ETD

• 1:5 scale
• acados
• advanced control synthesis
• as-rti-mpc
• autonomous vehicle
• benchmarking
• compute units
• cone detection
• constrained optimization
• delaunay triangulation
• efficiency
• ekf-slam
• embedded real-time control
• f1tenth
• gnss/ins
• graph-slam
• hardware integration
• lidar
• localization
• mit racecar
• modular communication
• neural networks
• nonlinear model predictive control
• path planning
• perception
• pid
• platform development
• power systems
• racing applications
• real-time iteration
• robustness
• ros2
• safety systems
• safety-critical robotics
• sensor fusion
• smartdrive
• software architecture
• spline fitting
• stereo camera
• tracking accuracy
• trajectory generation

This thesis presents the design, development and validation of SmartDrive project, a 1:5 scale autonomous vehicle platform integrating advanced hardware, software and control for racing applications. Inspired by initiatives like F1tenth and MIT RACECAR, it advances embedded real-time control for racing scenarios.
The work covers three core areas: platform development, software architecture and advanced control synthesis. The hardware integrates high-fidelity sensors (stereo camera, LiDAR, GNSS/INS), robust compute units and engineered power and safety systems to ensure reliable operation and emergency intervention.
The software, built on ROS2, enables modular subsystem communication. Perception combines neural networks and sensor fusion for cone detection; localization leverages EKF-SLAM and Graph-SLAM. Path planning applies a two-phase strategy: exploration, with centerline computation via Delaunay triangulation and spline fitting, followed by optimized trajectory generation minimizing distance or curvature.
A key contribution is the Advanced-Step Real-Time Iteration Model Predictive Control (AS-RTI-MPC), addressing nonlinear MPC challenges in on board systems. Implemented with the acados solver, the controller meets stringent timing while preserving nonlinear constrained optimization benefits.
Performance benchmarks show AS-RTI-MPC outperforms PID and standard MPC in tracking accuracy, efficiency, and robustness. This work delivers a complete methodology for autonomous racing platforms and advances real-time nonlinear control in safety-critical robotics.