Cluster Sampling Kalman Filter for Urban Canyon Navigation
Yeongkwon Choe, Jin Woo Song, Chan Gook Park · Proceedings of the Satellite Division's International Technical Meeting (Online)/Proceedings of the Satellite Division's International Technical Meeting (CD-ROM) · 2018
This paper discusses a navigation system using multiple rangefinders and 3D maps for urban canyon environment. In urban canyon environment, it is difficult to use global navigation satellite system (GNSS) due to lots of buildings blocking its signals. To overcome this challenging environment, we utilize the obstructing buildings. By matching the ranges from a vehicle to nearby buildings with 3D maps, it can be realized. The relationship between the measured ranges and the position of the vehicle, that is, the measurement model, is modeled as a discontinuous piecewise linear function because ordinary buildings are a combination of finite-size planes. In general, nonlinear filters will be required to treat this nonlinear measurement model. However, it is not easy to fully implement a nonlinear filter based on numerical methods to the navigation computer of small-size, multi-rotor type unmanned aerial vehicles. Therefore, we proposed a filter scheme similar to the Gaussian particle filter that deals with nonlinear problems using sampling techniques and regards a resultant probability density as a Gaussian distribution. However, approximating the result to a single Gaussian distribution may be inadequate for situations where multimode occurs. Therefore, we employed the expectationmaximization algorithm for clustering and selected the best cluster by Kullback-Leibler divergence. For verifying the performance, Monte-Carlo simulations were performed assuming actual urban environment, and a flight experiment was conducted in an area similar to urban canyon. Simulation results and an experimental result confirm that the proposed method improves the navigation performance compared to a previous method.