Thesis title: Real-Time and Efficient Control for Autonomous RacingSubject description: Ideally, Cyber-Physical Systems need sophisticated control algorithms to guarantee system performance, but the limited computing resources of edge computing / embedded platforms restrict the real-time execution of advanced algorithms. This thesis takes autonomous racing as the investigation case and realizes the performance guarantee in two aspects: functional correctness (i.e. the effectiveness of algorithm outputs), and temporal correctness (i.e. the response timeliness of real-time tasks).Highlights:1. A trigger-based control framework is proposed to enable real-time execution of Nonlinear Model Predictive Control (NMPC) algorithm: time budget constraints are respected; trajectories remain optimal; the computational load is reduced by approximately 50%.2. A mixed integer programming (MIP) based method is used to encode overtaking decisions of race cars, which is integrated into NMPC for generating time-optimal and collision-free trajectories. A simplified version is proposed as an alternative to achieve real-time, safe, and efficient overtaking actions. 3. Taking into account perception modules (i.e. opponent detection, self-localization) and control modules, a generic architecture for LiDAR-based autonomous race cars is proposed. The data dependency between different components is represented as a directed acyclic graph (DAG), serving as a design reference for similar autonomous systems.4. A task scheduling model is proposed to map parallelizable DAG nodes to computing units in a multi-core embedded system, which increases the control refresh frequency to 4 times that of single-core computing mode for a more timely response.5. The aforementioned approaches are implemented on the embedded platform NVIDIA Jetson TX2, using Hardware-in-the-Loop (HiL) simulation to validate both functional and temporal correctness.

Development of algorithms for the TurtleBot3 robot on the ROS platform.

Inference module of Deep Neural Networks on embedded processors.