The industrial sector has long used robotics platforms for tasks such as material handling, product manufacturing, and stock management. Today, autonomous mobile robots (AMRs), collaborative robots (cobots), and advanced assembly and inspection lines leverage artificial intelligence (AI) to achieve next-level operational autonomy as part of a new field known as “physical AI.”
Historically, many AI workloads have been associated with cloud computing, owing to its abundant resource potential and the traditional power and performance constraints of embedded hardware. As AI-driven tasks have advanced, from optical character recognition (OCR) to automated defect detection, for example, compute requirements have continued to increase.
However, modern processing platforms are enabling many sophisticated AI workloads to run directly on edge hardware. As a result, these technologies are changing how engineers approach industrial robotics design.
The operational advantages of local intelligence
While cloud computing offers ample resources for AI processing, it can introduce issues for industrial robotics:
The first is processing latency, which combines network latency, task scheduling within cloud servers, and in-cloud processing time. High-speed, highly coordinated operations—e.g., robotic product inspection and rejection on assembly lines—rely on strict timing to maximize throughput. Variable latency reduces determinism and therefore operational efficiency, as speeds must be lowered to compensate.
Non-local processing also presents barriers to operational continuity. During network outages, cloud-reliant equipment may experience degraded functionality when processing resources become unavailable. While the resulting downtime can be costly to manufacturers, network outages can also impact safety for staff working alongside cobots and AMRs if key sensory processes are interrupted.
Additionally, network bandwidth creates cost and throughput concerns in data-heavy use cases, such as quality inspection cobots with high-resolution cameras. Moreover, raw data leaving the device for analysis can increase data privacy and security risks.
By shifting to on-device intelligence, robotics system developers can address these non-local processing concerns:
Processing data close to its source typically results in lower latency. Developers can also increase determinism by using hardware-software combinations designed for real-time operation, enabling the high levels of coordination required for assembly line robotics systems.
Similarly, moving all essential processes directly onto robotics platforms helps support network-free operation, reducing the risk of downtime and degraded performance that can affect productivity or robot safety. Edge processing also eliminates the cost and bandwidth constraints associated with cloud-based data analysis while greatly improving data privacy and security, since raw data is not required to leave the device.
For industrial robotics teams, an edge-first approach increases hardware engineering focus on the processing and interfacing requirements of physical AI systems. Development is further complicated by implementing advanced application software and edge AI deployment workflows. Still, on-board intelligence offers a path to greater robotic autonomy.
How cloud infrastructure supports industrial robotics
While edge processing offers many advantages for individual industrial robotics platforms, cloud infrastructure still plays an important role in supporting deployment-scale operations:
Cloud-based fleet management allows teams to coordinate both daily and scheduled operations across many robots. Effectively, this turns discrete systems into a scalable network that operates as one, reducing operational bottlenecks such as AMRs blocking each other’s routes in shared spaces.
The cloud additionally provides a centralized space for supervisory control, observability through device telemetry, and visualization of operational insights; for example, defect detection rates and manufacturing throughput on an assembly line, rather than raw data. Most importantly, cloud platforms centralize essential software lifecycle management tasks such as over-the-air (OTA) updates that improve robot performance and provide remediation against security vulnerabilities.
Combining local intelligence with overarching cloud platforms is a powerful strategy for modern industrial robotics. Nevertheless, this process can be complex, so robotics teams benefit from production-ready solutions that can accelerate hardware and software development within a unified edge-to-cloud framework.
Scalable solutions for building edge-to-cloud robotics deployments
SECO offers a comprehensive ecosystem for building robotics systems for deployment-scale industrial operations. Still, a key challenge for engineers is finding processing platforms that can support advanced edge AI workloads. SECO solves this with its SOM-SMARC-Dragonwing-IQ8, SOM-COMe-CT6-Dragonwing-IQ-X, and SOM-SMARC-QCS6490 system-on-modules (SOMs).
Based on the Qualcomm Dragonwing IQ8, Dragonwing IQ-X, and Dragonwing QCS6490 processors respectively, these highly integrated edge platforms reduce operational dependence on the cloud by allowing demanding AI tasks to be executed locally. This is achieved via heterogeneous processing, featuring:
- Multiple CPU cores for advanced application processing, simultaneous localization and mapping (SLAM), and parallel communications between robotics peripherals such as motor control and sensing subsystems.
- An integrated GPU for image preprocessing and graphics rendering tasks, such as during product inspection or when driving human-machine interfaces (HMIs).
- Dedicated AI acceleration for local execution of person detection, object recognition, sound classification, and other essential tasks for cobot environments.
SECO offers these advanced processors on open-standard computer-on-modules (COMs), streamlining integration and manufacturing at scale while providing industrial interfaces to support robotics applications. Once engineers have built a suitable carrier system, the SOMs’ long lifetime support and interoperability with COMs of the same standard reduce vendor lock-in and minimize redesign efforts if alternative solutions must later be sourced.
Accompanying these hardware platforms, SECO Clea provides a secure, industrial-grade foundational layer for scalable, edge-to-cloud deployments. Clea OS offers a modular, highly customizable software framework based on Yocto Linux, enabling developers to fully utilize heterogeneous processing while allowing unnecessary software components to be omitted, maximizing edge resource potential.
Additional Clea modules help developers build the cloud layer for fleet management, data orchestration, and secure OTA updates that serve long-term operations for deployment-scale industrial robotics. Furthermore, SECO App Hub and Clea Studio AI provide verified AI models and training environments for accelerating AI development.
Advancements in edge AI are changing industrial robotics, but SECO provides the solutions and expertise to help engineers gain the full benefits of edge hardware when building physical AI systems. To learn more, talk to our experts and stay tuned for the next installment of this series exploring the evolving landscape of physical AI and industrial robotics.