Validation of the HD map data for autonomous vehicle navigation
The ITEA project XIVT, consisting of partners from Sweden, Germany, Turkey, Canada and Portugal, targets specific challenges of testing highly-configurable variant-intensive industrial systems from the automotive, railway, telecommunication, and industrial production domains.
The automotive industry faces many challenges; one such pain-point is raising the confidence level of High Definition (HD) map data for autonomous vehicle navigation. HD map data generated from LIDAR-like sensors might have gaps and other defects. As a result, a methodology to fully evaluate the HD map data is necessary. This methodology improves map quality (identify and fill gaps) within a short time frame. Current map data validation relies on brute-force (exhaustive) search to select map routes for validation. This approach suffers from a random route selection. In turn, this leads to redundancy (repeated test cases) in the generated test cases. Redundancies have a two-fold effect. First, they increase the cost of testing. Second, they lower map coverage levels when there is a limitation on the available time for validation.
QA Consultants, a Canadian partner in the project, developed a solution for HD map data validation called MapTest. It is the brainchild of QA Consultants and an OEM. MapTest automatically generates the minimum number of test cases required to validate the HD map. It consists of an intelligent route planner that guarantees 100% coverage of HD map features in a limited time and removes redundancy in the testing process. Additionally, it identifies map variants and focuses on new map portions for validation. This feature further lowers testing time and expenses.
MapTest's preliminary results are promising, based on case studies. When implemented, MapTest 's case studies achieved full map feature coverage with zero repeated test cases in the generated test cases. Its execution time for generated test cases was ten times faster compared to current brute-force based approaches. All in all, we are extremely confident with how MapTest will perform in the future.
