ARMdevices.net

Umar Ahmad, AI@Edge Evangelist at Advantech, presents a practical overview of how edge AI projects move from concept to production, and why hardware choice is only one part of the equation. The talk focuses on how CPUs, GPUs, and NPUs fit different workloads, and why deployment, software readiness, thermal design, and lifecycle support often decide whether an AI product succeeds in the field. [https://www.advantech.com/](https://www.advantech.com/)

A central theme is the full AI workflow: data collection, transfer learning, model optimization, format conversion, application development, edge deployment, monitoring, and retraining. Advantech positions itself as a partner across that chain, with board support packages, drivers, benchmarking tools, SDK support, and engineering services designed to shorten the path from trained model to production-ready embedded system.

The compute discussion is grounded in real trade-offs. CPUs are described as a strong fit for general-purpose processing, rule-based AI, and lighter sensor or time-series workloads. GPUs remain the preferred option for deep learning, vision, and higher-throughput edge inference, while NPUs target lower-power AI acceleration for industrial automation and embedded vision. The point is not that one architecture wins, but that each one matches a different deployment profile.

One of the more useful parts of the presentation is the warning against treating TOPS as the only metric that matters. Umar Ahmad explains that TOPS mostly reflects raw INT8 compute and can be misleading without context. In real edge AI design, latency, throughput, power efficiency, thermal behavior, memory bandwidth, framework support, operating system compatibility, and development environment maturity are often more relevant than a headline performance number.

Recorded at Embedded World 2026 in Nuremberg, this Advantech presentation also touches on practical platform selection across Intel, NVIDIA, Qualcomm, Rockchip, NXP, and accelerator vendors such as Hailo, along with Ubuntu Pro support on NXP i.MX8 through Canonical. The result is a useful summary of how edge AI moves beyond demos into maintainable, scalable products built for 24/7 operation in industrial and vision-centric environments

source https://www.youtube.com/watch?v=DArqsOHP3Bw

This conversation looks at how Advantech and NXP are turning a new generation of Arm-based edge hardware into deployable industrial platforms rather than just another SoC launch. The focus is the i.MX 95 family, positioned here as the step up in compute, graphics, and edge AI capability that lets Advantech build board-level and system-level products for HMI, vision, automation, and embedded inference. The message is less about raw peak numbers than about a usable stack: silicon, modules, carrier design, software enablement, and long product availability. https://www.advantech.com/

A key theme is scalability across form factors. Advantech describes a range built around NXP processors, from compact OSM modules to SMARC and ready-to-integrate boards, so customers can move from a tiny embedded node to a more feature-rich edge computer without changing ecosystem too much. That matters in industrial design, where display support, graphics, AI acceleration, power budget, thermal envelope, and I/O density all need to be balanced against enclosure size and certification path.

The i.MX 95 part stands out here as the higher-performance option, combining stronger multimedia and graphics capability with on-chip AI processing for edge workloads. Alongside it, the i.MX 93 appears as the smaller and lower-power route, especially relevant for compact SOM designs and cost- or energy-sensitive devices. The discussion around OSM Size-S and related module formats underlines a practical market shift: customers want standardized, highly integrated compute blocks that shorten design cycles while still leaving room for custom carrier boards and application-specific I/O.

Software and compliance are just as central as hardware in this interview. Advantech highlights Ubuntu Pro support on NXP-based platforms, with the appeal of a long maintenance window and a cleaner path toward European Cyber Resilience Act requirements coming into force in 2027. In other words, the value proposition is not only edge AI or graphics, but lifecycle management, security updates, BSP stability, and a more predictable route from prototype to deployed device in regulated industrial environments.

What makes the partnership credible is the industrial framing: longevity, open ecosystem thinking, and time-to-market. NXP’s long availability commitments and Advantech’s module and off-the-shelf product strategy give OEMs flexibility to choose between a finished platform and a scalable embedded design-in path. Filmed at Embedded World 2026 in Nuremberg, the interview captures a familiar reality of the embedded market: the winning products are usually the ones where silicon roadmaps, open software, module standards, and long-term maintenance line up into one coherent platform.

All my Embedded World videos are in this playlist: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

source https://www.youtube.com/watch?v=dcznGypWJVE

Würth Elektronik presents edge AI here as a practical hardware stack rather than just a component catalog. The core story is a compact open-hardware single-board computer built for local inference, where computer-vision and sensor workloads run directly on the device instead of depending on cloud processing. That makes the platform relevant for access control, object detection, robotics, condition monitoring, and industrial IoT systems where latency, bandwidth, privacy, or offline operation matter. https://www.we-online.com/en/components/products/GRINN-GENIOBOARD-EDGE-AI-SBC

At the center of the demo is the Grinn GenioBoard Edge AI SBC, based on a MediaTek platform with an integrated NPU rated at 4 TOPS for machine-learning workloads. In practice, the board is positioned as an out-of-box Linux platform with HDMI and DisplayPort output, USB-C power, Ethernet, dual MIPI-CSI camera inputs, onboard storage, and M.2 expansion for SSDs or AI accelerators. That combination makes it easier to move from prototype to custom carrier or production design with fewer architectural changes at the edge.

The live use cases make the positioning clear. One demo runs face recognition locally for entry control, while the broader pitch extends to museum protection, home automation, and robotics vision. The same idea also applies beyond cameras: pressure, acceleration, temperature, and humidity data can be processed on-premise for event detection, anomaly classification, or predictive-maintenance style inference, which is often more efficient than streaming raw telemetry to remote servers all day long.

What makes the presentation more interesting is that Würth Elektronik is not just showing a compute board, but wrapping it with the company’s own design support around EMC, power architecture, thermal behavior, RF integration, antenna matching, and compliance. Expansion options for LTE, Wi-Fi, and Bluetooth Low Energy turn the board into a bridge between local inference and connected IoT deployment, so a customer can choose when data stays on-device and when selected results move to the cloud. Filmed at Embedded World 2026 in Nuremberg, the demo feels like a reference design for companies that need a starting point for edge AI without building every subsystem from zero.

The bigger message is that edge AI development is shifting from isolated silicon demos to more complete system recipes. Würth Elektronik and Grinn are using the GenioBoard, GenioSOM-700, wireless modules, antennas, and sensor add-ons to show how computer vision, sensor fusion, wireless connectivity, and hardware design services can be combined into a deployable platform. For developers working on machine vision, smart devices, industrial automation, or embedded Linux products, this is less about a flashy demo and more about shortening the path from proof of concept to a certifiable product at scale.

All my Embedded World videos are in this playlist: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

source https://www.youtube.com/watch?v=OFlSQrnsG58

Advantech is showing how medical edge AI is moving from generic compute boxes to compact, purpose-built platforms around NVIDIA Jetson Thor. The core of this demo is the AIMB-294 mini-ITX board, designed for healthcare imaging, diagnostics, and smart device integration, with enough compute for real-time video analysis and sensor-driven inference in a much smaller footprint than older GPU systems. https://www.advantech.com/en-us/products/ec92f1c7-a7bd-4d47-bf13-dd6d159778d0/aimb-294/mod_26dae8fd-dcdc-40aa-ae62-5d84523841a6

What makes the platform interesting is not just raw TOPS, but the stack around it. Advantech positions this hardware with NVIDIA Holoscan support, which matters for low-latency streaming pipelines in medical imaging, endoscopy, surgical visualization, and other environments where camera input, AI inference, and display output need to stay tightly synchronized. In the demo, that shows up as live object detection and tracking, but the wider point is deterministic edge processing for medical workflows rather than cloud-dependent AI.

The hardware story is also practical. Advantech is pairing high compute density with medical-ready system design, including a certified box platform, compact thermal design, and modern I/O. One detail highlighted here is USB-C display integration, where power, touch, USB, and audio can run over a single cable. For medical device manufacturers, that can simplify arm-mounted systems, reduce cable bulk, lower weight, and make cleaning and service access easier in hospital and clinical settings.

A small transcript glitch seems to call the Intel platform “Panel Lake”, but the broader message is clear: Advantech is combining NVIDIA Jetson Thor for high-end AI workloads with Intel-based medical display and device integration options, giving OEMs multiple paths depending on compute, certification, and interface needs. Filmed at Embedded World 2026 in Nuremberg, this is less about a flashy demo than about how compact edge hardware, Holoscan-ready software, and medical integration features are starting to converge into deployable healthcare AI equipment.

All my Embedded World videos are in this playlist: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

source https://www.youtube.com/watch?v=cZO5HI7wH8o

Advantech’s Embedded World interview focuses on how NVIDIA Omniverse is being used as a practical robotics and warehouse digital twin workflow rather than a concept demo. The key idea is that a robot can map a storage environment with cameras and sensors, generate a 3D model of the space, and keep syncing live operational data back to the simulation layer. That makes the virtual model useful for path planning, coordination, and validation in environments where multiple AMRs or mobile robots may need to share space and avoid conflicts. https://www.advantech.com/

What stands out here is the feedback loop between physical and virtual systems. Instead of manually building every 3D scene from scratch, the robot contributes spatial data that feeds the digital twin, while real-world telemetry continues to update the model. In industrial terms, this is where simulation starts to matter: route optimization, obstacle avoidance, testing of robot behavior before deployment, and more reliable orchestration of fleets in logistics or smart factory settings. OpenUSD-based collaboration and Omniverse Enterprise also fit naturally into this kind of workflow, especially when different teams need to work on the same operational model.

On the hardware side, the demo is tied to Advantech edge AI infrastructure rather than a fixed single compute stack. In the interview they point to the AIR-420, an Edge AI HPC platform based on AMD Ryzen Embedded and EPYC Embedded options, designed for GPU-heavy workloads and scalable AI deployment. That matches the broader trend around industrial edge servers that can handle sensor fusion, real-time visualization, AI inference, and digital twin workloads close to the machine layer instead of sending everything to a distant data center.

The wider story is physical AI at the edge: robots perceiving space, generating useful world models, and acting on live data with enough compute nearby to keep latency under control. That is why this Embedded World 2026 demo in Nuremberg is interesting beyond the booth itself. It connects warehouse automation, AMR fleet behavior, 3D scene reconstruction, edge GPU computing, and industrial digital twin software into one readable example of where robotics infrastructure is heading.

All my Embedded World videos are in this playlist: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

source https://www.youtube.com/watch?v=-vke_HMi-pk

Advantech frames the Cyber Resilience Act as a system-level engineering problem rather than a paperwork exercise. In this interview, the focus is on building compliance from the hardware upward: trusted platform hardware, supported operating systems, and an embedded stack that is already aligned with industrial cybersecurity requirements before a customer adds its own software layer. The message is less about a single box to tick and more about shortening the path from embedded computer selection to deployable, regulation-aware products. https://www.advantech.com/

The core idea is pre-certification. Joe describes how selected Advantech platforms are already being validated against IEC 62443-4-2 with third-party involvement, then used as a baseline for a broader internal “pre-certified” process across more hardware lines. That matters because the CRA pushes device makers toward traceability, vulnerability handling, and repeatable evidence, especially for connected industrial and edge systems. Advantech’s current material around CRA and IEC 62443 makes the same point: start with hardened platforms, then reduce downstream certification work for OEM and system-integration teams.

A big technical piece here is software composition and vulnerability visibility. The video points to ONEKEY as the tool Advantech is using to address SBOM generation, CVE monitoring, and the ongoing software side of compliance. That is important because CRA readiness is not only about secure boot or TPM-backed roots of trust; it is also about knowing what is inside the firmware and software supply chain, then monitoring exposure over time. Advantech’s ONEKEY material specifically highlights automated binary analysis, one-click SBOM generation, continuous monitoring, and CI/CD integration, which fits well with the interview’s emphasis on repeatable, scalable workflows rather than one-off audits.

What makes this discussion relevant is that it connects EU regulation, IEC 62443 certification practice, and day-to-day embedded product development in one flow. The promise to customers is practical: save months in certification work, lower validation cost, and reduce risk by starting from pre-qualified industrial hardware, then adding gap analysis for the final application stack. Filmed at Embedded World 2026 in Nuremberg, it shows how industrial computing vendors are moving cybersecurity compliance closer to the board, BIOS, firmware, SBOM, and lifecycle-monitoring level, where CRA preparation becomes an ongoing product-management task rather than a last-minute legal check.

All my Embedded World videos are in this playlist: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

source https://www.youtube.com/watch?v=fdaF7laR4rg

Vantron presents a broad embedded portfolio built around customization rather than a single fixed platform. The interview highlights board-level designs, rugged tablets, compact wireless modules, medical and AI-capable systems, and multiple compute form factors aimed at OEM and system-integration work. A key theme is processor flexibility, with support across Intel, NVIDIA, Qualcomm, TI, MediaTek, NXP and ST, which makes the lineup relevant for industrial HMI, edge computing, embedded vision, IoT gateways and application-specific device design. https://vantrontech.us/

What makes this interesting is the mix of hardware breadth and deployment pragmatism. Vantron is positioning itself as processor-agnostic and form-factor agnostic, covering everything from Q7-style modules and Raspberry Pi-based designs to ruggedized field hardware and higher-performance edge AI platforms that can accommodate GPU acceleration. That matters for teams balancing BOM targets, thermals, power draw, Linux or Android support, and long-term product availability across different embedded roadmaps.

The software discussion shifts the story from hardware selection to fleet operations. Vantron’s management platform is described as a SaaS layer for Linux-based and Android devices, handling firmware and application updates, agent deployment, remote troubleshooting, monitoring and recovery in the field. The real value is at scale: device lockdown, remote viewer capability, debugging workflows similar to ADB tracing, and centralized configuration become much more important when a deployment grows from dozens of units to tens of thousands across retail, healthcare, industrial or other distributed edge environments.

Taken together, this is less about a single demo than about an edge-to-cloud deployment model where hardware choice, remote lifecycle management and service economics are tightly linked. The per-device monthly model reflects how embedded vendors are moving toward recurring platform management rather than one-time hardware shipment alone. Filmed at Embedded World 2026 in Nuremberg, the conversation gives a useful snapshot of where embedded IoT is heading: modular hardware, mixed silicon ecosystems, remote operations, secure update pipelines and practical fleet management for real devices already out in the field.

source https://www.youtube.com/watch?v=xzaNgGa7Ejs

Advantech uses this demo to frame humanoid robotics as a perception and compute problem: the robot needs to fuse multiple camera feeds, depth data, and scene understanding fast enough to act in real time. At the center is the AFE-A702, a Jetson Thor based robotic control system built for high-bandwidth sensor input, AI inference, and robot I/O, so the conversation is less about a single robot and more about the full edge-AI pipeline needed to make one practical. https://www.advantech.com/emt/products/8d5aadd0-1ef5-4704-a9a1-504718fb3b41/afe-a702/mod_13487539-d213-4c8f-a027-4be489e0fe1a

The live view makes that idea concrete. Four GMSL cameras and depth sensing are used to build a machine view of the scene, while segmentation separates people, objects, and other obstacles into semantic classes. That is the core requirement for humanoids, AMRs, and service robots: not just video capture, but low-latency perception, sensor fusion, and AI models that can support navigation, obstacle avoidance, workspace awareness, and interaction in dynamic environments at the edge.

What makes the platform interesting is the surrounding ecosystem rather than the compute figure alone. Advantech is positioning the AFE-A702 inside a broader robotics stack with integrated camera drivers, support for LiDAR, IMU and other sensors, JetPack 7, ROS 2 oriented tooling through Advantech Robotic Suite, and links to Isaac ROS, simulation, and deployment workflows. Filmed at Embedded World 2026 in Nuremberg, the demo fits the current shift toward production-ready robotics platforms that reduce custom integration work and shorten the path from prototype to field.

The technical message is clear: humanoid and mobile robotics now depend on scalable multi-sensor vision more than on isolated controller boards. Advantech’s recent updates around validated GMSL camera integration and Jetson Thor class robotics controllers suggest it wants to cover the whole route from thermal and mechanical design-in to perception software and AI inference. In that context, this demo is really about how a robot can turn synchronized camera and depth streams into a usable understanding of people, obstacles, and free space fast enough to deploy

All my Embedded World videos are in this playlist: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

source https://www.youtube.com/watch?v=IHD9JFthSiY

Advantech is showing how an AMR compute stack comes together around multi-camera perception, edge AI and integration support rather than just raw processor specs. The demo centers on the AFRS-761, a Rockchip RK3588-based controller connected to four GMSL cameras, building a 360-degree view for person detection, obstacle awareness and autonomous navigation in warehouse or factory robots. The point is clear: perception is the front end of robot intelligence, and the computer has to ingest, synchronize and process several camera streams in real time. https://www.advantech.com/

What makes this interesting is the emphasis on practical sensor integration. In the demo, GMSL is presented as a preferred interface for current robotics deployments because it simplifies multi-camera wiring across mobile platforms, while Advantech also supports other camera options including MIPI-CSI. That matters for AMRs, forklifts and mobile service robots where reliability, cabling distance, ruggedness and low-latency video all affect how safely a machine can move through a busy environment.

The broader message is that robotics perception is no longer a single-board story. Advantech positions the compute module as the robot brain, the cameras as the eyes, and the motion stack as the actuators behind wheels, arms or other mechanisms. Alongside the Arm platform, the company also highlights an Intel Core Ultra based AMR controller, showing that robotics developers increasingly want a choice of CPU and AI architectures depending on power budget, software stack and workload mix, from object detection to depth processing and scene understanding.

Software is a big part of the pitch as well. Advantech’s robotics approach combines hardware with integration work, driver support, ROS 2 oriented tooling and partner software for fleet management, navigation and deployment. In this setup, Node Robotics provides the higher-level AMR software visible in the demo, while Advantech focuses on making sensor and compute combinations easier to bring into real projects. That is often the hard part in robotics: not proving a concept once, but making perception pipelines stable enough for deployment.

Filmed at Embedded World 2026 in Nuremberg, this interview gives a useful snapshot of where industrial robotics is heading: closer coupling between cameras and edge compute, more multi-sensor perception at the vehicle level, and more modular ecosystems for AMRs, AGVs and warehouse automation. The small robot on the booth is the simple visual example, but the real topic is scalable perception architecture for robots that need to see people, avoid obstacles and keep moving reliably in dynamic spaces.

All my Embedded World videos are in this playlist: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

source https://www.youtube.com/watch?v=V0vkzAmaiSA

Würth Elektronik gives a broad view of how a passive-component supplier moves up the stack into practical power design. The interview centers on compact DC/DC power modules that integrate the inductor, capacitors and key support circuitry, so engineers can build a regulated supply with minimal external parts and much less layout effort. That makes the story less about single components and more about power architecture, EMI behavior, thermal paths and time-to-design in embedded hardware. https://www.we-online.com/

The most interesting angle is how the portfolio connects discrete magnetics, capacitors, quartz and oscillators with module-level building blocks such as the MagI3C family. In real designs, that means one vendor can cover timing, filtering, isolation, galvanic separation and point-of-load conversion across industrial and compute boards. The video also touches wireless power, where Würth Elektronik’s coil and transformer know-how feeds transmitter and receiver designs similar to split-transformer architectures used in inductive charging, from consumer devices up to higher-power transfer.

Automotive qualification is another key theme. The company highlights parts built for stricter reliability targets, including AEC-Q200 qualified components for harsher electrical and thermal environments. That matters in body electronics, infotainment, motor control and power conversion, where low loss, stable magnetic behavior and controlled EMC can matter as much as raw current rating. The discussion around molded inductors is especially relevant here, because shielded constructions help reduce stray magnetic fields and support cleaner high-efficiency converter layouts.

Seen in the context of Embedded World 2026 in Nuremberg, the demo is really about breadth: passive components, power modules, optoelectronics, LED control, wireless power and application examples with partners such as STMicroelectronics and Analog Devices. The closing focus on efficiency and thermal management is the right one, because embedded systems now span everything from nanoamp energy-harvesting nodes to high-current rails for GaN-based power stages, and both ends of that range depend on better magnetics, lower losses and tighter integration today.

All my Embedded World videos are in this playlist: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

source https://www.youtube.com/watch?v=wBCMA45Vd3I

I used current Advantech material on Ubuntu Pro for Devices, its Canonical collaboration, and its Qualcomm edge AI module lineup to tighten the wording and add relevant technical context around OS support, lifecycle, and edge inference. Advantech says Ubuntu Pro for Devices brings 10 years of LTS support, expanded security maintenance, and management tooling, while its Qualcomm QCS6490-based AOM-2721 platform supports Yocto, Windows, and Ubuntu across edge AI scenarios.

Advantech is showing how software support can be just as important as silicon in modern Arm-based edge systems. One part of the demo focuses on Ubuntu Pro running on NXP i.MX 8M, aimed at developers who want a more complete Linux environment for industrial IoT, gateways, robotics and embedded AI without spending time rebuilding drivers, kernel support and interface validation from scratch. The point is not just that Ubuntu boots on Arm, but that the platform is prepared for deployment with long-term maintenance, security updates and a usable BSP from day one. https://www.advantech.com/

The discussion also highlights why this matters for real products. On embedded platforms, OS readiness, driver coverage, graphics support, wireless connectivity and patch management often decide how quickly a team can move from evaluation to shipping hardware. Here the value proposition is a development-ready stack around NXP and Canonical, where Ubuntu Pro adds 10-year lifecycle support, expanded CVE maintenance and large-scale device management options that fit industrial environments better than a minimal custom Linux image.

The second demo moves to Qualcomm and a more explicitly AI-focused workflow. Advantech shows a small OSM-based edge platform running live object detection with YOLOv8, using the SoC’s heterogeneous compute resources rather than pushing everything onto the CPU. That is the real multi-OS story in this video: Yocto, Ubuntu and Windows support on Arm platforms where CPU, GPU and dedicated AI acceleration can be balanced depending on latency, power budget, camera pipeline and application needs.

What makes the conversation interesting is the practical emphasis on optimization. The interview keeps coming back to a familiar edge AI issue: strong hardware alone does not guarantee good results if the software stack is not tuned to the accelerator, memory bandwidth and available drivers. Advantech positions itself as the layer between silicon vendors and product teams, helping customers understand whether a workload belongs on CPU, GPU or NPU, and what software dependencies come with that decision.

This makes the video less about one benchmark and more about reducing engineering friction in embedded AI. The combination of long-term OS support on NXP, multi-OS enablement on Qualcomm, containerized AI workflows and board-level software integration reflects where many Arm deployments are heading now. Filmed at Embedded World 2026 in Nuremberg, it captures a shift from raw edge AI hardware announcements toward the harder question of how to make these platforms maintainable, secure and actually usable in production.

All my Embedded World videos are in this playlist: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

source https://www.youtube.com/watch?v=FWlD2-CJulU

Octavo Systems focuses on System-in-Package design: taking a microprocessor, DDR memory, power management and key passives, then collapsing them into a molded BGA that removes much of the hardest board-level integration. In this interview, that idea is shown not as an abstract packaging story but as a practical way to reduce layout risk, shrink PCB area, and accelerate bring-up for embedded Linux, edge AI, industrial control and audio products. https://octavosystems.com/

The most memorable demo is a Chaos Audio multi-effects guitar pedal, where the SiP handles the real-time audio DSP while a phone or tablet acts mainly as the control surface over Bluetooth. The point is not just miniaturization; it is deterministic local processing with effectively no audible latency, which is exactly what musicians need when switching tones, stacking effects, or building a digital pedalboard that still feels immediate under the fingers.

From there the discussion moves to Octavo’s newer processor-module direction, including the TI AM62-based OSD62PM and the STM32MP2-based OSD32MP2-PM reference platform. The pitch is very specific: processor plus DDR4 in a package roughly the size of the DRAM footprint, with major savings in area and routing complexity compared with a discrete MPU-and-memory design. Camera interfaces, DSI, LVDS, PCIe, Ethernet and built-in AI capability make these parts relevant for HMI, vision, smart gateways, robotics and compact edge compute gear.

What makes the booth tour useful is the range of deployed examples. Octavo shows SiP designs inside ROS 2 robotics modules, a retail people counter, a programmable smart torque drill for manufacturing, a compact SOM, an AMD Zynq UltraScale+ MPSoC platform for ADAS-style video inference, and an industrial automation controller with RS485, CAN and cloud connectivity. That broad spread makes the technology easier to understand: SiP is not a single market play, but a packaging and productization strategy that fits many embedded workloads.

A recurring theme is that SiP is not only about size. Octavo argues that pre-validating the processor-to-memory subsystem removes non-differentiating engineering work, reduces design spins, and in some cases can even compete on BOM cost when compared with sourcing the processor and DRAM separately. Filmed at Embedded World 2026 in Nuremberg, this is a grounded look at how integration, thermals, Linux-class processing and edge AI are being pushed into much smaller hardware footprints without turning every product into a custom high-risk board design.

All my Embedded World videos are in this playlist: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

source https://www.youtube.com/watch?v=C9RZC3o2RHE

HS Devices presents Atronx as a compact embedded development platform built around a microQ7 computer-on-module and an Octavo System-in-Package approach, aimed at teams that need a ready hardware base for custom products rather than a finished end device. The pitch is clear: shorten hardware bring-up, keep software portable, and give engineers a practical Linux-based platform they can adapt for industrial control, monitoring, and edge-connected systems. https://www.hsdevices.com/

What stands out in the demo is the browser-based terminal and device management layer. Instead of treating the module as a black box, HS Devices shows telemetry, module details, CPU load, memory usage, temperature, and direct command-line access in one web interface. That makes the board feel less like a static eval kit and more like a remotely manageable embedded node, which is useful for prototyping field devices, service access, scripted deployment, and debugging across distributed installations.

The hardware story is about modularity and reuse. Atronx is positioned so developers can keep the same carrier or development board and swap compute modules depending on the target: more multithreaded Linux performance, lower power operation, or stronger real-time behavior. That kind of separation between carrier design and compute module is valuable in embedded product design because it reduces redesign cycles, preserves I/O investment, and lets teams move faster when requirements change late in development.

There is also an interesting small-company angle here. HS Devices is a startup from Niš, Serbia, focused on PCB design, circuit design, and embedded hardware engineering, and this product reflects that mindset: practical board-level integration, standardized software foundations, and attention to communication interfaces. Filmed at Embedded World 2026 in Nuremberg, the interview shows a company trying to turn board design expertise into a flexible embedded platform that engineers can actually build on, not just evaluate.

source https://www.youtube.com/watch?v=SIz9Ln5vz68

RISC-V CEO Andrea Gallo presents here not as a single chip vendor story, but as the governance layer behind an open instruction set architecture that lets semiconductor firms, IP providers, tool vendors and device makers build to the same ISA while keeping their own implementation details proprietary. The interview explains why that distinction matters: RISC-V is an open standard, not an open-source core, so the shared asset is the specification itself and the portability it enables across software stacks, supply chains and long product cycles. https://riscv.org/

A central theme is standardization. Andrea Gallo frames the 2025 milestone of RISC-V International becoming an ISO/IEC JTC 1 PAS submitter as more than a badge: it is a path toward formal international recognition for the ISA, which can matter in procurement, compliance, functional safety and regulated industrial design. That is especially relevant as RISC-V moves further from microcontrollers into application processors, automotive platforms, security architectures and compute infrastructure where interoperability and long-term governance carry real weight.

The conversation also gets into how technical consensus is built without turning the ISA into bloat. New extensions are expected to solve real multi-company problems, not one-off requests, and that discipline is what keeps the architecture coherent while still expanding into vectors, matrix processing and AI-friendly data handling. The software point is critical: vendors may differentiate in silicon, microarchitecture and performance, but developers still want one PyTorch, one TensorFlow backend, one toolchain target and a stable compliance model rather than fragmented ports.

Another useful insight is how RISC-V is organizing itself around both horizontal technologies and industry verticals. Alongside the core technical groups, the ecosystem is pulling in requirements from automotive, safety, data center, space, intelligent edge and robotics so that recommendations can map real workloads to the right ISA profiles, extensions and software expectations. That makes the story less about ideology and more about practical system design: where the standard should stop, where vendors should compete, and how to keep portability from compiler to firmware to OS and AI runtime.

What comes through most clearly is that RISC-V is no longer just a university-origin ISA associated with embedded experimentation. It is becoming a neutral coordination point for global compute development, backed by formal process, public technical review and a growing base of engineers, researchers and students who are treating the architecture as production infrastructure. Filmed at Embedded World 2026 in Nuremberg, this interview captures that transition well: from open ISA theory to the harder work of profiles, extensions, safety, matrix acceleration, ecosystem alignment and real deployment at scale.

source https://www.youtube.com/watch?v=4IoVgheSB2o

Espressif’s booth video is really about how far the ESP32 family has moved beyond basic IoT nodes into embedded vision, motion control, touch UI, wireless audio, and higher-bandwidth connectivity. The main demo centers on an ESP32-P4 robotic arm using on-device computer vision to detect colored blocks and trigger pick-and-place motion, which is a good fit for the P4’s dual-core RISC-V architecture, AI instruction extensions, MIPI camera/display support, hardware pixel processing, and H.264-capable multimedia pipeline. https://www.espressif.com/

What makes the robotic arm section interesting is that it combines local inference with networked control instead of treating edge AI and cloud AI as opposites. In the demo, OpenCV-style vision runs directly on the chip for offline detection, while wireless connectivity is used for function-call style interaction and remote control. That fits Espressif’s broader direction for the P4 platform: richer HMI, camera-based edge computing, and low-cost embedded systems that can still expose modern interfaces and automation logic. The handheld controller also points to ESP-NOW as a practical low-latency device-to-device control layer for responsive robotics and peripherals.

The middle part of the video broadens that story with touch and audio demos rather than staying narrowly focused on robotics. The piano example shows how Espressif is positioning capacitive touch as a stable UI input method for compact devices, while the small talking character demo shifts attention to voice interaction, directional audio capture, and sensor-driven movement. That combination matters because Espressif is increasingly covering the full edge stack: sensing, local processing, audio I/O, display control, and wireless backhaul, all in platforms that stay closer to MCU economics than full application-processor designs.

Another useful part of the booth tour is the segmentation across chips. The ESP32-H4 appears in the BLE audio and touch-control demos, which lines up with its role as a low-power dual-core RISC-V SoC for Bluetooth 5.4 LE, IEEE 802.15.4, LE Audio, PAwR, direction finding, and battery-powered devices with an integrated DC-DC converter. The sensor shuttle concept then shows Espressif’s modular approach to quick prototyping, where IMU, magnetic, environmental, display, lighting, microphone, speaker, and battery functions can be mixed around a compact controller rather than rebuilt for each proof of concept.

Filmed at Embedded World 2026 in Nuremberg, the last stretch of the video gives a glimpse of where Espressif is expanding next: not just low-power 2.4 GHz IoT, but also stronger wireless transport. The ESP32-C5, which reached mass production in 2025, brings dual-band Wi-Fi 6 plus Bluetooth LE and 802.15.4, while the newer ESP32-E22 adds tri-band Wi-Fi 6E as a connectivity co-processor across 2.4, 5, and 6 GHz. Put together, the booth is less about a single hero demo and more about Espressif building a ladder from simple sensors to edge AI vision, LE Audio, robotics, and higher-throughput connected devices.

source https://www.youtube.com/watch?v=21dSHwdn7pQ

Espressif uses this demo to show how far its MCU roadmap has moved beyond classic sensor nodes and simple connectivity. The centerpiece is the ESP32-P4, a dual-core RISC-V MCU aimed at richer HMI, multimedia and lightweight edge vision, paired here with a Riverdi 12.1-inch 1280×800 high-brightness industrial touch display. What stands out is not raw headline performance alone, but the fact that this class of GUI can run in an MCU environment with ESP-IDF and LVGL rather than requiring a heavier application processor. https://www.espressif.com/en/products/socs/esp32-p4

The discussion makes clear that Espressif is positioning the P4 as a serious display and interface device: MIPI support, camera input, vector instructions, pixel-processing acceleration, and a software stack that stays accessible to embedded developers. That creates an interesting middle ground between traditional microcontrollers and Linux-class SoCs. For product teams building control panels, industrial terminals, smart appliances, medical interfaces or compact vision-enabled devices, that balance of cost, power envelope and graphics capability is likely the real point of interest.

Another theme is software portability and ecosystem depth. The demo moves between ESP-IDF, LVGL, Embedded Wizard and Slint, while also touching on Rust support and open-source inference examples. Espressif’s approach remains closely tied to accessible tooling, broad community adoption and low barrier to entry, which is one reason the ESP32 family continues to show up in both commercial products and fast prototyping. The partner angle with Riverdi also matters, because industrial display vendors can turn a reference platform into something closer to a deployable subassembly.

Power management is the other major thread. The ESP32-C6 demo highlights Espressif’s split between high-power and low-power cores, showing how software design affects current draw far more than many teams initially expect. That is especially relevant for battery devices, wireless panels and always-on IoT endpoints. Filmed at Embedded World 2026 in Nuremberg, the booth tour also gives a useful snapshot of how Espressif now spans makers, industrial users and HMI developers rather than sitting in only one of those camps.

The wider portfolio shown at the booth reinforces that trajectory. Alongside P4-based HMI and camera demos, Espressif points to C-series RISC-V parts tailored for different wireless and memory requirements, plus the newly announced ESP32-E22 as a tri-band Wi-Fi 6E connectivity co-processor covering 2.4, 5 and 6 GHz. Put together, the story here is about modular architecture: compute where you need it, radio where you need it, and a path from compact MCU designs to more display-heavy and connected embedded products without abandoning the familiar ESP development model.

source https://www.youtube.com/watch?v=uVr0JxrOTfU

This interview frames N-iX as a broad engineering partner rather than a narrow outsourcing vendor. The key point is its one-stop model: embedded software, hardware design, mechanical engineering, connectivity, cloud, and data work can be combined into a single product-development flow, which matters when companies need faster prototyping, tighter hardware-software integration, and fewer handoffs across suppliers. https://www.n-ix.com/

The embedded team describes the practical side of that model well. Instead of focusing only on firmware, they talk about building real devices end to end, including enclosures, electronics, and product-level design decisions. That makes this less about coding capacity and more about full-cycle embedded engineering, where board design, RTOS or Linux software, wireless connectivity, mechanical constraints, validation, and manufacturability all need to line up.

A useful detail in the conversation is how N-iX uses platforms such as Raspberry Pi and Arduino. These are presented mainly as prototyping tools, but also as fast paths for proof-of-concept work where teams need to validate sensing, control logic, motion, and obstacle avoidance before moving to a more production-oriented architecture. The robotic arm demo fits that pattern: rapid iteration around edge control, object handling, and system behavior, with the prototype acting as a bridge between concept and deployable product.

The mention of Nordic Semiconductor also points to a more specific technical direction: low-power connected devices. That usually means Bluetooth Low Energy, battery-optimized wearables, asset trackers, sensor nodes, and other designs where power budgeting, radio performance, firmware efficiency, and long maintenance cycles matter as much as raw compute. Seen that way, the video is really about how an engineering services company positions itself across the full embedded stack, from early prototype hardware to connected edge and IoT product development. The interview was filmed at Embedded World 2026 in Nuremberg.

source https://www.youtube.com/watch?v=aZFiVUrnBLw

CTRL+N presents itself here as a Serbian engineering startup building both hardware and software around embedded systems, with a clear focus on IoT, RTLS, wearables and AI-enabled digital platforms for industrial use. In this interview, the company frames its value around practical field devices rather than generic demos, showing how sensing, positioning and human-machine interaction can be combined into compact products for real deployments. https://ctrln.tech/

The strongest use case in the video is railway safety. CTRL+N shows a digital signalling and worker-safety concept built on embedded electronics, wireless connectivity and precise location awareness, so field personnel can be tracked relative to infrastructure and hazards. That points to a broader architecture built around RTLS, low-power radios such as Bluetooth Low Energy, edge sensing and alert logic, where worker position, status and alarm conditions can be fed into a supervision layer rather than handled as isolated devices.

The wearable element is especially relevant because it turns the system into something operational at track level. A wrist-worn or body-worn node that can vibrate, flash alarms and report location or basic vital-state data is a practical embedded design problem: power budget, ruggedization, wireless reliability, latency and usability all matter more than consumer-style features. In that sense, the video is less about a gadget and more about occupational safety infrastructure built from embedded hardware, firmware and connected software.

Another interesting detail is the AI-assisted multimeter concept. Instead of treating a measurement tool as a passive instrument, CTRL+N describes a compact tester with a chatbot-style interface that helps technicians investigate rail-track faults and interpret readings locally. That suggests a direction where field diagnostics blend measurement electronics, embedded UI, contextual guidance and AI support, giving junior engineers faster troubleshooting workflows while reducing dependence on constant access to senior staff. The interview was filmed at Embedded World 2026 in Nuremberg, where that mix of rail tech, wearables, RTLS and AI-backed maintenance made CTRL+N stand out as a systems-oriented engineering company rather than a single-product vendor.

source https://www.youtube.com/watch?v=16uPvs8tFE8

Altium positions Octopart Discover as a step beyond classic component lookup, turning Octopart from a parts search engine into a system-level discovery workflow. The core idea in this interview is persistent design intent: engineers can start with requirements, narrow options by context such as power, performance, lifecycle status or sourcing constraints, and carry those decisions through architecture, PCB design and procurement instead of losing that reasoning between tools. https://octopart.com/octopart-discover

What stands out is the shift from part-centric filtering to electronics system design. Rather than only searching for a specific IC or passive, the platform is shown handling reference designs, functional blocks, simulation assets, CAD data, lifecycle flags, alternates and distributor availability in one flow. That makes the tool relevant not just for component engineers but also for embedded software teams, hardware architects, sourcing specialists and manufacturing teams trying to converge earlier on a viable BOM.

The demo also suggests a more interactive reference design workflow. Users can inspect schematics and PCB context, view board layers and 3D geometry, drill into component properties, compare alternates, and preserve technical questions asked to field application engineers around a given design choice. That is important because many embedded projects fail less on raw part search than on handoff friction: why a device was chosen, whether it remains recommended for new design, and what constraints shaped the original decision.

On the Octopart side, the scale still matters. The demo references a component database in the tens of millions, live pricing and stock visibility, distributor and manufacturer normalization, and BOM-level purchasing flows that can move from architecture to preferred sourcing channels with fewer spreadsheet exports. For engineers dealing with second-source strategy, compliance, availability windows, regional supply conditions or cost-down work, that combination of technical metadata and sourcing context is where the platform becomes more than search.

Filmed at Embedded World 2026 in Nuremberg, this conversation is really about how EDA, supply chain data and early system architecture are starting to merge. Octopart Discover is presented not as a closed CAD feature but as an open, cross-ecosystem layer that can connect reference designs, component intelligence, distributor data and downstream implementation tools. If Altium executes on that open workflow, it could make early-stage embedded design more traceable, more procurement-aware and much faster to move from concept to production.

All my Embedded World videos are in this playlist: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

source https://www.youtube.com/watch?v=SL3y-r2sSuM

SiliconAuto is positioning itself as a new automotive semiconductor player focused on the control layer that sits between high-level compute and real-time vehicle behavior. In this interview, the company frames its first XMotiv M3 microcontroller as part of a broader move toward automotive HPC, MCU and high-speed interconnect architectures built for low latency, functional safety and deterministic motion control rather than consumer-style compute alone. https://www.siliconautotech.com/

The technical story is really about partitioning. Instead of forcing a central SoC to absorb every sensor, control and housekeeping task, SiliconAuto argues for distributing work across a safety-oriented MCU and companion devices that handle timing-critical jobs closer to the edge. That matters in ADAS and automated driving, where sensor fusion, bounded latency, power limits and fail-operational behavior all shape the system architecture more than raw TOPS alone.

The XMotiv M3 itself is described as a TSMC 28 nm automotive MCU built around an Arm Cortex-M33 at 160 MHz, with TrustZone, HSM, random-number generation, CAN FD, SPI, I2C, UART and a large GPIO budget. In the demo, it drives an adaptive driving beam reference design with matrix LED control, regional dimming, steering-linked light shaping and welcome-animation features. The interesting angle is not just the headlamp demo, but the attempt to move premium lighting control, reference code and faster integration paths into more mainstream vehicle programs too.

A second thread in the video is SiliconAuto’s work with ZF on an award-winning I/O architecture shown at Embedded World 2026 in Nuremberg. Here the MCU acts as a safety and system-management companion for a chip handling camera and sensor pre-processing, image signal processing, radar-related data paths and AI-assisted detection. The broader implication is a chiplet-friendly automotive compute stack where OEMs can mix performance SoC, AI accelerator and I/O domains with more flexibility, while reducing CPU overhead, DDR traffic and sensor-interface bottlenecks.

The digital-twin demos push that idea further by showing software, AI inference benchmarking and even robotic-arm control before final silicon is available. That early virtual-platform workflow is increasingly relevant for automotive, robotics and drone development, where validation time, toolchain maturity and faster concept-to-production cycles can be just as important as the silicon itself. Overall, the video shows SiliconAuto less as a single-chip launch and more as an attempt to define a modular automotive compute model around safety MCU, sensor pre-processing, ADB lighting, UCIe-era integration and real-time motion.

All my Embedded World videos are in this playlist: https://www.youtube.com/playlist?list=PL7xXqJFxvYvjgUpdNMBkGzEWU6YVxR8Ga

source https://www.youtube.com/watch?v=sxBNICzdrFo