Drone deliveries often use suspended payloads, keeping untrained users receiving the payload away from the actual equipment. But while it's dangling from the drone via tether, how can you maintain control of the payload? Guanrui Li, Alex Tunchez, and Guiseppe Loianno have been developing the answer to that question at NYU's Agile Robotics and Perception Lab, demonstrating a drone keeping rock-solid control over a suspended payload in a recent video.
The drone uses its camera and inertial measurement unit (IMU) to determine the behavior of the drone itself along with the suspended payload. It must also predict what the payload is going to do next at any given moment and how that will factor into the drone's next movements. The team at NYU developed a model predictive control (MPC) system to accomplish this, adding perception constraints to ensure that the drone's behavior keeps the payload in view at all times. This proposed control method, utilizing both state estimation and the MPC system, guarantees the respect of the system dynamics and actuator constraints as well as constant payload visibility, critical design aspects of safety and resilience for the delicate task of transportation.
The control method also presents some interesting design challenges, as they sought to use a minimal sensor suite for autonomous navigation, i.e., only the single downward-facing monocular camera and IMU, along with the drone's Snap. The system must concurrently estimate the load and vehicle states. Unlike previous systems, which have leveraged GPS or motion capture systems for state estimation, this considers both perception and physical constraints when approaching a solution. Notably, this could make drone delivery a reality in warehouses or dense urban areas where the GPS signal is absent or overshadowed.
While this approach is currently constrained to a speed of 4 m/s, it is adaptable to varying object sizes and masses. The limits of the system are established by the actuating and sensing systems; faster speeds can introduce motion blur while decreasing the load tracking precision. The system has demonstrated the ability to estimate changes in the vehicle and load configurations. It may adapt in dynamic conditions when the load's characteristics may be variable or unknown. n open research question in making the system adaptable to many situations is that of aerodynamics. Rather than increasing speed, this presents the next challenge for the team to tackle. Speed increases will bring aerodynamic disturbances and destabilize the drone. Conversely, introducing an actuated cable could compensate for windy conditions and increase delivery precision and reduce flight time and agility.
In the future, the lab is focused on tackling these challenges presented by aerodynamic conditions and adapting their system to accommodate multiple drones cooperatively managing tethered payloads that are too large for one quadcopter to handle alone. All of this research points to significant increases in drone delivery systems' viability, particularly for systems of drones delivering supplies where GPS signal is unavailable.