SensiML to open source Analytics Studio for TinyML code for IoT sensor apps

QuickLogic subidiary SensiML (rhymes with “sensible”) recently ventured forth with an ambitious initiative: offering  designers complete open-source access to its automated machine learning engine called Analytics Studio software.

SensiML’s Analytics Studio has previously been offered as a fully managed SaaS cloud service, which will still be available, but the software soon will be available open source. Its tools support what the company calls a broad array of MCUs, AI/ML accelerated SoCs and AI engines. They are designed to help developers build edge AI/ML datasets using flexible tools that aren’t tied to specific vendors, chipsets or inference engines.

With open-source community support, SensiML hopes to extend Analytics Studio to GenAI model development and other capabilities such as object recognition from images and video. Users will have the option to choose the open-source version or SensiML’s fully managed SaaS cloud service approach.

Why open source?  Mainly to build up comprehensive AutoML development tools for IoT edge devices, according to Chris Rogers, CEO of SensiML.

Analytics Studio is designed to bring smart sensing capability to a range of IoT edge devices, including offering predictive maintenance and anomaly detection sensors the ability to recognize and react locally to faults. This means, for example, a smart electric drill might be shut off before stripping a screw being drilled by using AI insights directly on the drill without a connection to the cloud, Rogers said.

In that example, the drill would have a smart clutch, using an accelerometer and gyrometer and the input of the motor voltage, then sensing how much load the motor has.  In an interview with Fierce Electronics, he said such a smart clutch might be integrated into a drill or other device for about the cost of a mechanical clutch using an 8-bit or 16-bit microcontroller.

“By open-sourcing Analytics Studio, we are are hoping to take all the GenAI and edge learning we hear from the academic space and get that set of smart people to put it into implementation,” he said. “For us, it was an easy decision if we want to see real innovation. We have to give something to get something. Hopefully we can get people excited about things that need to be done.”

In the overall AI realm there are already plenty of open source frameworks throught Tensor Flow, PyTorch and others used by experts and data scientists, but for end-to-end automated workflows, “there’s nothing out there,” he added.

SensiML already has customers for its AI tools, including an unusual agriculture customer that uses software and sensitive microphones  to detect when chickens being raised in massive chicken houses are in distress or diseased. Anomaly detection can compare normal sounds with sounds of stress. Recent bird flu scares are a great example of the importance of such an application.

The company said in a statement that it expects to launch a public GitHub repository and AutoML engine document in early summer. Developers can sign up at  At Sensors Converge in Santa Clara, Calif., June 24-26, Rogers expects his above-mentioned smart electric drill concept to be on display at booth 1333.

SensiML’s tagline is “making sensor data sensible,” with a focus on building product-worthy TinyML code for IoT sensor apps.  Rogers said his team of 10 developers is excited for moving to open source. “Anything that takes in sensor data needs to transform the data into some insights such as pattern recognition, which is distilled down to the endpoint. We can’t wait to ship our data to the web and wait for the response,” Rogers said.

Soon after making the open source announcement in mid-May, SensiML got a very favorable response to the idea, quickly receiving 100,000 LinkedIn impressions, he said. “It’s not just us who think it’s a good idea. There’s shared excitement about seeing something become open source and it’s a great sign for us.”

blocks in a puzzle