TinyAIoT

Energy- and resource-efficient artificial intelligence for modern IoT applications

...

The idea

The rapid growth of the Internet of Things fueled the design of devices that are based on microcontrollers, equipped with sensors, and capable of exchanging data. These devices - used, e.g., in smart home applications or to build environmental monitoring stations - enable the collection and analysis of large amounts of data and the development of potentially powerful applications. However, applications are currently limited by the need to exchange collected data via cloud services to use state-of-the-art AI processes, which consumes significant resources in the form of energy, material, and bandwidth. The aim of the TinyAIoT project is to reduce these resource requirements by developing efficient and tiny AI models that can be used on the microcontrollers themselves. This not only extends the range of possible use cases to more powerful applications, but also reduces the required bandwidth of applications, enabling microcontrollers to operate autonomously for several weeks to years.

Network partners

The project will be carried out as a joint project between the University of Münster and Reedu GmbH & Co. KG (Dr. Thomas Bartoschek).

...

re:edu is a start-up and spin-off from the Institute of Geoinformatics. Since 2018, re:edu is the producer of the senseBox and offers a wide range of services around the senseBox and the fields of Digital Education, Citizen Science and Smart Cities.

...

The University of Münster participates with the Institute for Geoinformatics(Prof. Dr. Angela Schwering) and the Institute for Business Informatics(Prof. Dr. Fabian Gieseke).

Associated partners

In addition, various application scenarios are to be realized together with four associated partners, Stadtwerke Emsdetten GmbH, Stabsstelle Smart City of the City of Münster, Naturschutzzentrum Kreis Coesfeld e.V. and Hof Homann eG. Additionally, subcontracts are to be awarded to two further companies (opensenselabgGmbH and Budelmann Elektronik GmbH).

...

Stadtwerke Emsdetten GmbH

...

Stabstelle Smart City der Stadt Münster

...

Naturschutzzentrum Kreis Coesfeld e.V.

...

Hof Homann eG

...

opensenselabgGmbH

...

HANZA Tech Solutions GmbH

Resource efficiency

The main goal of the TinyAIoT project is to further reduce the resource requirements of existing implementations and to adapt further AI models accordingly. In particular, the resource and energy requirements are to be reduced to such an extent that the underlying microcontrollers can be operated autonomously by means of batteries over a longer period of time. A special focus shall be on the special combination of microcontrollers of the Arduino family and the LoRaWAN network protocol (e.g. very small main memory and limited bandwidth of LoRaWAN). The results will eventually be used to adapt and extend the senseBox and associated sensor networks, leading to a 'smart' version of the senseBox -the TinyAI-senseBox- that can be operated autonomously for longer periods of time Combination

Potential for resource efficiency.

  • Microcontroller: Arduino Pro Mini (5 volts) with a power consumption of 22mA under full load and approx. 3.6mA in sleep mode
  • Number of IoT devices: 10 billion microcontrollers are assumed for the estimate; estimates are for about 75 billion IoT devices in 2025
To estimate the potential energy savings, a specific microcontroller and an estimate of the number of corresponding future IoT devices are used as examples:

This microcontroller has a yearly energy consumption of roughly (22, 1 · 24 ·365mAh·5V )/1000000 ≈ 0,97 kW/h in normal mode and (3, 6·24·365mAh·5V )/1000000 ≈ 0,16 kW/h in sleep mode. Looking at 10 billion devices one can derive an annual energy consumption of 7 TW/h (normal mode) and 1.6 TW/h (sleep mode). Assuming that more efficient implementations of AI techniques could result in a reduction of approximately 15mA per device (e.g., sleep mode, reduced data collection/transmission, etc.), a savings of approximately 657 kW/h per device per year could be achieved, analogous to the calculations above. For 10 billion devices worldwide, this would lead to a savings of approximately 57 TW/h, which corresponds to approximately 1.3% of Germany's net electricity consumption in the year 2020 (488 TWh).

Use cases

The TinyAIoT project was partly inspired by the existing Birdiary project.

Detecting dangerously close takeover manoeuvres with ToF imaging technology

Measuring and sending fill height of public trashcans

Counting people based on step vibrations

Relation to potential negative environmental impacts

The number of IoT applications as well as corresponding AI-based microcontrollers will increase dramatically in the future, both for direct applications to protect our environment and nature as well as for numerous economic applications (with possibly also positive environmental balance). However, increasing digitalization and especially more IoT applications will lead to additional energy consumption. In addition, more efficient implementations could lead to additional smart microcontrollers being used. However, it can be assumed that the implementation of corresponding scenarios will be largely determined by the economic benefits and not by energy consumption. In this respect, more efficient implementations are desirable in any case.

Funding

This project is funded by the Federal Ministry for the Environment, Nature Conservation, Nuclear Safety and Consumer Protection (BMUV) and expires on December 31, 2025.