Precision landing for drones as any landing method that is accurate to less than or around 10 inches. Precision landing is particularly important since the conventional GPS on drones are only accurate within 10 feet radius. Therefore it has proposed different techniques to achieve precision landing especially in use cases such as package delivery, surveillance, drone in a box, charging stations, and indoor flight.

Two standards methods have been identified from several solutions that are currently in use [7]. The first method involves mounting a computer vision-enabled camera on the drone to track an image designated as the landing pad on the ground (or on the carrier). The second method involves using an IR-lock for tracking an IR emitting beacon from the ground (carrier). Both methods can reach the precision level landing that is required for a drone in a box condition. In terms of cost, the IR-transmission method cost more to purchase the hardware but is easy to integrate (plug and play) and deploy while the computer vision-enabled camera hardware is not costly but required the development of customised solutions to detect and track the landing page and also integrate with the drone’s internal solution. Also, the computer vision-enabled camera method is affected by visual attenuation such as low/no light and bad weather. However, the authors of [3] proposed a method that works both day and night. The method makes use of a marker image with a width and height of about 1 meter and has three inner circles that are divided into eight areas of evenly distributed black and white sections. The marker further consists of simple geometric lines and circles based on colour detection of white and black. Adaptive thresholding and Hit and Miss morphology algorithms are used in night conditions.

[1] employed a downward-facing fisheye lens camera which provides a large Field-of-View to ensure that the UAV can land accurately. They also provided an algorithm (pose estimation) that ensures that the position estimate relative to the take-off path of the drone is used to guide the drone during landing in an unstructured environment (follows the take-off path in reverse). The proposed method could correct drift error and land with an accuracy of about 40cm. This method requires an extra camera (Fisheye camera), laser range finder, and a computing device such as NVIDIA Jetson TX2 for on-boarding pose estimation.

Tracking and landing on a moving platform is another challenge faced by UAVs. This is particularly challenging because of the limited battery life of drones. In addition, the size of the dock since it can be confined and finite and the velocity of the platform need to be considered. A method such as a velocity estimation is not effective in the long run as the drone loses reference to its surrounding environment. Using low-quality sensors, enough spatial localisation information can be communicated between the ground vehicle and the UAV. [4] mounted a camera on a UAV to provide feedback based on a visual marker tag located on the landing platform. In addition, a mobile device on the moving platform (around 50 km/h) sends GPS, inertial measurement unit data, and headings that are relative to the landing pad. Kalman filter was used in the estimation of the position of the landing pad.

[5] employed reinforcement learning to land drones on a moving platform. Deep deterministic policy gradients (DDPG) algorithms were used to create a rewarding system for successful landing which makes the UAV learn how to minimise the position differences of the moving vehicle and also monitors the velocity difference between it and the platform.