Background

According to Global status report on road safety 2023, there are an estimated 1.19 million road traffic deaths in 2021 and Motorcyclists and other powered two- and three-wheeler riders account for 30% of these deaths. So, Powered Two Wheelers (P2W) are involved in a lot of safety events while on road. A lot of research has been dedicated to the detection of these events in an offline or real-time settings to improve the safety strategies. More recently data driven approaches have become popular that can make use of timeseries sensor data collected from these vehicles to develop a predictive model for these safety events. But most machine learning (ML) -based methods require a large amount of data to develop a robust prediction model. On the other hand, collecting P2W riding data in real-world setting or even in controlled experiments can be expensive and time consuming. This becomes even more challenging when we consider the fact that most severe safety events happen rarely, and hence leading to the lack of good enough representation in the dataset to learn a robust prediction model. One promising direction is to exploit the advanced ML techniques to generate synthetic time series data to augment the smaller original dataset.

Aim

In recent years, a lot of research has been done for synthetic data generation techniques in the computer vision domain and they are found to be well established. However, the equivalent research for timeseries data generation has not yet been so mature. Recently there have been some efforts to use Generative AI techniques to generate timeseries data which is statistically similar to the input data. However, so far there is not much research on how effective these generative methods can be when applied to synthetic time series data collected from sensors (e.g., IMUs) placed on a P2W involved in various safety events under realistic driving conditions.

Main aim of this study is to develop an ML-based method based on Generative AI techniques such as GANs (or other emerging techniques) to generate synthetic timeseries signal data for various kind of sensors that are typically mounted on a P2Ws such as IMUs, magnetic sensors and speed sensors. Other state of the art existing or novel ML techniques may also be explored. Developed model should be able to cover various type of crash scenarios such as frontal crashes to another vehicles, angled crashes, high siders, and low siders etc.

Algorithm should be able to work on the sensor data streams generated from inertial sensors (IMUs) at the very least.

Project will also explore some validation techniques to assess the effectiveness and quality of the generated data. One desirable approach would be to use the synthetic data alone or possibly in combination with the organic data (already collected inhouse) to train an improved safety event detection model in terms of prediction performance. Techniques to incorporate physical domain knowledge in the model will be highly desirable.

Objectives

• Data analysis, and pre-processing
• Problem formulation and developing suitable ML-based (Generative AI) approach for synthetic data generation
• Devising suitable augmentations for the timeseries data
• Using developed model to generate synthetic data for various driving scenarios
• Asses the effectiveness of the generated data and the developed model

Study design

• Literature study on current state of the art
• Timeseries data pre-processing & analysis
• Devise suitable ML approach, train ML methods, and validate their predictions
• Detailed performance analysis under different scenarios and identify limitations

Suitability

• 1-2 Master students with engineering background
• Good programming skills, statistics and ML
• Familiarity with a suitable programming language, preferably Python
• Knowledge in Machine learning & Deep-learning methods
• Experience with the analysis of time-series data generated from sensors is desirable

Application

If you find this opportunity interesting and in line with your profile, do not wait with your application! We will start the recruitment process immediately and the positions could be filled before the final application date, 2024-12-01.

If you have any questions, you are welcome to contact the supervisor:

Jawwad Ahmed, jawwad.ahmed@autoliv.com