Autonomous Vehicle Positioning with GPS in Urban Canyon Environments By: Youjing Cui and Shuzhi Sam Ge Presented by: Trung Ngo Class: ICS 280 Donald Bren School of Information and Computer Sciences Department of Informatics University of California, Irvine
Motivation GPS signals in urban canyon environments Blocked by high rise buildings Not enough available satellite signals Existing approaches Increase the number of visible satellites Example: GLONASS – Make eight or more satellites available Integrate receivers with sensors Inertial navigation system (INS) Use external references Such as altimeter or a precise clock Find a constrained solution: In some cases, the altitude can be considered constant and assumed to be known
Pseudorange equation At least for satellites are needed to solve the standard pseudo-range equation User position: (x, y, z) N satellites (i = 1.. N) Pseudorange measurement: pi = f(x, y, z) + Br + vi 4 unknowns -> need 4 equations Can be eliminated with DGPS Clock diff. Actual range Known value
New approach Observation: Most pieces roads are straight lines, arcs, or other simple smooth curves Thus, the user position (x, y, z) can be simply modeled as (x, f1(x), f2(x)) where f1 and f2 are known based the road models In this paper, a new constrained solution is provided to solve the problem by approximately modeling the path of vehicle by pieces of curves in the urban canyon environments
New Pseudorange equation pi = f(x, y, z) + Br + vi (n equations) y = f1(x) Z = f2(y) Total we have n+2 equations There are 4 unknown variables, thus with n = 2 we can solve the problem. Two mathematical approaches proposed in the paper for this, including an extended version of Kalman Filtering technique (EKF) Thus, the minimum number of satellites required drops to two
Road intersection problem If user travel in only one road, the problem becomes simpler. However, in real urban environments, the road segments are connected by intersections It is important to know which road the vehicle takes when crossing road intersections Vehicle comes to an intersection
Map Representation Road segments are connected by intersections Road segments can be known based on city maps For each road, the following information need to be stored Road shape (e.g., line, arc …) Positions of intersections on the road Indexes of roads connected at each intersection Additional parameters of the road model Roads segment and intersection representation
Determine next road at intersections In this paper, the interacting multiple model (IMM) algorithm employed to solve the problem. This algorithm has been widely used in multi-target tracking applications. Other statistical techniques commonly used in robot navigation applications can be employed also Neighbor algorithm Tracking-splitting filter Join-likelihood algorithms Markov approaches
Simulations Grey horizontal plane frames represent buildings Building heights ranged from 60-180m Vehicle travel from point A to point I
Results Error (m) 1 1 2 5 3 2 15 3 15 5 Mean error = 0.436m
Conclusion A constrained method proposed Approximately modeling the path of the vehicle in the urban canyon environment as pieces of curves Helps to reduce the minimum number of available satellites reduces to two