Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Contextual Intelligence Using Ultra Low-Power Sensors and ML
The project aims to compare and evaluate different sensing technologies for contextual intelligence that can detect and classify the presence and movement of people at workplace- or room-level. The current implementation, using infrared sensors, shall be compared to novel ultrasound time-of-flight sensors that are still not commercially available. The project is interested in the achieved tradeoff between accuracy, power consumption, and system limitations.
Keywords: Embedded Systems, Low-power Sensing, Signal Processing, Machine Learning, Energy Harvesting, Wireless Communication, PCB Design
We at peaKFlow developed a completely self-sustaining sensor platform for contextual intelligence that can be integrated into HVAC, blinds, and lighting systems for building automation. It can detect and classify the presence and movement at workplace- or room-level whilst still consuming less than 80μW of power. In the current state, we want to explore different sensing technologies that could potentially replace our current implementation using infrared sensors. This could include alternative signal processing techniques with the same sensor or the use of novel ultrasound time-of-flight sensors that are still not commercially available.
Your task in this project will be to compare and evaluate different sensing technologies for contextual intelligence using low-power devices and TinyML. This includes infrared technology as mentioned above and ultrasound time-of-flight sensors for accurate low-power movement and presence detection and could be extended to a full-fledged people counter, flow estimator, or alike. If time allows, the final optimized system shall be finalized with a custom PCB and working prototype. Of interest is the achieved tradeoff between accuracy, power consumption, and system limitations.
**Prerequisites**
- Embedded Systems - MCU programming, sensor interfaces (I2C, SPI etc.)
- Wireless Communication (Bluetooth Low-Energy) is a plus
- Circuit design (Altium, KiCad)
- MATLAB or Python for evaluation and data processing
- Basic knowledge in Machine Learning
**Character**
- 20% Literature study
- 60% Signal processing, data acquisition, hardware design
- 20% Evaluation and validation
We at peaKFlow developed a completely self-sustaining sensor platform for contextual intelligence that can be integrated into HVAC, blinds, and lighting systems for building automation. It can detect and classify the presence and movement at workplace- or room-level whilst still consuming less than 80μW of power. In the current state, we want to explore different sensing technologies that could potentially replace our current implementation using infrared sensors. This could include alternative signal processing techniques with the same sensor or the use of novel ultrasound time-of-flight sensors that are still not commercially available. Your task in this project will be to compare and evaluate different sensing technologies for contextual intelligence using low-power devices and TinyML. This includes infrared technology as mentioned above and ultrasound time-of-flight sensors for accurate low-power movement and presence detection and could be extended to a full-fledged people counter, flow estimator, or alike. If time allows, the final optimized system shall be finalized with a custom PCB and working prototype. Of interest is the achieved tradeoff between accuracy, power consumption, and system limitations.
- Wireless Communication (Bluetooth Low-Energy) is a plus
- Circuit design (Altium, KiCad)
- MATLAB or Python for evaluation and data processing
- Basic knowledge in Machine Learning
**Character**
- 20% Literature study
- 60% Signal processing, data acquisition, hardware design
- 20% Evaluation and validation
**Project Tasks**
- Test different sensors in different environments
- Interpret the data using signal processing
- Create an algorithm for detecting presence/occupancy
- Compare the results with state-of-the-art
- If time allows, create a dedicated PCB
**Project Tasks**
- Test different sensors in different environments
- Interpret the data using signal processing
- Create an algorithm for detecting presence/occupancy
- Compare the results with state-of-the-art
- If time allows, create a dedicated PCB
- Tiago Salzmann (tiago.salzmann@pbl.ee.ethz.ch)
- Yvan Bosshard (byvan@ethz.ch)
- Dr. Michele Magno (michele.magno@pbl.ee.ethz.ch)
- Tiago Salzmann (tiago.salzmann@pbl.ee.ethz.ch)
- Yvan Bosshard (byvan@ethz.ch)
- Dr. Michele Magno (michele.magno@pbl.ee.ethz.ch)
Thesis_proposal_1.pdf
