Latest / The Edge Computing Podcast with Fexingo: Local Compute, CDNs, and Distributed Infrastructure / How Edge Computing Is Enabling Real-Time Inventory Tracking in Warehouses
Transcript
- Lucas: So you walk into a massive warehouse — we're talking a million square feet, racks towering forty feet high — and you need to know exactly where every single pallet is, in real time. Luna: And without walking the aisles with a clipboard, I assume. Lucas: Right. The old way was barcode scans at receiving and shipping, and maybe a periodic manual count. But in between, stuff gets misplaced, misrecorded, or just vanishes. A typical warehouse runs about 3 percent inventory error. That's three dollars out of every hundred in goods that you can't find when you need them. Luna: That adds up fast when you're moving millions of units a day. Lucas: Exactly. So a big logistics company — I won't name them, but think top-three global parcel carrier — decided to tackle this with edge computing. They installed edge nodes on every forklift. Each node has an NVIDIA Jetson module, a LoRaWAN radio, and a pair of cameras. The idea: as the forklift moves through the aisles, it reads RFID tags on pallets and shelves, plus does real-time computer vision to confirm location and condition of the goods. Luna: And this is where edge matters — because you can't stream 4K video from every forklift to the cloud and wait for a response. Lucas: Exactly. The latency requirement is brutal. They need to know within a few hundred milliseconds whether a pallet was dropped off at the right location. If you send the video to the cloud, round-trip latency on a good day is maybe 100 milliseconds, but with congestion it can balloon to half a second or more. And that's just for one forklift. Multiply by hundreds, and your cloud bill goes through the roof. So they process everything locally on the Jetson — inference for object detection, RFID decoding, even some on-device SLAM for mapping. Luna: Quick honest thing — a handful of listeners chip in monthly through buy me a coffee dot com slash fexingo, and that's literally what funds making this many episodes. If today's tech conversation gave you something usable, a small contribution keeps it going ad-free. Lucas: Yeah, it's a small group, but it makes a real difference. So back to the forklifts — the company reported that the edge system cut inventory discrepancies by 97 percent, bringing error rates from 3 percent down to under 0.1 percent. Luna: That's huge. How does that translate to dollars? Lucas: They said it saved roughly $2 million per warehouse per year in lost goods and labor for manual counts. And they've deployed this across forty facilities so far, with plans for another sixty by end of next year. Luna: So what's the hardware setup cost per forklift? Lucas: Around $2,500 per node, including the Jetson module, cameras, RFID reader, and LoRaWAN gateway. That's about the cost of a single lost pallet of consumer electronics. So the ROI is measured in months, not years. Luna: And the software stack? I imagine you're not running a full Linux desktop on those Jetsons. Lucas: No, it's a custom Yocto Linux build, stripped down to just what's needed. They run a containerized inference engine — TensorRT for the vision models, and a lightweight MQTT broker to publish location updates to the local edge server. That edge server aggregates data from fifty forklifts and syncs with the cloud only when inventory changes or at the end of each shift. Luna: So the cloud gets maybe 20 percent of the data it would have gotten otherwise. Lucas: More like 80 percent reduction in data sent to the cloud. That means lower bandwidth costs, lower cloud compute costs, and less latency. They also claimed their inventory updates are now sub-20 milliseconds from the moment a pallet is scanned to the digital twin updating. Luna: Digital twin — that's the key concept here. They're building a real-time mirrored version of the physical warehouse, right? Lucas: Exactly. Each shelf location has a digital representation that knows exactly what SKU, what quantity, and what condition. And because the edge nodes are doing continuous inference, they can detect anomalies — like a pallet that's leaning or a damaged box — and alert a human immediately. Luna: So it's not just counting, it's quality control too. Lucas: Right. The computer vision model is trained to spot crushed corners, torn shrink wrap, or water damage. That's something RFID alone can't do. And since it's running on the forklift, the driver gets a notification on a tablet in the cab within seconds. Luna: What about edge cases — literally — like when a forklift goes into a dead zone or the network drops? Lucas: Good question. The nodes have local storage and queue up events if the LoRaWAN link to the edge server is lost. Once connectivity resumes, they sync in order. The Jetson can buffer a few hours of data, so even a prolonged outage doesn't lose inventory events. And the edge server itself runs on a redundant pair of ruggedized servers in the warehouse office, with battery backup for up to four hours. Luna: So it's resilient. What about the human side — did the warehouse workers push back on being monitored by cameras? Lucas: That was a concern. The company says the cameras are pointed only at pallets and shelves, not at workers. The vision models are trained on objects, not people. And the RFID scans don't track individual workers. They also involved the union early in the design process, which helped. Luna: Smart. So is this approach spreading beyond parcel warehouses? Lucas: Yes. A grocery retailer is piloting a similar system for perishable inventory. They're adding temperature sensors to the edge nodes, so if a pallet of leafy greens sits in a warm aisle too long, the system flags it for immediate transfer to a cold zone. That's a huge deal for reducing food waste. Luna: And I'd imagine pharmaceutical warehouses, where temperature and location tracking are regulatory requirements. Lucas: Exactly. Pfizer and McKesson are both rumored to be testing edge-based cold chain monitoring. The key is that the edge nodes can validate not just location but environmental conditions in real time, with audit trails stored locally. Luna: So the same hardware platform can serve multiple industries with just software changes. Lucas: That's the beauty of edge compute. The Jetson module is a general-purpose accelerator; you just swap the models and sensors. One warehouse might use computer vision for pallet damage, another for expiration date detection. The edge server handles model updates over the air. Luna: What about smaller warehouses? The $2,500 per forklift might be prohibitive for a mom and pop operation. Lucas: There's a lower-cost variant from a startup called ShelfWise that uses a Raspberry Pi Compute Module 4 with a Coral TPU for inference. It's about $600 per node, but you lose some accuracy and can't run as many models simultaneously. Still, for a 50,000-square-foot warehouse with ten forklifts, that's $6,000. The ROI is still there if you have high-value inventory. Luna: So we're seeing a price-performance gradient. And the trend is toward cheaper, more capable hardware every year. Lucas: Exactly. And as 5G private networks become more common in industrial settings, you might even offload some inference to a local MEC server instead of doing it all on the forklift. That could drive costs down further. Luna: So where do you see this going in the next two to three years? Lucas: I think we'll see edge inventory systems become standard in any warehouse that handles high-turnover or high-value goods. The digital twin becomes the source of truth, and the physical warehouse is just the execution arm. Over time, these systems will link up across the supply chain, so a manufacturer can see exactly where their parts are in real time, from the supplier's warehouse to the assembly line. Luna: That level of visibility could transform just-in-time manufacturing. Lucas: Absolutely. And it all starts with that edge node on a forklift, processing data in milliseconds instead of sending it to the cloud. It's a concrete example of how edge computing turns a mundane operational problem into a strategic advantage. Luna: So next time you track a package and see 'in warehouse' for two days, you'll know the edge might be why it finally moves. Lucas: Hopefully. Or at least you'll know what's happening behind the scenes. Thanks for listening.