Automated Reverse Logistics

By Valerie M. Thomas, Georgia Institute of Technology

By combining computer vision, data from sensors, machine learning techniques, and robotics, reverse logistics can become easier and cheaper. Higher value items can be more easily identified and retrieved and can enter higher value supply chains. Reverse logistics is ripe for disruption.

Automated reverse logistics will use the same technologies that are being used today in retail stores that allow for grocery shopping without stopping at the check-out counter: Amazon Go is an example in the US, among a number of others. Combinations of cameras and other sensors identify items as the customer shops; there is no need for scanning at the check-out counter.

There are multiple reverse logistics applications that can be automated: items placed in any number of places - trash bins, recycling bins, self-storage facilities, industrial facilities, donation centers – could be identified as they are deposited. The resulting inventory can be used for sorting and retrieving valuables and for improved recycling and disposal of waste materials.

Automated product identification in the reverse supply chain can be achieved with several technologies. The decades-old UPC code, used on groceries and other items, provides easy identification if there is line of sight to a scanner. Other visual labels, such as the two-dimensional MaxiCode and QR code, work the same way while providing more information. This labeling technology is ubiquitous and basically free. The two limitations are the need for line-of-sight, and the fact that these codes are often used on the product packaging rather than on the product itself. If the code is on the packaging, it is thrown away immediately after the consumer buys the product. This cuts the cord that easily connects the product to the reverse logistics market.

Today, there are a small number of products for which the product code remains tightly embedded in the product: books and other media for example. Making use of the International Standard Book Number (ISBN), used books were one of the first products to take off on reselling sites including Amazon and Ebay.

A step up from the visual bar code is radio frequency identification (RFID). These work by reflecting a radio signal from the RFID scanner. RFID tags are now quite common, for example in clothing. These can have a long lifetime, and can support automated reverse logistics

Computer vision offers a much expanded opportunity for product identification in reverse logistics. Many types of items currently have little in the way of permanent labels yet could easily be identified visually: furniture, toys, housewares, china, and crystal, linens, building materials, and scrap metal, to name a few.

Computer vision is a type of deep learning. From the appearance of an item, computer vision applications can identify not only what the item is, but also, by drawing on other information available in databases, what can be expected to be inside the item. Many manufactured products have standard parts inside; the internal parts can in effect be identified by a combination of computer vision and extended deep learning techniques.

Truly automated reverse logistics requires robotics to sort and transport the products. The robotic requirements are the same as those already in use in recyclables sorting, package sorting and manufacturing facilities.  While specific applications need to be developed, the technology is already available.

Deep learning, including computer vision, has changed the landscape for reverse logistics. Previously, labels and specialized database were needed to automate reverse logistics, and the failure of labels and databases could derail the logistics systems. Labels and specialized databases are still needed. But deep learning fills in the gaps. Deep learning with computer vision can identify the products that do not have labels, and can create databases using generative AI, with updates also generated as conditions change.  

Reverse logistics still faces challenges of costs and value, as in all industries. The coupling of machine learning with the existing reverse logistics technologies will reduce costs and add value.