Fig. 5

Estimated vehicle trajectories and mean-squared errors on real-world data. Our approach (a) produces trajectories that consistently have smaller positioning errors as compared to trajectory from J-PF (b) and KF-RAIM (c). The horizontal RMSE values are computed by averaging over 20 runs with different randomness values in initialization and propagation. In regions where multiple measurements contaminated with bias errors, our approach is able to localize better than the baselines by assigning high measurement weights to measurements that are consistent with the particle filter probability distribution of vehicle state