SwarmAero 🚁

Autonomous 3-drone swarm with quaternion EKF state estimation, LQR attitude control, minimum-snap trajectory generation, and graph Laplacian formation control.

Built on ROS2 Humble + PX4 SITL + Gazebo | Author: Maanav Shah | Purdue ECE

---

System Architecture

IMU (250Hz) ──→ Quaternion EKF ──→ State Estimate GPS (30Hz) ──→ │ │ Gazebo GT ──→ NIS/NEES Validation │ ↓ LQR Attitude Controller │ ↓ Minimum-Snap Trajectory Follower │ ↓ Graph Laplacian Formation Control (3 agents)

Modules

1. Quaternion EKF (src/swarm_ekf/)

• State vector: [pos(3), vel(3), quat(4), accel_bias(3), gyro_bias(3)] — 16 states
• Exact discrete quaternion update: q_{k+1} = q_k ⊗ exp(0.5 * ω * dt)
• Unit quaternion re-normalization after every prediction step
• Gyro + accel bias modeled as random walk with tuned process noise
• Joseph form covariance update: P = (I-KH)P(I-KH)^T + KRK^T
• NIS/NEES within 95% chi-squared bounds confirming filter consistency

2. LQR Attitude Controller (in progress)

3. Minimum-Snap Trajectory Generation (in progress)

4. Multi-Agent Formation Control (in progress)

---

Results

NIS/NEES Filter Consistency

---

How to Run

Terminal 1 — PX4 SITL + Gazebo:

cd ~/PX4-Autopilot && make px4_sitl gz_x500

Terminal 2 — ROS2 Bridge:

MicroXRCEAgent udp4 -p 8888

Terminal 3 — EKF Node:

source ~/SwarmAero/ros2_ws/install/setup.bash
ros2 run swarm_ekf ekf_node

Hit Ctrl+C to stop and auto-save NIS/NEES plots to results/.

---

Dependencies

• ROS2 Humble
• PX4-Autopilot (SITL)
• Gazebo 8
• micro-XRCE-DDS-Agent
• Python: numpy, scipy, matplotlib, transforms3d