Artificial intelligence is breaking free from the cloud giants. During our "AI Leaves the Cloud" conference at the Smile booth at VivaTech, Christophe Brunschweiler (Director of Industrial & Product Systems) explained this revolution. By directly integrating itself into the heart of Edge and Embedded AI, artificial intelligence is pushing back technological boundaries and giving rise to unprecedented use cases.
What is embedded AI?
Embedded AI involves running machine learning models directly on local hardware (NPUs, microcontrollers, SoCs) without relying on the cloud. It offers four major benefits : ultra-low latency, enhanced data security, complete offline autonomy, and reduced bandwidth costs.
Why decentralize? The 4 pillars of Edge AI
The evolution of hardware and software imposes a new paradigm: the Continuum of Intelligence . The execution of algorithms moves from the data center to the network edge.
- Minimal latency: local processing without server round trips for real-time applications.
- Confidentiality & Security: retention of sensitive data (imaging, biometrics) on physical equipment.
- Operational autonomy: maintaining service in the event of network connectivity failure.
- Cost optimization: drastic reduction in bandwidth and cloud storage consumption.
Architecture: Hardware and frameworks under constraints
Deploying AI in constrained environments presents a significant engineering challenge. On the hardware side, the revolution is driven by Neural Processing Units (NPUs) , chips specifically designed for AI matrix computing, offering very high performance with negligible power consumption of just a few watts or milliwatts. On the software side, the challenge lies in compressing and reducing the size of models using suitable frameworks such as TensorFlow Lite or ExecuTorch (to name just a couple), some of which even allow algorithms to run on simple microcontrollers.
Sector-specific use cases for embedded AI
While the theoretical potential of embedded AI is immense, its concrete impact is already being felt in the field: all sectors of activity are affected . From healthcare to heavy industry and even sovereignty, AI at the grassroots level is reinventing critical applications.
- The medical sector: In healthcare, local processing ensures strict adherence to medical confidentiality and patient data privacy. Use cases range from portable medical imaging (pocket-sized ultrasound scanners integrating AI to aid real-time diagnosis) to implantable or wearable medical devices capable of instantly detecting cardiac or respiratory abnormalities without relying on a network connection.
- Industry and Smart Manufacturing: Edge AI is revolutionizing the production chain with predictive maintenance directly integrated into PLCs and real-time visual quality control on the assembly line. Local analysis of vibrations, temperatures, and machine wear makes it possible to anticipate critical breakdowns without overloading the factory network bandwidth.
- Sovereign defense and security: In disconnected, isolated, or electromagnetically jammed theaters of operation, decision-making autonomy is a vital necessity. Embedded AI equips drones, tactical vehicles, and individual equipment for obstacle detection, pattern recognition, and signal analysis in the field, while ensuring strict containment of strategic data.
Lessons learned: Industrial AI benchmark
To illustrate this mechanism, we shared two concrete case studies during the conference:
- Computer Vision: A flagship use case for embedded systems, "Computer Vision" perfectly illustrates the power of AI. One of the major challenges lies in the hardware's ability to process increasingly large images at a higher frame rate, while maintaining exemplary energy efficiency (!). Chips are evolving, but they still need to be properly utilized. This is the core focus of our project, and the results speak for themselves: switching the CPU to the NPU of an iMX8MPlus reduced image processing time by a factor of four, dropping it from 652 ms to just 150 ms!
- Proactive personal assistant: Beyond vision, we have demonstrated the potential of agentic AI directly at the edge. Integrated into a vehicle, this hybrid-architecture assistant possesses offline spatial awareness to react to the environment and ensure use cases ranging from passenger comfort to safety.
The potential of embedded artificial intelligence and edge computing is transforming the landscape of uses by turning passive sensors into truly autonomous decision-making entities.
By harmoniously orchestrating the Cloud, the Edge and local sensors, the future belongs to this perfect symbiosis that is the “Continuum of Intelligence”.
Thank you to everyone who came to chat with us at the Smile stand!
Want to take it further? Contact the Smile teams to discuss it.
