We test geometric control, estimation, and learning algorithms through indoor, outdoor, and shipboard flight experiments. Our flight hardware and software developed in-house provide direct access to sensing, onboard computation, and control, allowing new methods to be evaluated under real disturbances, measurement uncertainty, and computational constraints.
Indoor flight testing in the lab’s netted motion-capture space.
Selected experiments
These demonstrations connect the methods described on our Research page to physical systems. Each highlights a different challenge in control, learning, or navigation.
Agile flight and learning
Large rotations, wind disturbances, and coupled vehicle–pendulum motion test how geometric controllers and learned policies perform beyond nominal flight conditions.
Flying inverted pendulum
Tests geometric control and learned dynamics corrections while a quadrotor balances an upright pendulum in flight.
Equivariant reinforcement learning
Evaluates the transfer of control policies that incorporate the symmetries of quadrotor dynamics.
Modular reinforcement learning
Tests learned flight control with separate policies for translational and yaw motion.
Quadrotor backflip in wind
Tests adaptive geometric control during a large-angle maneuver under wind disturbances.
Quadrotor backflip
Demonstrates flight control through a complete rotation and recovery to upright flight.
Quadrotor hovering in wind
Evaluates position regulation and disturbance rejection during hovering flight.
Constrained attitude control
Examines attitude maneuvers with restrictions on the directions a rigid body may point.
Aerial transportation
Transporting a suspended load couples the motion of the aircraft, cable, and payload. These experiments test how controllers coordinate that motion while tracking a desired trajectory.
Autonomous payload transportation
Tests trajectory tracking for a suspended payload while controlling the coupled motion of the vehicle and load.
Autonomous aerial transportation
Demonstrates aerial load transportation, where vehicle motion must account for the dynamics of the suspended load.
Perception and navigation
Autonomous navigation depends on useful state estimates despite incomplete observations, sensor delays, and moving surroundings. These demonstrations include indoor maritime emulation, shipboard flight, and aerial and ground exploration.
Indoor maritime validation
Tests onboard perception, estimation, and autonomous flight with a physical UAV and an indoor visual emulation of a maritime environment.
Autonomous shipboard landing
Demonstrates autonomous landing during shipboard flight, complementing controlled indoor validation.
Navigation with delayed RTK GPS
Tests autonomous flight using state estimates that account for delayed localization measurements.
Laser-guided landing
Demonstrates the use of laser measurements to guide a UAV to a landing surface.
Autonomous aerial exploration
Demonstrates mapping and exploration with an aerial robot.
Autonomous ground exploration
Demonstrates autonomous exploration with a ground robot.
Facilities and platforms
Flight spaces and motion capture
- A 40 ft × 20 ft netted indoor flight space with 12 Vicon Valkyrie motion-capture cameras.
- A 20 ft × 20 ft netted outdoor flight space.
Onboard sensing and computation
- Flight hardware based on NVIDIA Jetson modules, including Orin.
- RTK GPS localization.
- Flight software developed in C++ and Python, with ROS integration.
Computing infrastructure
- A GPU server with two NVIDIA A100 80 GB GPUs and one NVIDIA A100 40 GB GPU.
Open-source software
We share selected control and simulation tools through the FDCL GitHub organization, supporting implementation and further development of the methods used in our research.
- Geometric UAV Control: implementations of geometric quadrotor controllers in C++, Python, and MATLAB, including standard and decoupled-yaw formulations.
- UAV Simulator: a Python, ROS, and Gazebo simulation environment for a UAV with geometric control.