Advanced driver assistance systems for high-performance embedded automotive platforms

Abstract

Advanced driver-assistance systems (ADASs) have become a salient feature for safety in modern vehicles. They are also a key underlying technology in emerging autonomous vehicles. State-of-the-art ADASs are primarily vision based, but light detection and ranging (lidar), radio detection and ranging (radar), and other advanced-sensing technologies are also becoming popular. In the first chapter, we present a survey of different hardware and software ADAS technologies and their capabilities and limitations. In the first chapter, we discuss approaches used for vision-based recognition and sensor fusion in ADAS solutions. We also highlight challenges for the next generation of ADASs. Then in the second chapter, we discuss the high-performance embedded platforms using in automotive domain. Since there is a tight relationship between trust of the costumers and comfort in autonomous vehicles with the higher autonomy levels, we focused on the most important issue of the comfort, motion sickness, that impacts on many people. The outcome of our research work in the thesis was two novel methods to mitigate the motion sickness that we discuss in the third and fourth chapters. To extend our work, we decided to use the machine learning techniques for motion prediction. The motion prediction techniques had conducted us to use the traffic rules for having the outperformance in the intersections. Therefore, we developed a state-of-the-art motion prediction system that works in intersections that will be described in the fifth chapter. In the third chapter a Motion Sickness mitigation system is introduced. Current full- and semi- Autonomous car prototypes increasingly feature complex algorithms for lateral and longitudinal control of the vehicle. Unfortunately, in some cases, they might cause fussy and unwanted effects on the human body, such as motion sickness, ultimately harnessing passengers' comfort, and driving experience. Motion sickness is due to conflict between visual and vestibular inputs, and in the worst case might causes loss of control over one’s movements, and reduced ability to anticipate the direction of movement. In the chapter two, we focus on the five main physical characteristics that affect motion sickness, including them in the function cost, to provide quality passengers' experience to vehicle passengers. We implemented our approach in a state-of-the-art Model Predictive Controller, to be used in a real Autonomous Vehicle. Preliminary tests using the Unreal Engine simulator have already shown that our approach is viable and effective, and we implemented and evaluated using Motion Sickness Dose Value and Illness Rating and then tested it in an embedded platform. We have also developed another novel alerting system to minimize the motion sickness describing in the fourth chapter. Current intelligent car prototypes increasingly move to become autonomous where no driver is required. If an automated vehicle has rearward and forward-facing seats and none of the passengers pay attention to the road, they increasingly experience the motion sickness because of the inability of passengers to anticipate the future motion trajectory. In the chapter three, we focus on anticipatory audio and video cues using pleasant sounds and a Human Machine Interface to display and inform the passengers about the upcoming trajectories that may lead to make the passengers sick. To be able to anticipate the next moves, we require an evaluation system of the next 1 kilometer of the road using the map. The road is investigated based on the amount of the turns and the maximum speed allowed that lead to lateral accelerations that is high enough based on Motion Sickness Dose Value to make the passengers sick. The system alerts the passengers through a Human Machine Interface to focus on the road for prevention of the Motion Sickness. We evaluate our method by using Motion Sickness Dose Value. Based on this work, we can prevent the sickness due to lateral accelerations by making the passengers to focus on the road and decrease the vestibular conflict. Finally, to extend our works into the machine learning techniques, in the fifth chapter, we started researching on motion prediction area and we developed a state-of-the-art motion prediction model. As declared, one of the motion sickness sources is ability to anticipate the direction of movement. Therefore, having a trustable prediction on the next trajectories, can even help decreasing the motion sickness and increasing the comfort. In the other hand, autonomous driving motion forecasting is essential to have a correct and reliable planning. The influence of the road agents on each other makes it even more challenging. However, most prior works have not considered these interactions and planning against fixed predictions would reduce the ability to represent the future interaction possibilities between different agents. In this chapter of the thesis, we propose a model that predicts the agents’ behaviour in a jointly manner. We take advantage of using masking strategy as the query to our model. Our model architecture uses a unified Transformer architecture by employing attention across the road elements, agent interactions and traffic rules in intersections. We evaluate our approach on autonomous driving datasets for behavior prediction and test it on python. Our work demonstrates that motion forecasting by a model with a masking strategy and having attentions and traffic rules can lead us to a state-of-the-art model. For the last three chapters mentioned above, I succeeded to publish the related publication as bellow: 1. Moazen, I., & Burgio, P. (2021). A Full-Featured, Enhanced Cost Function to Mitigate Motion Sickness in Semi-and Fully-autonomous Vehicles. In VEHITS (pp. 497-504). 2. Moazen, I., Burgio, P., & Castellano, A. (2022, August). Motion Sickness Minimization Alerting System Using The Next Curvature Topology. In 2022 IEEE International Conference on Mechatronics and Automation (ICMA) (pp. 635-640). IEEE. 3. Submitted: The Advantage of Using Traffic Rules for Motion Prediction in Intersections (TRMPI), In 2023 IEEE International Conference on Mechatronics and Automation (ICMA)