An Autonomous VIO Quadrotor

This project was about getting a quadrotor to fly on its own through cluttered spaces without GPS, using only its onboard cameras and IMU. I tied together global planning, smooth trajectory generation in the quadrotor’s differentially flat space, low-level control, and visual–inertial state estimation into one end-to-end system, then added real-time replanning so the drone could react to obstacles and drift as it flew.

• Planned collision-free 3D paths on a voxel map, then converted them into minimum-jerk 5th-order polynomial trajectories in the flat outputs.

• Tuned a nonlinear SE(3) controller (with simple drag compensation) to track fast, agile motions.

• Fused stereo vision and IMU data with an error-state Kalman filter to estimate the drone’s full 6-DOF pose without GPS.

• Added a real-time local replanning state machine that detects impending collisions or drift and quickly switches the drone onto a freshly computed short-horizon path.

Results demonstrated high tracking accuracy across benchmark scenarios (EuRoC MAV datasets), with replanning improving robustness and reduced upfront planning time.