Bayesian optimization and deep learning for steering wheel angle prediction.

Area of Science:

• Artificial Intelligence
• Robotics
• Computer Science

Background:

• Automated driving systems (ADS) are rapidly advancing, promising to revolutionize transportation.
• Deep Learning (DL) has shown great potential in developing innovative ADS solutions.
• Optimizing deep neural network architectures and hyperparameters is computationally expensive and time-consuming.

Purpose of the Study:

• To optimize hyperparameters for a Spatiotemporal-Long Short Term Memory (ST-LSTM) network for ADS.
• To develop an accurate model for predicting steering angle in automated driving systems.
• To leverage Bayesian optimization (BO) for efficient hyperparameter tuning.

Main Methods:

• Bayesian optimization (BO) was employed to tune the hyperparameters of an ST-LSTM network.
• The optimized model, termed BO_ST-LSTM, was trained and evaluated on a public dataset.
• Performance was compared against traditional end-to-end driving models.

Main Results:

• The BO-optimized ST-LSTM model (BO_ST-LSTM) achieved superior accuracy in steering angle prediction.
• BO identified an optimal model within a limited number of trials, saving time and resources.
• The proposed method outperformed existing classical end-to-end driving models.

Conclusions:

• Bayesian optimization is an effective strategy for tuning deep learning models in ADS.
• The BO_ST-LSTM model offers a highly accurate and efficient solution for steering angle prediction.
• This research contributes to the advancement of reliable and performant self-driving car technologies.