Towards Optimal Information Gathering in Unknown Stochastic Environments

A method for achieving lane-level localization in global navigation satellite system (GNSS)-challenged environments is presented. The proposed method uses the pseudoranges drawn from unknown ambient cellular towers as an exclusive aiding source for a vehicle-mounted light detection and ranging (lidar) sensor. The following scenario is considered. A vehicle aiding its lidar with GNSS signals enters an environment where these signals become unusable. The vehicle is equipped with a receiver capable of producing pseudoranges to unknown cellular towers in its environment. These pseudoranges are fused through an extended Kalman filter (EKF) to close-the-loop with the lidar odometry, while estimating the vehicle's own state (three-dimensional position and orientation) simultaneously with the position of the cellular towers and the difference between the receiver's and cellular towers' clock error states (bias and drift). The proposed method is computationally efficient and is demonstrated to achieve lane-level accuracy in different environments. Simulation and experimental results with the proposed method are presented illustrating a close match between the vehicle's true trajectory and that estimated using the cellular-aided lidar odometry over a 1 km trajectory. A 68% reduction in localization error is obtained over the lidar odometry-only approach.