Tightly coupled position determination with LPS, Odometry and IMU

Performance of LPS (Local Positioning System)

Fig. 1 shows the performance of the Local Positioning System (LPS) with a model train. The positions of three anchors and the robot are jointly estimated. The track of the model train is a closed loop, which enables an analysis of the repeatability of the position solution. The enlarged view provides two insights: First, the point cloud at (3.3 m, 1.95 m) refers to the initial static position, and has a standard deviation of a few centimeters only. Second, the multiple parallel lines refer to different rounds of the model train and indicate a consistent position solution.

Fig. 1: Performance analysis of Local Positioning System (LPS) with a model train.

Comparison of LPS/ IMU and GNSS/ IMU RTK Positioning

Fig. 2 includes a comparison between the ANavS tightly coupled LPS/ IMU and the ANavS tightly coupled GNSS/ IMU RTK positioning. The closed-loop track is installed at a roof-top with open-sky conditions, i.e. both satellite signals and anchor signals are received without obstructions. Both systems are coupled with an IMU and provide consistent solutions with an uncertainty of less than 10 cm for most timestamps. The systematic offset between both positioning solutions around the lower left part and also at the rightmost part of the track are LPS errors that occur if the angle between an LPS antenna  plane and the signal path is very small.

Fig. 2: Comparison between ANavS tightly coupled GNSS/ IMU RTK positioning and ANavS tightly coupled LPS/ IMU positioning with model train.

Fusion of LPS, Wheel Odometry and IMU

Fig. 3 shows a comparison between the horizontal position estimates of the ANavS Multi-Sensor Fusion of LPS (Local Positioning System), wheel odometry and inertial sensor (IMU) and the horizontal position estimates of the tachymeter-based reference solution (the “truth”). In principle, both solutions are well aligned for almost all timesstamps. A slight offset of the LPS/Odometry/IMU solution can be observed near the start at (0,0) since the Kalman filter needs some time to converge. The tachymeter solution has occasional gaps due to the lack of a line of sight between tachymeter and robot. Moreover, a temporary reduction of accuracy can  be observed for the LPS/ ODO/ IMU solution in areas where the LPS signals from at least one anchor point were shadowed or blocked, e.g. around (1.5 m, -3.0 m).

Fig. 3: Comparison of horizontal position estimates between the ANavS Multi-Sensor Fusion of Local Positioning System, wheel odometry and inertial sensor with the tachymeter-based reference solution.

Fig. 4 includes a quantitative assessment of the positioning accuracy of the LPS/ INS/ odometry sensor fusion using a tachymeter as reference. The position offset remains below 15 cm for 95 % of the epochs.

Fig. 4: Histogram of horizontal position deviation between ANavS Multi-Sensor Fusion of Local Positioning System, wheel odometry and inertial sensor and tachymeter-based reference.

Conclusion

The autonomous driving of robots requires a precise and reliable positioning. In this paper, we analyzed the sensor fusion of GNSS-RTK, LPS, INS and odometry. The focus was put on LPS, and their integration into the sensor fusion. The paper provided a quantitative performance analysis with real measurements, and showed that centimeter-level positioning accuracy is feasible with ANavS MSRTK Module and LPS systems.