Course pages 2014–15
Sensor Fusion and Location Aware Computing
This module aims to provide a solid introduction to sensor fusion via the field of navigation and location tracking. this field is particularly appropriate for demonstration because it is a difficult problem that is very topical in industry; there are myriad sensor technologies available; and the application area allows non-abstract demonstration. Both robotic and human positioning are considered.
Lecture allocations are flexible and given for guideline only.
- Absolute positioning techniques. ToF, TDoA, AoA. GNSS systems. Micro-location. [2 lectures]
- Inertial Navigation. Core principle. IMU grades. Sources of error. Calibration techniques. Naïve ZUPTing. [2 lectures]
- Bayesian Filtering and the Kalman Filter. Mathematical formulation. EKF. Example applications. [2 lectures]
- Particle Filtering. Core principles. Motion and sensor models. Resampling schemes. Dynamic particle numbers. Example use in robotics and PDR wall fusion. [2 lectures]
- PDR. Step-and-compass systems. Step detection algorithms. Heading gyro/mag fusion. [1 lecture]
- Signal Fingerprinting. Surveying. Matching metrics. Propagation Models. Regression Models. Example signals. Incorporation into a Bayesian Filter. [1 lecture]
- Simultaneous Localisation and Mapping. Loop closure. DP-SLAM. Graph-SLAM. WiFi-SLAM. [2 lectures]
On completion of this module, students should:
- Be aware of, be confident with, and understand the trade-offs between the key sensor fusion algorithms of today
- Understand and compare the various absolute and relative positioning techniques available
- Be able to implement a complete location system based on fusing multiple sensors
- Understand the constraints and limitations of todays sensor technologies for location
A set of sensor-rich datasets of a user walking around a building will be provided along with an appropriate floorplan.
Students will be required to develop a complete location system for pedestrians that demonstrates as a minimum: PDR, wall sensitivity, and SLAM loop closure. A 4,000 word report on the design, implementation and evaluation of this system will be used for assessment (see below). High marks will be associated with optimising the algorithms, implementing more advanced versions of the algorithms, handling floor changes, etc.
There will be three ticked exercises associated with implementing the core components of a location system. A tick will be awarded on successful demonstration of each component. The components need not work together for the ticks. A total of four one-hour practical sessions will be provided with demonstrators on hand to assist both with the ticks and the subsequent integration.
Assessment will be based on ticked practical exercises (25%) and a written report of a complete system (75%).
Thrun, Burgard and Fox (2005). Probabilistic Robotics.
Groves (2007). Principles of GNSS, Inertial, and Multisensor Integrated Navigation Systems. Artech House (1st ed.).
Titterton (2004). Strapdown Inertial Navigation Technology. IEE (2nd ed.).