Retail Robotics Case Study

Retail Robotics

INTRO

The customer is a robotics company that manufactures inventory control robots for use in retail stores. Their AI-based image capture and analysis engine provides real-time, on-shelf data and analytics for merchandising, product location, promotional displays and more for the retail industry.

BACKGROUND

The customer’s robots roam the aisles in a store taking photos of shelves and making suggestions about what items need to be restocked. In order for the robots to make accurate predictions, they must understand what full, empty and low-stock shelves look like. They must also be able to identify individual units of inventory on the shelves.

Because of the high number of variables in the retail industry, when training the system on new products or on edge cases, the system can sometimes misidentify individual units or may combine multiple items into one unit. This will cause the inventory counts on the shelves to be inaccurate and therefore items won't be restocked correctly. These issues create delays in launching the technology in stores, resulting in significantly longer development cycles.

THE SOLUTION ( Quality Control )

To get started, DataPure's team deep dives into the customer's use cases and understands the complexities. Once DataPure is deployed, the team reviews the images that the robots produce and corrects the machine vision when failure occurs. The team works in sprints on small batches of datasets. The edge cases and instructions may change from batch to batch, depending on the items that are being labeled in each bounding box. The customer uses this output to retrain the robots to rapidly increase the accuracy of their predictions. Additionally, DataPure also has a live human-in-the-loop solution that can also correct the AI in real-time. The perfect balance of technology and human augmentation helps the customer go-to-market faster.