Overview
Designed and compared three control strategies for autonomous parallel parking: Feedback Linearization, LQR (Linear Quadratic Regulator), and Sliding Mode Control. The modified Sliding Mode Controller (with chattering reduction) achieved a 21% performance improvement over the baseline, validated in MATLAB–CarSim co-simulation with realistic vehicle dynamics.
The study provides a systematic, quantitative comparison of classical and modern control approaches applied to a real autonomous driving task, contributing insight into which controller properties matter most for tight geometric maneuvers like parallel parking.
The Challenge
Parallel parking is a challenging autonomous driving task — it requires precise path following in tight spaces with hard geometric constraints. The parking slot dimensions allow little margin for error, and the reverse maneuver requires accurate lateral and longitudinal control simultaneously. Controller design must balance tracking accuracy, smoothness (chattering in SMC is a well-known problem), and robustness to disturbances.
Standard controllers often fail to satisfy all constraints simultaneously: Feedback Linearization requires an accurate dynamic model, LQR is optimal only near the linearization point, and standard SMC produces chattering that causes mechanical wear and passenger discomfort. These limitations motivated a comparative study that also explored modifications to address the chattering problem.
Technical Approach
- Feedback Linearization controller: exact linearization of the nonlinear vehicle kinematic model, transforming the parking path-following problem into a linear control design task
- LQR controller: optimal linear controller minimizing a quadratic cost function on state deviation and control input — balances tracking accuracy against control effort
- Sliding Mode Controller (SMC): robust nonlinear controller with guaranteed finite-time convergence to the sliding surface, providing strong disturbance rejection at the cost of chattering
- Modified SMC with boundary layer: chattering reduction via a smooth saturation function replacing the discontinuous sign function, improving practical smoothness while retaining robustness
- MATLAB–CarSim co-simulation: MATLAB/Simulink controller driving the CarSim vehicle dynamics model for realistic validation — CarSim provides high-fidelity tire and suspension models beyond simple kinematic models
- Performance metrics: path tracking RMS error, maneuver execution time, control input smoothness, and robustness to initial position perturbations
- 21% performance improvement measured for the modified SMC versus the unmodified SMC baseline, quantified across the full set of performance metrics