Published back in 2Q 2023, an ABI Insight titled “Developments in AI-Powered Image Processing Look Set to Propel Industrial Machine Vision Applications to New Heights” discussed the increasing applications of Machine Vision (MV) within a warehousing context and provided an outlook for how Artificial Intelligence (AI)-powered MV would propel these use cases further. The rapid development and applications of AI since then have allowed very standard camera systems to become tools for myriad use cases, covering efficiency, quality, safety, digital twins, and asset tracking.
Most recently, Honeywell partnered with Stereolabs—a provider of Three-Dimensional (3D) vision cameras for robotics, automation, and spatial analytics—to develop a tool that can automate the measuring of boxes containing new Stock Keeping Units (SKUs) (e.g., size, weight, and volume) to be entered into a Warehouse Management System (WMS). The solution is a combination of Stereolab’s ZED X series cameras and Honeywell’s SwiftDecoder software. Also, within the last month, Gather AI announced its first dealer network partnership with Burwell Material Handling to help bring its drone-based MV solution to Burwell’s customers for inventory management purposes.
Industry Software Driving Effectiveness
|
IMPACT
|
Taking visual feed and turning it into actionable data is of high value to industrial environments. Full visibility, not just of assets and people, but also environments and workflows allow operators to develop a more holistic understanding of their operations. MV has typically been used for item scanning and basic defect detection, and is mounted at fixed points. However, more recent developments are using 3D cameras and AI for advanced spatial analytics and environment perception, and applying them to mobile points (e.g., Autonomous Mobile Robots (AMRs) and drones) for broader use cases.
The tool being developed through the Honeywell and Stereolabs partnership can remove the need for scales, tape measures, and handheld devices when obtaining the proper material master information of a new SKU and registering it in the WMS. This is a crucial step in enabling all subsequent picking and loading operations, and can often be a bottleneck in warehouses, particularly when workers are bringing new SKUs from multiple loading docks to a central point for measuring. Such requirements are even more important for autonomous material handling systems that require a full set of product details to store and handle them correctly.
KNAPP offers a comparable solution with MultiScan, which uses standard scanning and scales to measure SKUs and connect with the KiSoft Genomix app to log all relevant attributes of the product. Where the two differ is that KNAPP’s MultiScan takes a more modular approach, being able to add accessory packages to the hardware for different sector-specific requirements (e.g., the TEXpress module that compresses textiles to determine precise volumes), while the camera-based systems of Honeywell and Stereolabs remain more minimalist and fully camera based. This is a good representation of how MV is establishing itself as a key technology in the industrial sector. While it’s not a complete replacement for many of the well-established scanning and measuring tools currently being used for material handling, its ability to register complex information based on low-cost hardware acts as a key differentiator.
Any vision system is entirely a product of its enabling software. Honeywell’s SwiftDecoder, a software development kit that allows developers to integrate advanced MV into custom applications for various operating systems and devices through Honeywell’s proven data capture algorithms, is the driving force behind the integration of Stereolabs cameras within a warehouse context. Back in March, Honeywell partnered with Corvus Robotics to allow its autonomous drone solution to perform batch scanning (capturing multiple barcodes at once) for rapid inventory management. While the solution provided by Corvus Robotics established its use case, Honeywell’s industrial software is what is allowing the solution to meet industry requirements both in terms of speed and scale.
Processing Requirements Remain a Concern
|
RECOMMENDATIONS
|
In a recent survey conducted by ABI Research of supply chain professionals assessing technology investment plans within warehousing, over 46% of respondents reported that they are either considering or evaluating a Proof-of-Concept (PoC) for MV solutions, while just under 25% were in the initial rollout phase of the technology. Compared to other warehouse technologies, this shows that the vast majority of organizations remain very much in the early phases of MV deployment and represents significant growth potential over the next 3 years.
Innovation is largely being driven by startups, focusing either on the processing software or supplying the advanced camera systems. Companies like Dori AI, Arvist, and alwaysAI have developed camera-agnostic platforms that enable a multitude of different use cases (e.g., defect inspection, worker safety, traffic management, etc.) based on either existing cameras, typically for safety and efficiency-based use cases, or new camera systems, typically for quality management. A key selling point for these providers is not just their ability to integrate with existing hardware, but also to integrate with an existing WMS or Programmable Logic Control (PLC) system. Such providers are seeing little to no competition from established warehouse productivity solution providers, due in large part to their competitive advantages of being agile with innovating new use cases and having low-friction integration capabilities.
Currently, these startups are focused heavily on large enterprise customers, with scale creating a greater need for MV support, but many small and medium enterprises (SMEs) are now testing MV applications to drive new efficiencies and augment the available workforce. While the hardware is cheap, many are still put off by the high processing costs associated with MV solutions, and don’t always have the necessary infrastructure to manage this at scale. Edge computing will need to be a key consideration for providers to reduce data transfers and processing expenses, but providers must also ensure that use cases remain isolated, and don’t get carried away making a single camera run multiple applications. This will allow processing models to remain lighter and more efficient, minimizing the computational burden without sacrificing accuracy.
