually have quite a few. One of the customers – pretty interesting – they started with six of our tuggers four years ago roughly. And they have doubled their production with the same number of tuggers.

Winn Hardin: [00:17:36] No kidding.

Saurabh Chandra: [00:17:37] Yeah.

Winn Hardin: [00:17:37] And the same number of shifts, everything else being constant?

Saurabh Chandra: [00:17:40] That’s right. So they run 24/7 already. And they’ve doubled their production area within the same facility. And the way it has happened is very interesting. So we kept integrating deeper and deeper into the software systems. So now their ERP directly tells our fleet manager where the material movement is. And we have been doing very interesting, intelligent stuff like batching, batching trips. For example, we would wait for a finite amount of time for multiple orders to come in so that we can drop them in the same trip, for example. So very interesting outcomes, which come in once you start thinking of these robots really as software-defined material-moving endpoints. And then you can do a lot of optimizations and increase overall productivity, which can’t be unlocked really with manual processes how much ever you may try.

Winn Hardin: [00:18:30] Are you using deep learning or AI to a certain extent – I think I know the answer to this question – just for process optimization and efficiency gains?

Saurabh Chandra: [00:18:38] Actually, these are simple, very simple problems which don’t require deep learning to be honest. Okay. I mean, these are very old optimization techniques which can be used. The deep learning that we typically use has been on the robot side. Like a pallet detection is very, very good because we use deep learning there.

Winn Hardin: [00:18:55] And you’ve got so many different pallet styles and everything.

Saurabh Chandra: [00:18:58] That’s right. And we can figure out their orientation. So especially in a brownfield environment, it’s not always going to be perfect. And I think that’s what AI is very good at, where things have a lot of variability. But what we do is we combine the deep learning thing with more classical on top to get also the reliability. With deep learning, you handle variability, but you don’t get the reliability often. You will get a certain percentage. And that’s where deep learning usually is very good in digital places where it’s okay to be wrong sometimes, but it’s not okay to be wrong on a factory floor.

Winn Hardin: [00:19:33] It could be a safety concern. There’s a lot of things that can o on: million-dollar-per-minute loss, downtime.

Saurabh Chandra: [00:19:37] Absolutely. So we make sure that we combine good AI techniques for what they are good at and combine them with classical techniques for what they are good at. So combine the handling of variability with the handling of predictability that we need.

Winn Hardin: [00:19:49] Gotcha, gotcha. So anything coming in the next year or two that we should be looking down and be excited about coming, that you can share?

Saurabh Chandra: [00:19:58] Absolutely. I mean, we recently launched the Sherpa Mecha, which is basically our point of view on the humanoid. In factories we really think of it like the super-worker that we need. We’re not trying to mimic merely a human being but to outperform what people do. Which is what you want machines to do, right? Machines have to do much more, lift heavier loads, do it faster, have more endurance.

Winn Hardin: [00:20:23] Without productivity gains, what’s the point?

Saurabh Chandra: [00:20:23] That’s right.

Winn Hardin: [00:20:24] Unless its dirty, dangerous, dull — which is a good reason too.

Saurabh Chandra: [00:20:26] That’s right. Correct. So we don’t want to do a bipedal. I think in a factory environment, wheeled base is the right way to do it. And we already have very good 3D navigation which we are able to reuse with the wheel base that we have and then the dual arms. We have done custom arms which are carbon fiber, such that they can do very heavy payloads. We are doing a variant which can do 50 pounds. So that’s the kind of stuff we are very, very keen on. Cameras in the hands, which can do inspection in a very different way. So really think of the humanoid as something which we can really make a super-worker of sorts for a factory.

Winn Hardin: [00:21:04] So when’s the next public event that everyone’s going to be able to see this new humanoid iteration?

Saurabh Chandra: [00:21:10] Q1 next year.

Winn Hardin: [00:21:10] Q1 next year. So definitely at Automate. And is that Detroit or Chicago?

Saurabh Chandra: [00:21:15] This time Chicago.

Winn Hardin: [00:21:16] Chicago. All right. Fantastic. So everyone keep an eye out for Automate 2026. Probably May-ish.

Saurabh Chandra: [00:21:22] Yeah, yeah. By that time for sure.

Winn Hardin: [00:21:24] Wonderful. I can’t wait to see it. Imean there’s a lot of hype about humanoids. There’s some good applications out there. I love the classic approaches, or actually it’s more of a traditional approach. I mean, let’s realize that a pedestal with two arms can be useful for a lot of purposes, but legs aren’t always the best possible option. Bringing a whole lot of complex physics into it, which brings in safety concerns, which brings in questions about regulation, standards, and everything else.

Saurabh Chandra: [00:21:50] That’s right. And the simplest thing: Where do you keep the battery? You have to keep it in the torso, which makes you top-heavy. And then you have an endurance of like four hours. So as soon as I see a robot, I’m always figuring out: Where’s the power source and how long does it run? And is it stable?

Winn Hardin: [00:22:10] Well, you’re doing full eight-hour, possibly 10-hour cycles. And you know, that’s fantastic, and you’ve got a two-minute swap-out. Did you mention what the charge time was?

Saurabh Chandra: [00:22:20] Oh, yeah. You swap out in two minutes, but when you charge: one hour.

Winn Hardin: [00:22:27] For a full eight-hour charge?

Saurabh Chandra: [00:22:29] That’s right. Yeah. And the Mecha, by the way, will also have a 10-hour run. So only humanoid which should have that kind of a run time.

Winn Hardin: [00:22:38] Fantastic.

Saurabh Chandra: [00:22:39] Yeah.

Winn Hardin: [00:22:39] Saurabh, thank you so much for joining me today. I know you’re taking time out of the busy Pack Expo week you’ve got here. I hope it’s been a successful week.

Saurabh Chandra: [00:22:45] Oh, yeah. I mean, everyone’s saying this is one of the best-attended Pack Expos, so it’s been good being here.

Winn Hardin: [00:22:51] Absolutely. Thanks again for joining us. We look forward to hearing some other great things from Ati Motors. Everyone keep an eye out. Make sure you try to visit them in Rochester Hills. If you’ve got any questions for Saurabh, feel free to send them to us. We’ll make sure you get to him. What’s the website again?

Saurabh Chandra: [00:23:04]  www.atimotors.com.

Winn Hardin: [00:23:07] Fantastic. You can catch other episodes of Manufacturing Matters at manufacturing-matters.com or wherever you find your favorite podcast hosts. And until our next show and until we meet in Detroit, hopefully sooner than that.

Saurabh Chandra: [00:23:19] Absolutely.

Winn Hardin: [00:23:19] Come back and see us in a few weeks or so. Actually, days. We’re producing stuff all the time. So like, subscribe, and we’ll see you soon.

Saurabh Chandra: [00:23:26] Thank you.