|
Measurement-Level Integration of CPDGPS and LASER for Outdoor Ground Vehicle Navigation
 CPDGPS and Laser-Based SLAM
Carrier-Phase Differential GPS (CPDGPS) enables high-precision positioning, which is critical for many outdoor autonomous ground vehicle (AGV) applications. Carrier-phase ranging information is more precise than code-phase data by several orders of magnitude, but its use requires that an unknown integer bias be estimated. If these integer cycle ambiguities are correctly resolved, centimeter-level accuracy is achievable.
However, robust CPDGPS navigation is restricted to open-sky areas because GPS satellite signals can be significantly attenuated or blocked by buildings, trees, and rugged terrain. In response, we augment GPS with two-dimensional laser-scanner measurements from surrounding obstacles, which are used as landmarks. Laser observations are available when GPS is not, and provide in addition, a means for obstacle detection.
Non-linear laser measurements are processed using an Extended Kalman Filter (EKF) -based Simultaneous Localization and Mapping (SLAM) procedure, which determines the vehicle position using previously unknown features in the environment. In practice, a raw laser scan consists of several hundreds of measurements, which cannot conveniently be fed into the EKF. Hence, two intermediary procedures are implemented: feature extraction aims at selecting the few measurements originating from consistently identifiable landmarks, and data association assigns these extracted observations to the corresponding landmark states in the EKF.
Measurement-Level Integration
An intuitive way to determine the user’s location based on GPS and laser-scanner information is simply to combine the individual positioning outputs of each sensor. In partially obstructed GPS environments such as urban canyons and forest roads, there are often less than four satellites available, which with this position-domain approach are left unused. In contrast, integration at the measurement level makes use of these few satellite signals with clear lines of sight by utilizing additional laser observations.
The range-domain GPS/laser integration architecture is realized using a measurement differencing EKF capable of handling the non-linearity in laser observations as well as time-correlated GPS signals. The unified algorithm simultaneously performs real-time vehicle and landmark positioning and on-the-fly carrier-phase cycle ambiguity estimation. The proposed approach is optimal in that it automatically combines all available information (GPS code and carrier, and also laser measurements) to determine cycle ambiguities. It is particularly well suited for AGVs passing through GPS obstructions, as it enables cycle ambiguities for re-acquired satellites to be immediately updated with vehicle position. In comparison, using off-the-shelf GPS and laser navigation solutions requires feeding back laser position updates into the carrier-phase GPS software.
Performance Analysis and Testing
In this work, performance analyses are structured around two scenarios: first, a ‘forest scenario’ where the vehicle roves across a GPS-unavailable area using tree trunks as landmarks in order to maintain a precise position estimate; second, an ‘urban canyon scenario’ describing the decisive contribution of a few GPS satellites to the integrated system, as compared to a position-domain implementation, which only uses laser measurements to buildings’ edges. Covariance analysis, Monte-Carlo simulations and experimental testing in the streets of Chicago demonstrate that the performance of the range-domain integrated system far exceeds that of a simpler position-domain implementation, in that it not only achieves meter-level precision over extended GPS-obstructed areas, but also improves the robustness of laser-based SLAM.
Direct Simulation of the Forest Scenario
LEFT FIGURE: Performance of the GPS/Laser Measurement-level Integration Algorithm
The AGV roves through a forest-type GPS-obstruction (black circles represent trees). The vehicle starts in a GPS-available area (yellow area), with a small positioning error (represented by a red ellipse). It then enters the transitional GPS-and-laser-available area (green area). As the system collects redundant observations of the landmarks (assumed stationary), and as the geometry between vehicle and trees changes due to the rover’s motion, the landmark position estimate improves steadily (blue ellipses are shrinking). In the middle of the forest, there are no more satellites available, and the vehicle’s position estimation relies on laser measurements only (blue area). Using laser-based SLAM and without a sensor to measure the vehicle’s orientation, the rover’s cross-track deviation drifts with distance. At the exit of the blue area, the cross-track deviation drift is stopped. Both GPS and laser measurements contribute directly to the estimation of carrier-phase cycle ambiguities.
UPPER-RIGHT FIGURE: GPS Satellite Visibility
This azimuth-elevation plot shows that, as the vehicle roves through the forest, satellites that are initially visible (blue circles) are blocked (red circles) by the tree canopy (whose edge is represented by a moving red line).
LOWER-RIGHT FIGURE: Laser Scan Processing
A laser-scanner emits pulsed infrared laser beams that are reflected from surfaces of surrounding objects and returned to the scanner. Hence a full laser-scan is made of hundreds of ranging measurements at regular angular intervals. The blue line represents a 360° laser-scan (in practice, we use two back-to-back 180° lasers) obtained for a vehicle at the center of the plot (location indicated by a triangle). The raw scan (underlying green line) is affected by impulse noise that is easily filtered out. Trees (black circles) are used as landmarks. Their cylindrical trunk’s center location is first extracted (as indicated with a red circle) and then associated (red square) with previously observed states that are estimated in the EKF.
|
