Zohdy, MohamedYan, HaoKaur, AmanpreetLi, LiMonroe, Ryan2025-07-112025-07-112024-01-01https://hdl.handle.net/10323/18806Advanced Driver Assistance Systems (ADAS) has become increasingly important in the automotive industry, because of customers’ need for higher safety and efficiency for transportation. Truck-trailer systems, which occupies a large market share in both commercial and recreational sections, have been facing special challenges for ADAS due to the complex vehicle system dynamics and special skills required for maneuvering, especially in reverse. In the current market products, traditional control methods consisted of path planning and classical control methods have been widely used, however, they often struggle to adapt to the changing scenarios in real-world applications. On the other hand, with the development of Machine Learning, one of its research field, reinforcement learning, shows great potential in robotics and control area. The reinforcement learning algorithm is a semi supervised machine learning approach, and it can provide a flexible solution by enabling systems to learn and improve through experience, potentially conquer the limitations of tradional approaches. This dissertation explores the development of control systems for Truck-trailer Robotics Vehicle (TTRV) system, especially during reverse driving and parking in low-speed, closed, and and unstructured environments. The research addressed challenges caused by the complex dynamics of TTRV systems, which is known for its nonlinearity, nonholonomic constraints, and under-actuation behaviours. And this research applied several modified reinforcement learning approaches to trailer control, as well as tradional control methods, to show the improved ADAS capabilities of reinforcement learning approaches, and effectiveness of satisfying the growing need for intelligent transportation solutions.This dissertation begins with a review of classical control methods which has been used for truck trailer system, including pole placement, Lyapunov controller, Proportional–Integral–Derivative (PID), Linear Quadratic Regulator (LQR), Model Predictive Control (MPC), and other nonlinear control methods. Multiple path planning approaches such as Dubins paths, Reeds-Shepp paths, and the A* algorithm are also applied and tested to the articulated vehicle. While these methods have shown capability in certain scenarios, changing environments, online calibration, and complex dynamics can make tradional algorithm struggle in real application situations. Moreover, these algorithms require pre-design and fine-tuning from engineers, which limits their application and self-evolution in real-world situations. All these problems slow down the popularity for advanced TTRV control systems in nowadays fast growing market. To overcome these limitations mentioned above, an end-to-end Deep Reinforcement Learning (DRL) network solution for TTRV reverse autonomous control is proposed in this dissertation. The study compares several DRL algorithms with different computational complexities and performances, including both on-policy and off-policy approaches, all of them showed the effective of DRL in solving the trailer control problem. Moreover, a hybrid model-based reinforcement learning approach is developed in this research, which integrates classical path planning techniques into the DRL algorithm to improve sample efficiency and performance. The proposed algorithm is tested in several osbtacle avoidance simulations, demonstrating its ability to handle target chasing, obstacle avoidance, and passenger comfort simultaneously. Also, an Immitation Inspired Probabilistic Ensembles with Trajectory Sampling (IIPETS) algorithm is developed on the base of Probabilistic Ensembles with Trajectory Sampling (PETS) algorithm, this model based approach significantly improved sample efficiency, and showed outstanding robustness and adaptability for controlling the truck trailer system. The results of this study contribute to the development of intelligent ADAS technologies for truck trailer systems, which can potentially satisfy the growing demand for efficient and safe transportation market needs. The developed Reinforcement Learning (RL) based controllers demonstrated the adaptation capability to changing scenarios and overcoming the limitations of classical control methods, paving the way for a more robust and flexible TTRV intelligent control systems.ADASPath planningReinforcement learningTruck trailerVehicle controlMachine learning