Applications must be submitted here: PhD Research Fellowship in Microelectronics / Machine Learning (199990) | University of Oslo (jobbnorge.no)
The Internet of Things (IoT) megatrend is a strong driver for more human friendly interaction between humans and technology. Traditionally, technology is controlled through specific and dedicated actions normally involving physical contact between humans and devices. Recently, more intuitive and natural ways of human-machine interaction based on radar sensor technology have been explored as exemplified by the Google Soli project (https://atap.google.com/soli/ ). Due to range limitations of the specific radar sensor utilized, the Soli project has so far been strongly biased towards hand gestures and close proximity sensing. The goal of this project is to extend the sensing range through use of ultra-wide band (UWB) radar sensor technology and at the same time focus more on utilizing natural body gestures and movements / gait combined with vital signs sensing to enable more intelligent and seamless human / technology interaction, enabling context awareness in the use-cases.
The Micro-/Nanoelectronics group has over the last 15 years focused on UWB radar sensor technology fully implemented in a CMOS chip. Applications ranging from remote presence detection to into-body vital signs detection and environmental analysis have been explored in addition to research on the CMOS radar sensor itself. So far, the algorithms used have been based on deterministic and statistical methods normally best suited for well-controlled and predictable use-cases. Moving to more unpredictable use-case scenarios as natural real-world environments, alternative methods must be explored where machine learning is a natural candidate.
The objective for this specific PhD project is to address challenges related to using UWB radar sensor technology as a method for natural and seamless interaction with humans. The project will be based on using already existing radar sensor modules providing high-quality radar sensor data with high-resolution range and velocity information. The goal for the PhD student will be to develop methods and machine-learning algorithms including HW prototypes solving specific challenges needed to support the use-case requirement. The first fundamental problem is to enable robust classification of one or more humans in the scene, to avoid false positives / negatives. The next step is to analyze the actual behaviour of individuals to understand and predict movements and intentions in order to interact with surrounding technology / environments. The project will also address processing strategies and how to optimize between in-sensor and remote / cloud computation including analyzing trade-offs between dedicated / customized HW and off-the-shelf processors. The project will be part of an ongoing program involving international and industrial collaboration. Some relevant background information is found here:
https://ieeexplore.ieee.org/document/7870299
https://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=8106658
https://dl.acm.org/doi/10.1145/3170427.3186473
Continue with Facebook
