Indicators for Self-driving Algorithm Performance Evaluation
- Delivery
- Available on this site
- Format
- Price
- Non-members (tax incl.)：￥1,100 Members (tax incl.)：￥880
- Publication code
- 20219070
- Paper/Info type
- Other International Conferences
- Pages
- 1-7(Total 7 p)
- Date of publication
- Sep 2021
- Publisher
- JSAE
- Language
- English
Detailed Information
| Author(E) | 1) Chen Chen, 2) Shengbo Eben Li, 3) Andre Wijaya, 4) Qi Sun, 5) Jianhua Jiang |
|---|---|
| Affiliation(E) | 1) Tsinghua University, 2) Tsinghua University, 3) Tsinghua University, 4) Tsinghua University, 5) Tsinghua University |
| Abstract(E) | Machine learning has been the most popular and effective method in the self-driving field, in which reward or loss function design is the key to algorithms training. However, indicators design in the past always focuses on multi-lane scenarios, especially in terms of safety, which cannot satisfy algorithm training needs on mix-traffic and intersection scenes. Besides, there is a lack of a unified standard that can support the judgment of autonomous vehicle performance on urban roads. This paper presents a novel driving safety evaluation model that is compatible with all scenarios. On this basis, fivedimensional indicators are established, and real-time and statistical calculation models for each indicator are provided. This research can, on one hand, help to the development of self-driving technology for general urban traffic environments; on the other hand, support the judgment if an autonomous vehicle is well prepared for widely test or application on open roads. |
