Classification of Ego Platform Motion for Platform Independent Plug and Play Navigation
Type of DegreeDissertation
Restriction TypeAuburn University Users
MetadataShow full item record
This dissertation presents a method of using these kinematic constraints without requiring this apriori knowledge of the application platform. The method is termed ``joint navigation and classification'' (JNC), and involves determining the platform type online and simultaneously using the platform type information to properly apply the kinematic constraints and improve accuracy. The JNC problem is solved with a form of multiple model particle filter which treats the platform type as a mode state upon which the navigation state vector is conditioned. The particle filter is marginalized using Rao-Blackwellization to improve computational efficiency. Additionally, this dissertation presents a method within the JNC particle filter of autonomously determining whether any constraints should be applied at all. Motivating this is a study which shows that applying constraints when the IMU is of high quality can actually hurt the solution. The final JNC particle filter is robust to this phenomenon. The algorithm is validated using data collected on three different platforms (ground vehicle, pedestrian, and aircraft) and with IMUs of varying quality. It is demonstrated that the JNC particle filter can autonomously determine the correct platform type and use that knowledge to improve the navigation solution. It is further demonstrated that the JNC filter can autonomously detect situations where it is advantageous not to apply the constraints, thereby avoiding the pitfall described above. The JNC particle filter offers a best of both worlds solution which is both flexible and optimized to the platform, in addition to being robust to situations with high quality IMUs. Many navigation systems take advantage of knowledge of the host platform type, such as a ground vehicle, to apply kinematic constraints to the system to improve the navigation solution. It has been well documented that such constraints can help reduce inertial drift, whether in concert with other aiding sensors or as the only aids to an otherwise unaided inertial system. However, using these constraints implies an apriori knowledge of the host platform type during the design phase. This is commonly done and is acceptable for stovepiped solutions for which flexibility is not a concern. However, this does not allow for flexibility in either the design phase or in the field.