Contact Us
Developing AI Drones that Fly Autonomously to Perform Inspections and Other Tasks
Project Summary
ARAYA's Autonomous Agent Team is developing technology that enables drones to fly autonomously based on image information from cameras mounted on the drones. Using this technology, the team aims to significantly improve the efficiency of inspection, inventory management, and other tasks that have traditionally been carried out by humans.
Social context of the project
In logistics and infrastructure inspections, there is an increasing need for manpower and work safety. For example, inventory in warehouses and inspections in factories require work in high places and extensive movement and coverage.
By using drones that fly autonomously, work at heights and over a wide area can be done automatically, in a short time, and at low cost.
Benefits of Araya's Autonomous Drone
The autonomously flying drones that ARAYA is developing have the following main advantages.
Application example 1: Inventory in the warehouse
The drone flies autonomously, tracing all the products, scanning the barcodes and QR codes on the inventory, and registering the inventory in the inventory management system.
The system is expected to save labor, improve inventory accuracy, and shorten shutdown periods.
[2020.5.18 updated] Demo has been uploaded. Please see this page as well.
Application example 2: Power line inspection
2020.5.18 Update] Application Examples and Demos have been uploaded. Please see this page as well.
The drone flies along a flexed power line and captures images / videos of the power line.
At that time, by controlling the transmission line so that it is kept at a certain distance, it is possible to always take a picture of the transmission line at a resolution appropriate for inspection.
Application example 3: Piping inspection in a factory
The drone follows the pipes that are installed in high places and over a wide area in the factory and inspects them for damage and abnormalities.
This is expected to save labor, improve safety, and increase inspection accuracy.
Technology to realize autonomous flight
By applying deep learning to drone control, we can autonomously determine the direction of movement of the aircraft based on the current camera image. Specifically, we use supervised learning and reinforcement learning to train a deep neural network that uses image data as input and outputs drone control commands (left turn, forward turn, right turn).
Example of current results: a drone that can fly along floor lines
It flies to follow the pink line drawn on the floor. It can also perform pre-defined tasks when it detects certain markers.
The drone is able to determine its flight direction autonomously from the camera image without being given a path in advance.
For the future
In the future, we are working to develop technology that will enable autonomous flight based on various indoor clues, even in indoor environments where there are no lines drawn on the floor.
