Development of a New Intelligent Algorithm to Improve Autonomous Car Operation

Autonomous Driving Systems (ADS) are transforming modern transportation by enabling safer, more efficient vehicle operation. Among their core components, local path planning remains a significant challenge due to the need for optimal navigation decisions in complex environments while balancing safety, smoothness, and computational efficiency. Existing methods suffer several drawbacks, including generating non-smooth paths, vulnerability to local minima, reliance on static parameters, random exhaustive search behaviours, and poor balance between exploration (searching new areas) and exploitation (refining known areas). This thesis aims to enhance local path planning and real-time vehicle control through a unified, optimisation-driven methodology. A novel algorithm, Dynamic eXplorative Multi-Operator Differential Evolution (DXMODE), is proposed to overcome the drawbacks of current optimisation methods. For path planning, DXMODE uniquely integrates optimisation, reinforcement learning, and interpolation techniques to improve path smoothness and consistency across repeated runs. In parallel, two additional algorithms are developed to enable real-time PID tuning of speed and steering control. The methodology is implemented within a modular architecture using the Robot Operating System (ROS), integrating all system components in a hardware-independent architecture. It is validated on two vehicle prototypes: a four-wheel differential drive platform and an Ackermann personal mobility scooter. Comprehensive validation is conducted across 50 simulated path-planning scenarios, 64 standard optimisation benchmarks from the Congress on Evolutionary Computation (CEC), and six real-world driving scenarios. Compared to 22 state-of-the-art algorithms and previous CEC winners, DXMODE consistently achieves top performance with a 100% ranking score, producing minimal median path lengths and collision-free trajectories. Control response validation further confirms the framework’s reliability: the DC motor speed control achieves an 86.11% reduction in overshoot and a 34.83% decrease in settling time, while the steering control yields a 93% improvement in settling time with zero overshoot and no steady-state error. These results establish the proposed optimisation and ADS framework as a comprehensive, high-performance solution for autonomous vehicle navigation and broader intelligent systems.