Driver prediction model for autonomous vehicle within a virtual testing platform tailored to Malaysia driving scenarios
Virtual simulation is a vital tool for testing autonomous vehicle (AV) systems in hazardous scenarios due to its cost-effectiveness, reproducibility, and safety. The reliability of such simulations depends on the accuracy of vehicle dynamics, environmental models, and, critically, driver models, whi...
| Main Author: | |
|---|---|
| Format: | Thesis (University of Nottingham only) |
| Language: | English |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://eprints.nottingham.ac.uk/81189/ |
| _version_ | 1848801301643657216 |
|---|---|
| author | Cheok, Jun Hong |
| author_facet | Cheok, Jun Hong |
| author_sort | Cheok, Jun Hong |
| building | Nottingham Research Data Repository |
| collection | Online Access |
| description | Virtual simulation is a vital tool for testing autonomous vehicle (AV) systems in hazardous scenarios due to its cost-effectiveness, reproducibility, and safety. The reliability of such simulations depends on the accuracy of vehicle dynamics, environmental models, and, critically, driver models, which must replicate human driving behaviour to ensure valid testing results. This study develops an artificial intelligence-based driver model tailored to the Malaysian driving environment, addressing significant differences in traffic behaviour between developing and developed countries. To achieve this, a non-linear 14 Degrees of Freedom (DOF) vehicle model was developed and validated through comparative analysis with experimental data to ensure accurate replication of vehicle handling characteristics. Real-world driving data were collected over 245 hours using an instrumented vehicle equipped with cost-effective off-the-shelf sensors, covering diverse road networks, including urban, rural, and highway scenarios. Additionally, a mixed-reality driving simulator, integrating IPG CarMaker with a 6-degree-of-freedom motion platform and virtual reality, was employed to capture realistic human driving behaviours. Thirty participants were invited, and their driving styles were classified into aggressive, normal, and slow categories. The model was trained using normal driver data to develop a baseline for human-like driving behaviour. A hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model, incorporating attention mechanisms to capture spatial and temporal dependencies in driving behaviour, was implemented. The model achieved 84.63% accuracy in predicting steering, throttle, and braking inputs under simulated conditions. However, when tested with real-world data, accuracy declined to 67.23%, highlighting a generalization gap due to underrepresented road types, varying time-of-day conditions, and environmental factors such as weather variations. To mitigate this issue, further training was conducted using a combination of real-world and simulation data, improving the model’s adaptability. The proposed driver model was benchmarked against existing deep learning-based driver models, demonstrating superior performance in replicating human-like driving behaviour within the Malaysian driving context. Despite its contributions, the study acknowledges limitations in data collection, including the limited number of participants, relatively short driving durations per driver, and insufficient representation of extreme driving behaviours. These constraints impact the generalizability of the model to all traffic scenarios. Future work should focus on expanding the dataset with more diverse driving conditions and optimizing the model to enhance its robustness in real-world applications. This research advances driver modelling by leveraging deep learning to create a more contextually relevant model for Malaysia, bridging the gap between virtual simulation and real-world driving behaviour. The developed model has significant implications for AV testing, driver training systems, and intelligent transportation applications in developing countries with complex driving environments. |
| first_indexed | 2025-11-14T21:05:17Z |
| format | Thesis (University of Nottingham only) |
| id | nottingham-81189 |
| institution | University of Nottingham Malaysia Campus |
| institution_category | Local University |
| language | English |
| last_indexed | 2025-11-14T21:05:17Z |
| publishDate | 2025 |
| recordtype | eprints |
| repository_type | Digital Repository |
| spelling | nottingham-811892025-07-26T04:40:23Z https://eprints.nottingham.ac.uk/81189/ Driver prediction model for autonomous vehicle within a virtual testing platform tailored to Malaysia driving scenarios Cheok, Jun Hong Virtual simulation is a vital tool for testing autonomous vehicle (AV) systems in hazardous scenarios due to its cost-effectiveness, reproducibility, and safety. The reliability of such simulations depends on the accuracy of vehicle dynamics, environmental models, and, critically, driver models, which must replicate human driving behaviour to ensure valid testing results. This study develops an artificial intelligence-based driver model tailored to the Malaysian driving environment, addressing significant differences in traffic behaviour between developing and developed countries. To achieve this, a non-linear 14 Degrees of Freedom (DOF) vehicle model was developed and validated through comparative analysis with experimental data to ensure accurate replication of vehicle handling characteristics. Real-world driving data were collected over 245 hours using an instrumented vehicle equipped with cost-effective off-the-shelf sensors, covering diverse road networks, including urban, rural, and highway scenarios. Additionally, a mixed-reality driving simulator, integrating IPG CarMaker with a 6-degree-of-freedom motion platform and virtual reality, was employed to capture realistic human driving behaviours. Thirty participants were invited, and their driving styles were classified into aggressive, normal, and slow categories. The model was trained using normal driver data to develop a baseline for human-like driving behaviour. A hybrid Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) model, incorporating attention mechanisms to capture spatial and temporal dependencies in driving behaviour, was implemented. The model achieved 84.63% accuracy in predicting steering, throttle, and braking inputs under simulated conditions. However, when tested with real-world data, accuracy declined to 67.23%, highlighting a generalization gap due to underrepresented road types, varying time-of-day conditions, and environmental factors such as weather variations. To mitigate this issue, further training was conducted using a combination of real-world and simulation data, improving the model’s adaptability. The proposed driver model was benchmarked against existing deep learning-based driver models, demonstrating superior performance in replicating human-like driving behaviour within the Malaysian driving context. Despite its contributions, the study acknowledges limitations in data collection, including the limited number of participants, relatively short driving durations per driver, and insufficient representation of extreme driving behaviours. These constraints impact the generalizability of the model to all traffic scenarios. Future work should focus on expanding the dataset with more diverse driving conditions and optimizing the model to enhance its robustness in real-world applications. This research advances driver modelling by leveraging deep learning to create a more contextually relevant model for Malaysia, bridging the gap between virtual simulation and real-world driving behaviour. The developed model has significant implications for AV testing, driver training systems, and intelligent transportation applications in developing countries with complex driving environments. 2025-07-26 Thesis (University of Nottingham only) NonPeerReviewed application/pdf en cc_by https://eprints.nottingham.ac.uk/81189/1/Cheok_Jun_Hong_PhD_thesis.pdf Cheok, Jun Hong (2025) Driver prediction model for autonomous vehicle within a virtual testing platform tailored to Malaysia driving scenarios. PhD thesis, University of Nottingham. AI‑based driver model; Malaysian driving environment; CNN‑LSTM with attention; mixed‑reality driving simulator; autonomous vehicle testing |
| spellingShingle | AI‑based driver model; Malaysian driving environment; CNN‑LSTM with attention; mixed‑reality driving simulator; autonomous vehicle testing Cheok, Jun Hong Driver prediction model for autonomous vehicle within a virtual testing platform tailored to Malaysia driving scenarios |
| title | Driver prediction model for autonomous vehicle within a virtual testing platform tailored to Malaysia driving scenarios |
| title_full | Driver prediction model for autonomous vehicle within a virtual testing platform tailored to Malaysia driving scenarios |
| title_fullStr | Driver prediction model for autonomous vehicle within a virtual testing platform tailored to Malaysia driving scenarios |
| title_full_unstemmed | Driver prediction model for autonomous vehicle within a virtual testing platform tailored to Malaysia driving scenarios |
| title_short | Driver prediction model for autonomous vehicle within a virtual testing platform tailored to Malaysia driving scenarios |
| title_sort | driver prediction model for autonomous vehicle within a virtual testing platform tailored to malaysia driving scenarios |
| topic | AI‑based driver model; Malaysian driving environment; CNN‑LSTM with attention; mixed‑reality driving simulator; autonomous vehicle testing |
| url | https://eprints.nottingham.ac.uk/81189/ |