Creating a playground for testing control frameworks and performance optimization for unmanned robots.
Year
2025-2026
Aquila is an end-to-end simulation and flight-software playground I built to study, validate, and push control algorithms and machine learning-based autonomy.
The study began by developing a ground-up 6-DOF rigid-body digital twin inspired by the Skywalker X8, implementing quaternion-based attitude propagation, aerodynamic force modeling, sensor emulation (IMU, GPS, barometer), and a cascaded GNC architecture for waypoint following, altitude hold, and airspeed control. Emphasis was placed on enforcing closed-loop stability under injected noise—mirroring the failure modes encountered in real flight software. Key challenges included control and plant mismatches, pitch and energy coupling, and actuator saturation, which I mitigated through model sanity checks and iterative controller simplification to isolate the sources of instability.
Building on this validated legacy engine, I transitioned to an industry aligned robotics stack on MacOS using containerized PX4, Gazebo Harmonic, and ROS 2 Humble to simulate a custom UAV quadrotor with vision and LiDAR sensing. The platform uses Micro XRCE-DDS for data exchange between the flight controller and off-board compute.
On top of this infrastructure, I am currently implementing factor graph-based state estimation (GTSAM) to fuse LiDAR and IMU data for localization in GPS-denied environments, and integrating a Soft Actor-Critic (SAC) reinforcement learning agent for autonomous landing and obstacle avoidance via a Gym-wrapped environment.
Aquila serves as a scalable testbed for validating modern control strategies (e.g., TECS, adaptive and derivative-free methods) alongside learning-based policies, bolstering my confidence to bridge classical GNC, high-fidelity simulation, and ML-driven autonomy into a market ready platform.
Disciplines
Machine Learning
Research
Physics
Network Protocols
Middleware
Simulation
Verification & Validation
System Testing
