Figures (11)  Tables (3)

    • Figure 1. 

      Overview of elastic multi-source fusion navigation system.

    • Figure 2. 

      Visual reprojection process considering time delay.

    • Figure 3. 

      LiDAR point cloud matching transformation considering time delay.

    • Figure 4. 

      Factor graph optimization framework.

    • Figure 5. 

      Trajectory comparison of different algorithms.

    • Figure 6. 

      ATE variation curve.

    • Figure 7. 

      Real-time calibration of camera and LiDAR extrinsic parameters.

    • Figure 8. 

      Online estimation of camera-IMU and LiDAR-IMU temporal offsets.

    • Figure 9. 

      ATE distribution of different sensor combinations.

    • Figure 10. 

      ATE changes during sensor loss and recovery.

    • Figure 11. 

      Multi-source fusion navigation effect of MAV platform.

    • Input: sliding-window factors, initial states, marginalization prior
      Output: optimized states, calibration parameters, active topology
      1: while a new sliding window is available do
      2: Add parameters and residual blocks of IMU, GNSS, visual, LiDAR factors.
      3: Perform the first LM optimization with N1 iterations.
      4: for each factor bk from GNSS, visual, and LiDAR measurements do
      5: Compute normalized squared residual $ \boldsymbol{\mathit{D}}_k=\boldsymbol{r}_k^T\boldsymbol{\Omega}_k\boldsymbol{r}_k $.
      6: if $ \boldsymbol{\mathit{D}}_k \gt \chi_{0.99,v_k}^2 $ then
      7: Remove bk from the factor graph.
      8: end if
      9: end for
      10: for each modality s $\in $ {visual, LiDAR} and frame i do
      11: Compute $ {\rho }_{s,i}=N_{s,i}^{removed}/N_{s,i}^{all} $.
      12: if ρs,i > 50% then
      13: Remove all factors of modality s in frame i and mark degraded.
      14: end if
      15: end for
      16: Perform the second LM optimization with N2 iterations.
      17: Update states and calibration parameters.
      18: end while

      Table 1. 

      Elastic sliding-window optimization with chi-square factor culling.

    • Test algorithmARE (deg)ATE (m)
      FAST-LIO23.041.90
      VINS-Mono0.673.82
      LE-VINS0.380.61
      VILOG (ours)0.210.18

      Table 1. 

      Comparison data of different algorithms.

    • Increment set
      (ms)      		 Statistics of camera (ms) Statistics of LiDAR (ms)
      Mean Median RMSE Mean Median RMSE
      5 5.08 5.05 0.43 4.79 4.81 0.24
      10 10.05 10.01 0.52 10.00 10.01 0.05
      15 15.04 15.04 0.43 14.90 14.93 0.14
      20 20.07 20.04 0.44 20.08 20.08 0.11

      Table 2. 

      Time offset increment estimation statistics.