Human-like driver model for emergency collision avoidance using neural network autoregressive with exogenous inputs
One of the most difficult challenges in providing a reliable and safe autonomous collision avoidance maneuver is developing the driver model that provides the planning system with the risk of an impending collision. Over the last few years, researchers have extensively studied several methodologies...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Book Chapter |
| Language: | English English English |
| Published: |
Woodhead Publishing
2024
|
| Subjects: | |
| Online Access: | http://umpir.ump.edu.my/id/eprint/41786/1/Machine%20Intelligence%20in%20Mechanical%20Engineering.pdf http://umpir.ump.edu.my/id/eprint/41786/2/Human-like%20driver%20model%20for%20emergency%20collision%20avoidance_ABST.pdf http://umpir.ump.edu.my/id/eprint/41786/3/Human-like%20driver%20model%20for%20emergency%20collision%20avoidance.pdf http://umpir.ump.edu.my/id/eprint/41786/ https://doi.org/10.1016/B978-0-443-18644-8.00017-4 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | One of the most difficult challenges in providing a reliable and safe autonomous collision avoidance maneuver is developing the driver model that provides the planning system with the risk of an impending collision. Over the last few years, researchers have extensively studied several methodologies to develop such a driver model. The human-like driver model, on the other hand, is a rarely discussed research topic. This work aims to develop a steering maneuver model in emergency collision avoidance that can imitate human emergency intervention. The neural network autoregressive with exogenous inputs (NNARX) is utilized to develop the model and autonomously predict the steering angle response. The work begins by collecting the avoidance maneuver driving data of the expert driver from the automaker company. In the data collection process, a controlled environment target scenario is used to ensure that all drivers encounter real emergencies. To investigate the performance of the developed model, a comparison prediction performance between the developed model and feed-forward neural network (FFNN) is presented. The finding shows that NNARX predicts the steering angle response with a lower prediction error during both training and testing compared to FFNN. |
|---|