A Tale of Two AI Implementations in Logistics: Cloud and Edge

Subscribe To Download This Insight

1Q 2020 | IN-5757

NTT and Mitsubishi formed a partnership in late 2019 with the aim of bringing emerging technologies to the Japanese food supply chain. This ABI Insight dives into how Artificial Intelligence (AI) can be implemented across the supply chain, from the cloud to the edge.

Registered users can unlock up to five pieces of premium content each month.

Log in or register to unlock this Insight.


NTT-Mitsubishi Partnership to Shake up Japanese Logistics Industry


In December 2019, NTT and Mitsubishi, two of the largest technology companies in Japan, formed a partnership that focused on the introduction of emerging technologies to the food industry in the areas of source tagging, inventory management, demand prediction, product classification, delivery route optimization, and information security. Underpinning all of these are big data analytics and Machine Learning (ML)-based Artificial Intelligence (AI). NTT projects this collaboration will bring in billions in revenue for the company as logistics players in Japan start to embark on digital transformation.

Not limited to Japan, the pressing issues of aging populations and rising transportation costs have slowly forced traditional logistics players across the world to improve their business processes by turning to emerging technologies such as big data and AI. In order to benefit from AI, logistics players need to understand how AI will be implemented in their processes.

Resource Optimization and Workflow Automation in the Cloud


There are two ways of AI implementation in logistics, namely the cloud and the edge, and the settings and requirements for both scenarios differ greatly.

For companies that are already running cloud-based telematics solutions, the adoption of AI comes via the integration of existing telematics databases in the cloud with ML algorithms that will be used to forecast future demands, optimize transportation routes, or assign job requests to different trucks or riders. Locus, an India-based startup, develops route deviation engines, order dispatch automation, and predictive analytics for major brands like Tata, Myntra, and Eurofins. By pulling in information from Customer Relationship Management (CRM) and Enterprise Resource Planning (ERP) software, logistics players can offer customer service based on conversational AI, which will improve over time, with the frequency of interactions with end customers. Netomi, for example, offers a conversational AI platform that allows businesses to activate, manage, and train AI to resolve tickets, address questions regarding availability and pricing, and provide assistance with complex issues.

Another key AI technology that has taken off in recent years is AI-based Optical Character Recognition (OCR). As more and more operations in logistics are being automated, primarily through Robotics Processing Automation (RPA), automated information retrieval and capturing has become the next frontier. KoiReader is one of the startups that offer OCR for logistics-related documents and videos and works well with key RPA software such as UiPath.

All these integrations are done via Application Programming Interfaces (APIs). However, given the different software language, platforms, and database architecture that the software runs on, it can take a colossal effort just to create that seamless integration. Not all Information Technology (IT) departments in logistics companies are well-equipped to handle this complexity, and this is where Systems Integrators (SIs) play a key role in ensuring all the processes are well in place, with support from vendors.

Low Latency Response and Data Collection at the Edge


When it comes to the edge, the deployment scenario becomes more complex. Autonomous long-haul trucks and last-mile delivery robots have garnered a lot of attention in recent years, as labor cost is the largest cost component for trucking companies and third-party logistics vendors. TuSimple and Starship Technologies are two of the leading examples with commercial deployment. In big warehouses, automatic picking solutions from the likes of Covariant, IAM Robotics, Kindred Systems, and RightHand Robotics are slowly tackling the issue of labor shortage, all of which are trained using ML.

At the same time, there is a hive of activity at the device and sensor levels. AI-based machine vision can be deployed in warehouses for item sorting, classification, and segmentation, as well as on Autonomous Mobile Robots (AMRs) and autonomous forklifts for localization and navigation purposes. Many embedded camera companies, including Basler, SICK, and FLIR Systems, have catered solutions to embedded vision systems in robotics. They are also partnering with Field-Programmable Gate Array (FPGA) and Application-Specific Integrated Circuit (ASIC) vendors, such as Intel and Xilinx, to develop cameras with AI inference capabilities.

Another group of edge AI chipset vendors is targeting edge applications. Companies like STMicroelectronics, Syntiant, XMOS, and Kalray are developing ultra-low-powered AI chipsets for embedded sensors and cameras in heavy machinery and infrastructure in logistics. These chipsets are able to perform singular AI inference tasks, such as computer vision and audio signal processing using quantized or even binary deep learning models. This allows truck and rail service providers to perform device management and predictive maintenance on their assets, or simple pallet counting or ambient control in their storage facilities.

In order for all these chipsets to perform at the highest level, the right data must be collected and fed back to the cloud for the training of AI models, which leads us back to the partnership between NTT and Mitsubishi. The connectivity solution provided by NTT will ensure edge AI is able to provide a low latency response and collect valuable data for enterprises.