Part II Project Ideas

The general Adaptive City research theme involves looking at anything related to the digital architecture of urban areas in a future where sensor deployments are extremely dense (IoT) and the infrastructure is autonomous (AI). This also assumes the processing of the incoming data will be in real-time e.g. a decision taken (by the infrastructure) to re-route the autonomous delivery of your pizza will rely upon low-latency processing of incoming data and real-time predictions of issues that may need to be mitigated. Clearly we can currently only nibble away at the various challenges that need to be solved, and a prosaic example of the some current research projects in the group is given here.

Basically the area is target-rich for computing challenges but ultimately there are $trillions at stake.

The "Project ideas" listed below are just that... ideas to help you form your own. If you'd like to discuss ideas you might have, feel free to contact me for a no-commitment informal discussion. Good luck with your Part II Project.

Smart Sensors and Actuators

Generally, can a sensor send data more intelligently than once-every-five-minutes (or invent your own time constant)?

Project idea: choose a sensor type that you think is relevant, program a smarter version, and show how your measurement and data transmission algorithm results in less data transmitted AND lower latency. The main trick is to analyse the data at the edge and send updates when something seems interesting, rather than the prevelant but dump method of sending the data periodically regardless of the information content.

Project idea: PREDICT something is going to happen on the edge, and send a message proactively. Measure the improved timeliness.

Async / real-time data handling

The techniques of keeping the data moving throughout your system are more complex than most people understand, so the vast majority of implemented systems are of the "collect data", "store data", "process stored data" architecture. Academia has woken up to real-time data processing in recent years, adopting the term "Stream Processing", while investment banks have been processing financial data in real-time for many decades. Recent system services such as MQTT (think of that as 'http' for real-time messages) have simplified the exploration of rel-time systems design.

Project idea: write an algorithm designed to process sensor data from one or more sensors as it arrives and implement a system that communicates any derived information in real-time. E.g. you could be predicting the arrival time of a bus, or predict congestion is going to occur in some area of Cambridge.

Data visualisation & interpretation

Human-readable views of the data are important to inform the users of the monitored assets (e.g. the building or the buses) but also to inform the research process. In general spatial views (such as a heatmap) are easier to design for humans to see patterns than temporal views (e.g. the idea that the chance of urban congestion is increasing).

Project idea: design a display that somehow communicates information derived from dense sensors over and above the simple readings from each sensor.

Machine Learning

Urban and in-building sensor systems present many use-cases to which machine learning techniques can be applied. These range from the most straightforward interpretation of images through to more complex temporal analyses and predictions.

Project idea: design a camera-based sensor that summarises a traffic situation and reports on it in a timely fashion.

Project idea: design a software module to monitor sensor data and learn what 'normal' looks like, such that a deviation from normal can be quickly identified. The deviation could be in an individual sensor, in the spatial distribution of readings, or in the temporal pattern of the readings.