In-situ sensing devices need to be deployed in remote environments for long periods of time, and minimizing their power consumption is vital for maximising both their operational lifetime and coverage.
We are exploring the construction of a versatile multi-sensor device (initially based around the ESP32 chipset) and designing an exceptionally low power consumption model by using an on-device reinforcement learning scheduler that can learn to cooperate with other nearby devices.
Our prototype device setup for learning schedules for biodiversity monitoring does pretty well against a number of fixed schedules; the scheduler captures more than 80% of events at less than 50% of the number of activations of the best-performing fixed schedule. You can read more about this in Terracorder: Sense Long and Prosper.
Cooperative Sensor Networks for Long-Term Biodiversity Monitoring extends the scheduler from one device to a cooperative network of them (which is where the power budget improvs a lot), and Poster: Towards Low-Power Comprehensive Biodiversity Monitoring is an early poster on covering this.
The second question is what tasks can usefully run on hardware this small. Benchmarking Ultra-Low-Power μNPUs benchmarks the ultra-low-power NPUs now turning up in this class of device, and Benchmarking Ultra-Low-Power -NPUs is the shorter GETMOBILE version of that work.
Energy-Aware Deep Learning on Resource-Constrained Hardware looks at energy-aware deep learning on the same constrained hardware. Steps towards an Ecology for the Internet steps back to ask what an ecology for the internet of billions of these devices might look like.
He would have presented a poster at ESA's BioSpace too, had he registered in time (ESA's first BioSpace conference seems a huge success). Woops ;-)