Intelligent Transport Systems and GNSS ITSNT 2017 ENAC, Toulouse, France 11/14-17 2017 Nobuaki Kubo (TUMSAT)
Contents ITS applications in Japan How can GNSS contribute to ITS? Current performance of GNSS with other sensors - Test results in the different environments. - Low-cost commercial products are only used
ACC (Adaptive Cruise Control) and Lane Keep Assist Very recently, one of the newspaper company tested the 6 cars in Lane keep assist. Express highway, difficult situation (heavy rain, backlight and lane is difficult to see) Very Good : Company M 99/110, Company V 95/110, Company S 89/110 Good : Company B 77/110 Not Good : 5 th 54, 6 th 18
50 km highway driving (Tokyo-Tsukuba) using only ACC (no accelerator/brake) Tsukuba Handle operation only! Tokyo Even in this crowded intersecting, I realized smooth lane change. During the heavy traffic, ACC was perfect.
Pre-Collision Braking System
V2X (Vehicle to N/V/I/P) V2X 1. Vehicle-to-cellular-Network 2. Vehicle-to-Vehicle 3. Vehicle-to-roadside-Infrastructure 4. Vehicle-to-Pedestrian
Advantages and Disadvantages Dead-reckoning GNSS Scan-matching IMU, Speed etc. GNSS receiver LIDAR etc. Advantages Can be used anywhere Can be used without map, Initial position, velocity and timing Disadvantages Error is accumulated Accuracy depends on the environment Instead of human eyes and accurate Precise 3D map and feature points are required Tunnel
How about GNSS? These services I introduce do not use GNSS. It might be used in V2X because position and velocity etc. of each car are very important. GNSS is used for precise 3D map generation map generation by self-vehicle shooting Location based charging system Recently, we are frequently asked by several manufactures regarding lane-recognition. It does not require cm-level but decimeter level. Maybe, decimeter level absolute position is the key to get GNSS used more in the field of ITS.
Different environments for GNSS Countryside 1. Cm-level 2. Low-cost Urban 1. Decimeter-level 2. Low-cost 3. Multi-sensors Tunnel and indoor 1. Impossible It is important to switch algorithm smoothly depending on the condition
Emerging correction service of GNSS One thing we need to consider... The performance of these global correction services is different from local correction service such as DGNSS and RTK. Generally, local-rtk is the best Performance in terms of accuracy, availability and reliability. We tend to discuss the best performance Investment from 5 companies : 1. Hitachi Zosen Corporation 2. Development Bank of Japan Inc. 3. DENSO Corporation 4. Hitachi Automotive Systems, Ltd. 5. Japan Radio Co., Ltd. MADOCA Multi-GNSS Advanced Demonstration tool for Orbit and Clock Analysis being developed by JAXA based on their technology for estimating satellite orbit and clock corrections
Current performance of GNSS with other sensors
Current situation (low-cost vs. survey grade) Decimeter-level accuracy is expected. It s time to use correction data even with consumer GNSS receiver. Multi-GNSS improves accuracy and availability. Performance improvement of low-cost receiver is remarkable. Target : maximum horizontal error is < 1.0 m (100 %) except for tunnel. >10m 5m 1m 10cm 1cm Target Accuracy #2 Product ($10-100) #1 Product ($10,000-100,000) 12
Test Results Low-cost receiver in normal urban areas Low-cost receiver with sensors in dense urban areas Low-cost receiver with sensors in metropolitan expressway Motivation 1. Where is the limitation of GNSS? 2. What is the key technique to improve the performance? Prerequisite 1. Base station is within about 10-20 km for all tests 2. GNSS Receiver is u-blox M8T 3. IMU/Speed sensors are car navigation grade
1. Test Results of GNSS receiver in normal urban areas (2015)
Normal Urban Areas Test (only GNSS) GNSS Observations DGNSS Position Doppler frequency derived velocity Test route Test configuration Tokyo, August 2015 Single frequency GNSS receiver (u-blox M8T) GPS/BEI/QZS (DGNSS) 20 minutes with 5Hz (3 times for same route) Reference positions : POSLV Normal urban areas except for several high-rise buildings Speed Slow or zero Normal Integration using Kalman filter Weighting Position <<< Velocity Position < Velocity Integrated navigation solution Position 15
Pseudo-range based position w/o C/N 0 check Without C/N 0 consideration With C/N 0 consideration We need to reduce the large jumps probably due to NLOS satellite as much as possible before coupling. C/N 0 based satellite selection is effective to some degree.
KF Loosely-coupling pseudo-range based position with velocity (derived from Doppler frequency) We tend to receive the strong multipath during the stopping in urban areas No speed consideration Speed consideration The normal weighting for positioning / velocity is 2m/0.025m/s. Speed consideration means we heavily rely on velocity when the car speed is very slow or zero. It means the weighting for positioning / velocity is 100m/0.025m/s.
Horizontal Errors and Cumulative Percentage No differential correction for NMEA Maximum error % less than 1.5 m Proposed method 1.86 m 99.5 % Receiver s NMEA 5.31 m 0 % (No differential correction) Results of other 2 tests were almost same. 18
New Test Results in Similar Course (u-blox M8T) +RTK-GNSS (GPS/QZS/BeiDou/Galileo) in 2016 Every system combination of this experiment (Non-consideration) Horizontal distribution of this experiment (More than 10 satellites) Route image 1.5km Under elevated road, bridge and high-rise building Minimum number of used satellites 6 7 8 9 10 RTK FIX Rate 73.9% 71.6% 68.3% 61.4% 52.3% Within 30cm in horizontal 97.35% 99.27% 99.57% 99.40% 99.86% Single-frequency RTK performance depends on the number of used satellites 19
Coupling past work with RTK-GNSS Past work GNSS Observables + RTK-GNSS FLOAT solution FIX AR FIX solution Position DGNSS Position integration Kalman filter Doppler frequency derived velocity DGNSS positions are replaced with RTK-GNSS positions when we have valid RTK solution Target Maximum horizontal error < 1.0 m Integrated navigation solution Position 20
Result [Horizontal error in time sequence] Through the truss bridge stop at a red light Sky image Blue Horizontal error[m] 21
Result [Cumulative distribution] Blue Previous method red +RTK-GNSS Max horizontal error % less than 1.0 m Previous method 1.80 m 92.8 % +RTK-GNSS 0.97 m 100 % 22
2. Test Results of GNSS receiver with sensors in dense urban areas (2016)
Overall of our Integration Method (GNSS/IMU/Speed) Accelerometer Vehicle sensor Gyroscope Speed Yaw rate Movement Decision GNSS Velocity Heading Heading filter Velocity filter DGNSS Absolute position Position filter RTK-GNSS Wrong fix Decision Navigation In fact, it is very difficult to obtain the reliable fixed solutions in dense urban areas Near future low-cost dual-frequency GNSS receiver helps a lot!
Meas.(!) σ 2 z6 = z 6 x 2 2 6 σ x6 Innovation Predict 25
How is the accuracy of the direction improved? Raw IMU direction error After considering the zero angular rate update After integration with GNSS velocity vector IMU direction and GNSS velocity direction are checked each other. Outlier detection of GNSS velocity is very important. Speed sensor output is also used to check the Jumps of GNSS velocity. Ave : almost 0 Std : 0.25 degrees 26
Pseudo-range based position with C/N 0 check Business district near Tokyo station Many high-rise buildings 1. Over 100 m errors can be easily seen. 2. Strict C/N 0 check is already performed. 3. Positioning rate of GNSS was about 76 % 4. Maximum interval with no GNSS was 68 s. 27
Final coupling horizontal results using GNSS/IMU/Speed Temporal horizontal errors (E and N) Maximum horizontal error was 2.9 m When we used dual-frequency multi-gnss receiver, the performance was improved a lot because we will have 20-30 % reliable RTK Fixed solutions. Horizontal Errors and Cumulative Percentage 28
3. Test Results of GNSS receiver with sensors in urban expressway (2017) 10-20 seconds outages are included in the case of using IMU/Speed
Test Configuration sec 3 2 1 4 Interval of non-gnss solutions Test route in the center of Tokyo We evaluated the period without long GNSS outages. Low-cost GNSS receiver (u-blox M8T), IMU and speed sensor were used. Base station was within 10 km. Reference precise positions was deduced from POSLV.
Only RTK Results About 27% FIX Rate Horizontal Errors and Cumulative Percentage
Test results of 4 different scenarios Without 10-20 second outages Percentile 90% 95% 99% Code 1.78m 2.38m 6.02m Code+DP 1.03m 1.40m 2.62m Code+DP+IMU+Speed 0.78m 0.98m 1.16m Code+DP+IMU+Speed+RTK 0.40m 0.58m 1.13m Horizontal Errors and Cumulative Percentage DP : Doppler frequency based velocity is used.
Summary Initial or absolute positions are still very important for future ITS applications. We might need to discuss about wide area precise correction service (we tend to show the test results in local area correction service). In normal urban areas, horizontal errors can be mitigated within 1 m. In dense urban areas, horizontal errors can be mitigated within 3 m (near future, 1-2 m possible). Appropriate protection level can be estimated using GNSS/IMU/Speed sensors. It is difficult to estimate using only GNSS.
PRN194 PRN193 PRN199 PRN195
Thank you very much! nkubo@kaiyodai.ac.jp
4. Protection level estimation for GNSS in urban areas (2015) Determines the confidence limits of position errors. It is called protection level. HPL : Horizontal Protection Level compares the protection level with the alert limit if the system is usable or not. HPL<HAL Available HPL HAL HPL>HAL Not available GNSS Sensors RTK Doppler DGNSS Speed IMU Position Velocity Direction It is very difficult to define the confidence limits using only GNSS because it has sharp increase errors. We will see more practical protection level integrating GNSS with other sensors.
POSLV RTK Doppler IMU Actual estimated protection level in dense urban area (GNSS/IMU/Speed)
Reliability and horizontal errors (m) Protection Level and Real Horizontal Errors Temporal horizontal errors Protection level Horizontal errors TOW (s)
Stanford chart (99.9%) Number of Points