Autonomous Vehicles & Quadrotor Control Systems
@ CMU — Modern Control Theory
Introduction
Reliable autonomous behavior requires control, estimation, and planning to work together — getting one right while ignoring the others produces systems that fail the moment conditions change. This course project implemented and validated a full autonomy stack for ground vehicles and quadrotors in simulation, from low-level control through state estimation and path planning.
Methods
Four control and estimation algorithms implemented in Webots simulation:
PID & LQR — baseline and optimal trajectory tracking for differential-drive vehicles and quadrotors under dynamic constraints.
MRAC — model reference adaptive control to compensate for unknown disturbances and parameter shifts without retuning.
EKF SLAM — full pipeline fusing noisy odometry with lidar-like measurements for joint pose and landmark estimation during navigation.
A* Path Planning — collision-free global path generation combined with receding-horizon local tracking and motor-loss failure recovery.
Results
LQR and MRAC controllers achieved near-zero steady-state error across all tracking criteria. EKF SLAM maintained reliable localization through loop closures despite sensor noise. A* planning with closed-loop tracking completed complex routes under injected disturbances and simulated motor failures. Quadrotor controller met all benchmarks for altitude hold, attitude regulation, and 3D waypoint transitions.
Discussion
The clearest takeaway: tracking performance is bounded by estimator quality, and planning quality is bounded by estimator consistency. Treating control, sensing, and planning as independent problems produces systems that work individually and fail together. Coupling them — designing each layer with the others' limitations in mind — is what makes autonomous behavior actually reliable.
My Contributions
Implemented full vehicle and quadrotor dynamic models in Webots.
Designed PID, LQR, and MRAC controllers meeting all performance criteria.
Developed EKF SLAM pipeline for joint state and map estimation.
Implemented A* path planning with closed-loop tracking and failure recovery.