State of Play: Computer Vision & Supply Chain
We are often asked by customers & investors to talk through the state of computer vision (CV) and supply chain. The first part of the story is simple -- massive investment in autonomous driving over 10 years drove down the cost curve for sensor tech generally. As widespread deployment for driverless cars/trucks seemed farther and farther away, that tech and the engineers building it began to seep out into the broader economy. Simultaneously, compute costs were falling and AI advances were looming on the horizon, both promising sufficient power to squeeze useful supply chain insights from the data streaming through newly cheap sensors. From there the story becomes more about individual companies making tough choices in uncertainty. Today, supply chain startups using CV have to make trades between HW/Sensor performance, cost, and what demands to put on the customer. In short, every decision impacts the only equation that matters:
Level of Capability
VALUE = ___________________________________________
(Added Cost) x (Added Risk) x (Added Operational Complexity)
Let's talk through where different groups of companies are putting down chips when it comes to CV.
Old School Industrial Equipment Incumbents: These companies have been making scales, cameras, dimensioners, etc.. for decades. The tools tend to perform one task very well, but usually demand that the customer alter their workflow to use it. When these incumbents deploy computer vision, they do so to make an existing process marginally faster, rather than eliminating entire categories of task or driving heretofore impossible optimizations.
Big Infra & Big Promises: Some startups are pushing large-format, high-cost sensor deployments with the promise of high-performance CV on docs / yards / racks, etc... . In other words, they increase the numerator in the equation above, but simultaneously increase the denominator by quite a bit. These startups can struggle as exquisite sensors tend to be sensitive to the punishing operational environments in logistics. In this category, I would also include many of the drone-based "CV for inventory management" firms.
Low/No Infra, Long on AI: Others are making the opposite bet. They want to plug into CCTV cameras and other existing infra at low cost and extract value by running models on the feeds and images. This has a very low denominator (i.e. cost and complexity), but the jury is still out on the numerator. Many of these companies have found that CCTV cameras (even with the AI assist) are not up to the task of capturing the data they desired at the spec required.
Dockware is making a high conviction bet on a middle path. We do not believe docks are the right environment to deploy exquisite hardware, nor do we believe that CCTV cameras will ever be capable enough to solve the biggest issues in supply chain. Instead, we have designed our own high-performance sensors (multiple cameras, LiDAR) capable of capturing granular data at operational tempos, but made every other decision to ensure they are cheap, durable, and out of the way. One "Dockware Vision" system covers many dock doors; it sits 20-30 ft off the ground; and you can install it in under an hour. Better yet, we can deploy many CV models & capabilities through the same system, without meaningfully increasing our underlying costs. In the future, the modular design will incorporate new sensor tech (i.e. thermal, RFID) and overlay that new data on the output of our existing models.