2020 Fiscal Year Research-status Report
A Study on Generating Accurate and Largescale LIDAR Maps Based on Graph-Slam Technology for Autonomous Vehicles
| Project/Area Number |
20K19893
|
|---|
| Research Institution | Kanazawa University |
|---|
Principal Investigator |
|
|---|
| Project Period (FY) |
2020-04-01 – 2022-03-31
|
| Keywords | Graph Slam / LIDAR Maps / Mapping Systems / Autonomous Vehicles / 2D Intensity Maps / LIDAR SLAM |
|---|
| Outline of Annual Research Achievements |
-The proposed framework has been successfully implemented. The initial results were published in The 2020 IEEE 91st Vehicular Technology Conference (IEEE-VTC2020-Spring). This step allowed to investigate the performance of generating precise maps in different road conditions.
-The detection of revisited areas that might contain mapping data with different global accuracy has been achieved in different road structures. The detection strategy was published in The 23rd IEEE International Conference on Intelligent Transportation Systems (IEEE-ITSC 2020) with granting the second best paper presentation award. This step enabled visually illustrate the relationships between vehicle position and compare the Graph Slam maps with GNSS/INS-RTK system.
-The evaluation method of the map quality in the detected revisited areas has been implemented to detect the ghosting effects in the map images automatically. The method demonstration was published in The 2020 IEEE/SICE International Symposium on System Integration (SII 2020). This technique significantly provided the capability to validate and tune the cost function of the proposed Graph Slam framework to accurately publish the maps in the Absolute Coordinate System.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
We confidently implemented the Graph Slam framework (GS) and the obtained results were very promising (VTC2020). We collected map data in different challenging environments of high buildings and dense trees in Kanazawa and Tokyo. The robust detection of the loop-closure events (ITSC2020) enabled to significantly compensate the relative position errors based on the environmental feature matching. Moreover, it allowed to automatically check the ground truth maps of an expensive GNSS/INS-RTK (GIR) system (~ 100K USD) and compare GS and GIR maps (SII2021). The proposed GS outperformed GIR maps in terms of accuracy and quality because of the unique utilization tactic of the environmental features and the robust design of the GS cost function to maintain a smooth updating of the road context.
|
| Strategy for Future Research Activity |
The next step will be to investigate the capability to combine two maps collected independently using GS. This is a key-step to generate largescale and precise maps regardless to the road structures. Accordingly, the loop-closure detection strategy will be extended to detect the map-combiner events globally between two maps. The GS cost function will be redesigned to acquire the compensations of global-position errors between two maps and publish the resultant map in Absolute Coordinate System. The effects of Coronavirus have restricted us to collect more map data in critical road segments in Tokyo. We will try to recover this shortage as soon as the situation permits. However, we had previously collected map data in Yamate Junction Tunnel allow us to proceed in the research considerably.
|
| Causes of Carryover |
I would like to transfer the unspent support in F-2020 to F-2021. Due to the current situation of COVID-19, many travel plans and conferences have been cancelled/postponed. Hopefully, the situation will be recovered and the activities will be attained as planned.
|
