A Strategy-Driven Multi-Agent Control Framework for Autonomous Racing

This project presents a strategy-driven multi-agent control framework for autonomous racing. The work studies autonomous racing not only as a low-level trajectory-tracking problem, but also as a hierarchical decision-making and control problem involving racing objectives, tactical planning, vehicle interaction, and controller execution.

The proposed framework connects three main layers: a high-level strategy layer that represents racing objectives and risk preferences, a tactical planning layer that translates these objectives into reference paths, speed targets, role assignments, and safety margins, and a low-level control layer that executes vehicle motion through feedback and optimal control methods. A consensus-inspired multi-agent structure is introduced to describe how multiple autonomous racing vehicles can coordinate under shared team-level objectives while still maintaining distinct vehicle roles.

The controller design component compares several control approaches, including PID, ADRC, LQR, and MPC. PID and ADRC are used as baseline and robustness-oriented controllers, while LQR and MPC are studied as optimization-based methods that can better connect tracking performance, control effort, stability, constraints, and strategy-dependent priorities. Simulation-based evaluation is used to compare controller behavior across tracking error, heading error, control effort, safety penalty, and strategy consistency.

The results suggest that autonomous racing should be viewed as a strategy-aware multi-agent control problem rather than only a path-following task. In particular, LQR provides a practical balance between performance and computational simplicity, while MPC offers a promising direction for future constraint-aware and predictive racing control. The project also identifies several limitations, including simplified vehicle dynamics, incomplete physical feasibility validation, and the need for a more rigorous strategy-to-control mapping.

Overall, this work provides a full-year project report and research foundation for future studies on autonomous racing, hierarchical control, multi-agent coordination, and strategy-aware vehicle autonomy.