By Michel Chabroux
When I talk about the intelligent edge, I like to refer to the electromechanical edge. This is, you might say, the very edge of the edge — the place furthest from giant data centers and closest to the real physical world. Equipment at the electromechanical edge is often controlled by a real-time operating system (RTOS) to give it the precision it needs to interact with the physical world.
The hallmarks of an RTOS at the electromechanical edge have always been predictability, determinism, and safety. But I think it won’t be long before a new characteristic makes an appearance: artificial intelligence (AI) and machine learning (ML). Here I’ll just call that pairing AI.
The heavy lifting of AI is the training of models using large amounts of data and intensive compute resources, and that part is likely to stay in the cloud and in data centers for some time to come. But the practical applications of AI — the use of trained models to infer conclusions from new data — are poised to revolutionize what an RTOS is capable of doing. What kinds of scenarios will we first see AI being used for in RTOS environments?
Consider Some Use Cases
One use case for AI that I expect to see as RTOS applications become mainstream is predictive maintenance. All kinds of data will be collected about how much a piece of equipment has been used and under what conditions, in addition to other data that could have predictive value, such as error rates, sound and vibration levels, and so on. Models will be built and trained to predict when equipment needs to be serviced or replaced, based on all the data collected. And those models will be embedded with an inference engine right in the RTOS, so that the equipment can predict its own need for maintenance.
Another likely use for AI will be in the realm of security. A system that learns to recognize its own normal behavior could be more resilient and able to spot unexpected deviations that might signal security incidents.
Based on conversations with leading robot manufacturers, I see AI being used to improve existing processes, such as motion control, by making use of actual observations rather than only mathematical models. Whether in self-driving cars or automated factories, AI embedded in real-time systems will bring a better ability to understand and adapt to the physical environment that the machines are interacting with.
AI scenarios like these require a significant transformation of the traditional RTOS, in addition to adaptation of data science and AI tools to perform in the more constrained environment that an RTOS inhabits.
Tooling Our RTOS for AI
At Wind River®, we are modernizing VxWorks®, our RTOS, to lay the foundation for embedded AI at the intelligent edge. At its heart, AI requires three elements: data, a model for generating predictions, and an inference engine to apply the model to the data and reach conclusions. Our first focus is on providing the engine, so that customers who have the data and the model can begin building their AI solutions.
We started with support for Python, the much-loved language that simplifies the development of neural networks, along with the NumPy library that turns Python into a data science powerhouse. Now we are adding the Pandas data analysis library as our first AI engine. Next we will add support for TensorFlow Lite, as we expand the tools available for AI solution development.
The data collected locally by the RTOS will need to be sent to the cloud for the compute-intensive operation of training the AI model. VxWorks supports a variety of communication protocols, so you can collect device information and send it to the cloud for use in AI training. Our goal in the future is to provide additional technology to make it even easier for you to ingest more data locally and transmit that data to the cloud, so that you can use feedback loops for continuous improvement.
Check out other ways that VxWorks is laying the foundations for AI on the intelligent edge: www.windriver.com/products/vxworks.