Quadrotor Landing on an Unstable Moving Platform
Advanced Controls Systems Integration
Carnegie Mellon University
Pittsburgh, PA
October 2023 - December 2023
In dynamic environments, particularly maritime settings where ship motion is influenced by ever-changing sea conditions, maintaining steady-state operation is often impractical. This project aims to address these challenges by developing a novel algorithm to land a quadrotor (the CrazyFlie) on a two-wheeled platform (the Tumbller) that follows a trajectory, simulating the difficulties of landing on a ship in rough seas.
Bitcraze CrazyFlie Quadrotor (right) and Elegoo Tumbller (left) used as analogs to an aircraft and ship in rough seas
The controller divides the CrazyFlie’s maneuver into two phases: the search phase and the land phase. Each phase operates an individual high-level model predictive controller to track the landing platform, The quadrotor’s maneuver is segmented into SEARCH and LAND phases, with a finite state machine handling phase selection and objective generation. Each phase is governed by a two-level control system. A high level Model Predictive Control (MPC) generates setpoints, while a low-level PID controller adjusts the quadrotor's motors to achieve precise control.​
Finite State Machine Logic Flowchart. The CrazyFlie starts in the Search Phase, where it attempts to track the Tumbller. If the X-Y error is below a tolerance, it transitions to the Land Phase. In the Land Phase, altitude is iteratively lowered when the X-Y error is within tolerance. Once altitude is within landing tolerance, the propellers are cut off, and the test ends.
The Tumbller platform employs a cascaded PID architecture to maintain balance and regulate wheel motion to ensuring accurate trajectory tracking, without global positioning feedback, relying instead on wheel encoder feedback. To enhance landing accuracy and mitigate controller delays, a reference prediction subroutine was integrated into the control algorithm, to allow the quadrotor to make predictive maneuvers despite not having prior knowledge of the Tumbller's future states.
This approach guarantees successful landings on a platform with a diameter twice the length of the quadrotor, even under constant acceleration trajectories at speeds up to 0.25 m/s. Testing demonstrated the control system's efficacy using two different landing pads on the Tumbller. The smaller pad provided a 50% greater tolerance for error and 30% lower platform instability, resulting in a 100% landing success rate at speeds below 0.25 m/s. At a maximum speed of 0.38 m/s, the quadrotor achieved a successful landing within 5.6 seconds, with a position error of 2.5 cm. However, larger landing pads introduced increased platform instability, impacting position estimates and controller performance. The overall success rate for landing at speeds below 0.38 m/s was 60%. This project also encompassed the hardware implementation of the controllers and the development of a ROS communication architecture for computation, showcasing the practical viability of the proposed control algorithms in real-world scenarios.
Circular Trajectory
Read the full report below: