Research of Vehicle Behavior Classification of Off-Road Motorcycles Using Gradient Boosting
- Delivery
- Available on this site
- Format
- Price
- Non-members (tax incl.)：￥1,100 Members (tax incl.)：￥880
- Publication code
- 20239517
- Paper/Info type
- SETC
No.2023-01-1817
- Pages
- 1-16(Total 16 p)
- Date of publication
- Oct 2023
- Publisher
- JSAE & SAE
- Language
- English
- Event
- Small Powertrains and Energy Systems Technology Conference 2023
Detailed Information
| Author(E) | 1) Yuki Uto, 2) Hisato Tokunaga, 3) Taichi Inaba, 4) Takayuki Higashi |
|---|---|
| Affiliation(E) | 1) Kawasaki Motors, Ltd., 2) Kawasaki Heavy Industries, Ltd., 3) Kawasaki Heavy Industries, Ltd., 4) Kawasaki Heavy Industries, Ltd. |
| Abstract(E) | Machine learning is used for the research and development of ITS services and the rider assistance for on-road motorcycle racing. Meanwhile, rider assistance systems for off-road motorcycles have yet to be developed, partly due to the complexity of the measurement conditions, as described in the previous paper. This research aims to create a reliable AI which is capable of classifying typical jump behaviors in off-road riding by machine learning to create a rider assistance system for off-road motorcycles. Motorcycle manufacturers and certain research institutes use motion sensors to collect data, but the data is obtained from a limited number of vehicles and riders. The creation of a rider assistance system requires a large amount of validation data. Furthermore, it is desirable to achieve the target with data that can be measured in mass-produced vehicles, which will make it possible to collect data even from general users. In addition, recent machine learning models are black boxes because it is difficult for people to understand the entire process, and it is necessary to evaluate the validity of the results. The approaches are as follows. (1) Using data that can be measured in mass-produced vehicles, the number of features was increased as a preprocessing step. (2) The validity of the machine learning model was evaluated by focusing on the SHAP value, one of the XAI techniques. As a result, (1) classification ability has improved. (2) correspondence between the number of features with large SHAP values and physical phenomena has been obtained. In other words, it has been confirmed that appropriate number of features have been selected for classification. These results have indicated that the created AI has a certain level of classification ability and that the judgment results can be trusted. |
