自駕系統大規模情境測試與驗證

工研院機械所開發ITRIMMSL-VT (Verification Toolset) 自駕系統驗證工具，包含臺灣情境資料，可以驗證自駕系統在臺灣運行的安全性。基於硬體設備、成本考量與多次迭代測試經驗，ITRIMMSL-VT使用兩周的時間，完成 1000 萬筆測試案例(Test Case)，並根據測試結果構建自駕系統的操作設計域 (Operational Design Domains, ODDs)，可以識別自駕系統的安全運行條件。為提升ITRIMMSL-VT驗證工具的測試情境覆蓋範圍，我們建置臺灣情境資料庫，涵蓋高密度機車流及汽機車混合交通情境，彌補現有開源資料集的不足。此外，藉由 ISO 34502 安全評估框架，識別安全關鍵情境(Safety Critical Scenarios)，並分析臺灣駕駛行為，生成貼近臺灣道路環境的自駕系統測試情境。我們進一步識別188種 ISO 34502 未涵蓋的安全關鍵情境，提升測試結果對自駕車在臺灣道路環境下運行安全性與測試數據的可靠度。

ITRIMMSL has developed ITRIMMSL-VT (Verification Toolset), an automated driving system verification tool incorporating Taiwan-specific scenario data to assess the safety of automated vehicles operating in Taiwan. Considering hardware constraints, cost factors, and iterative testing experiences, ITRIMMSL-VT completed 10 million test cases within two weeks. Based on the test results, it constructs the Operational Design Domains (ODDs) of automated driving systems, identifying the safe operating conditions. To enhance the coverage of test scenarios in ITRIMMSL-VT, we built a Taiwan-specific scenario database, encompassing high-density motorcycle traffic and mixed vehicle-motorcycle traffic conditions, addressing gaps in existing open-source datasets. Furthermore, leveraging the ISO 34502 safety evaluation framework, we identified Safety Critical Scenarios and analyzed local driving behaviors to generate test scenarios that closely reflect Taiwan’s road environment. Additionally, we identified 188 safety-critical scenarios not covered by ISO 34502, improving the reliability of test data and ensuring the safety of automated vehicles operating in Taiwan.