The Client



Our client is one of the largest commercial real estate companies in the world. They employ 70,000 people across more than 280 corporate offices, and have revenues of over $6 billion a year while managing more than 4 billion square feet of real estate worldwide.

The Problem

The client wanted to develop a proof of concept which would automate inventory management. This could be utilized in their own offices with thousands of pieces of equipment, or productized as a service to their large customers for their spaces. For example, a hospital has lots of large pieces of portable equipment that are constantly moving around. It pays big returns in both healthcare and dollars to know where each piece of equipment was last seen and if it has gone missing from the facility.


To solve this problem, Codelitt built the brains, or AI, behind a custom autonomous robot, that we affectionately named Schneider. In order to satisfy the very specific customer needs, we had to create our own indoor positioning system (IPS), which is more accurate than anything on the market. Indoor positioning is like GPS, but specifically for indoor applications. GPS can’t reliably be used indoors, because often there is no signal, or just an inaccurate signal. Part of what makes this such a difficult problem to solve is that you are dealing with very small spaces that require very high accuracy. In Google Maps, for example, GPS accuracy of 5 feet is not a problem. If the GPS shows you 5 feet to the south of your actual position, then the map will still show you on the road. However, if you’re indoors, 5 feet could mean the difference between being in an entirely different room on the other side of the wall. To get accurate results for indoor positions, you have to rely on data from sources such as low energy bluetooth, wifi, or RFID. These are technologies that are not specifically designed for location. We utilized a fingerprint, triangulation, and machine learning solution that enabled us to track the indoor position of the robot using these signals with very high accuracy. The more data that is collected from the bot roving, the better the machine learning algorithms work and the higher accuracy we get.

Use Cases

First, we adapted a self-mapping system. A self-mapping system utilizes LIDAR (lasers) and an artificial intelligence that senses walls, furniture, and other fixtures in a space, and then builds a point cloud. Schneider roams around our test space building this point cloud which it then turns into a 2D map of the space. Part of this self-mapping system is object avoidance, even for temporary objects like a chair that has been moved, or a person walking in front of the robot. We then outfitted our inventory with low energy bluetooth devices stuck on them. We built a mobile application for an Android device which has a bluetooth sensor built in and senses the bluetooth fingerprints from these inventory items. The moment the signal is strongest, the device sends a signal to a central server which then communicates with Schneider to retrieve its location on the map. The real position is then translated into pixels on a map and the inventory item appears in it’s current location on a web accessible map. The map allows the end user to be able to monitor current positions, past position, and if an item was not detected during a scan. These navigations can be scheduled, or done manually.



The final deliver of Schneider was an incredible success. Our client was very pleased with the result, and is hoping to roll out a pilot in select locations. The automated inventory management system could provide hundreds of thousands of dollars in savings for a large facility with expensive equipment, as well as savings on the human capital that would be needed to track these items.




You might also like