Inside GNSS Media & Research

NOV-DEC 2017

Issue link: https://insidegnss.epubxp.com/i/906409

Contents of this Issue

Navigation

Page 44 of 67

www.insidegnss.com N O V E M B E R / D E C E M B E R 2 0 1 7 Inside GNSS 45 cameras provide relative displacements with respect to a previ- ous time-step, or with respect to a map. us, measurement time-filtering is required, which makes integrity risk evaluation more challenging since past-time sensor errors and undetected faults can now impact current-time safety. Example LiDAR Navigation Safety Evaluation While safety quantification for GNSS and GNSS/INS has been rigorously performed for aviation applications, and is being researched for HAVs, navigation safety for LiDAR, radar, camera, and multi-sensor navigation is a widely unexplored research area. To provide a specific example on the research work that lies ahead, we have started developing safety risk evaluation methods for LiDARs. We selected LiDARs because of their prevalence in HAVs, of their market availability, and because of our prior experience. However, the techniques we are developing are general enough that radar, cameras, or any future sensor that returns range data can be substituted. Raw range data must be processed before it can be used for navigation. One technique, visual odometry, establishes corre- lations between successive scans to estimate sensor changes in pose (i.e., position and orientation). ese processes are highly computationally intensive, and have the same problems as other dead-reckoning techniques, such as wheel odometry over time. us, they can become inaccurate or cumbersome for HAVs moving over multiple time epochs. Although proprietary infor- mation regarding the use of visual odometry by HAV manu- facturers is unavailable, the research literature suggests that it is only used for short time scale operations. A second class of algorithms provides sensor localization by extracting static features from the raw sensor data and associating those features to a map. is is typically done in two steps, as illustrated in Figure 6 : feature extraction (FE) and data association (DA). e resulting information can then be iteratively processed using sequential estimators (e.g., Extended Kalman fil- ter or EKF), which has been readily used in many practical applications. ere are several problems that the FE and DA algorithms are addressing. First, landmarks in the environment are unidentified, and their observa- tions are not tagged in a manner similar to a GNSS satellite signal's Pseudo Random Noise (PRN) number. us, the feature extraction algorithm must isolate the few most consistently identifiable, viewpoint-invariant landmarks in the raw sen- sor data. ese features must be identifiable over repeated observations and distinguishable from one landmark to another. Features that are difficult to distinguish from each other can be found easily, but the possibility that the association is incorrect will greatly negatively impact the integrity risk. Second, range data based on extracted features must match those features with those from a fea- ture database or map. Data association algorithms accomplish this; however, incorrect associations commonly occur. ese can lead to large naviga- tion errors, as illustrated in Figure 6, thereby representing a threat to navigation integrity. FE and DA can be challenging in the presence of sensor uncertainty. is is why many sophisticated algorithms have been devised. But, how can we prove whether these FE and DA methods are safe for life-critical HAV navigation applications, and under what circumstances? ese research questions are currently unanswered. The most relevant publications on DA risk are found in literature on multi-target tracking. For example, in the paper Y. Bar-Shalom and T. E. Fortmann, an innovation-based nearest-neighbor DA criterion is introduced, which serves as basis in many practical implementations. e article by Y. Bar-Shalom, et alia, "e Probabilistic Data FIGURE 5 Example HAV Navigation System. Key Sensors are represented on top. Their measurements are processed to estimate HAV position, velocity and orientation, and then to predict safety risk and send alerts if needed. Odometer GPS INS Laser Feature Extraction Data Association Landmark Map Robot Controller Integrity Monitor Integrity Prediction Integrity Map Alarm Robot Dynamics Pose Estimator FIGURE 6 Impact of Incorrect Association on Vehicle Trajectory Estima- tion. The position-domain integration scheme on the right-hand side experiences a missed association when the measurement- integration scheme on the left-hand side does not. In this case, the left-hand side can be considered truth reference trajectory; because of the missed association, the flawed estimated right-hand side tra- jectory indicates that the vehicle drove into multiple building walls. Measurement-domain Integration –60 –40 –20 East (m) East (m) 0 20 –20 0 20 40 Position-domain Integration North (m) 120 100 80 60 40 20 0 –20 160 140 120 100 60 40 Misassociation estimated landmark location estimated vehicle trajectory reference traj.

Articles in this issue

Links on this page

view archives of Inside GNSS Media & Research - NOV-DEC 2017