Customer case

UAV autonomous flight software platform

2020-11-05 11:41:33 132

Case Introduction

· The Generalized Autonomy Aviation System (GAAS) developed by  Beijing Generalized Intelligent Technology Co., Ltd. ( gi) is an open source drone autonomous flight framework designed for UAVs and Urban Air Mobility (UAM). Through functions such as SLAM, path planning, and Global Optimization Graph, the UAV can provide autonomous flight functions without GPS and external communication.

·In  this case, through NVDIA Jetson TX2, GAAS realized the processing of vision sensor data on board the drone, helping the drone to patrol on a fully autonomous passenger plane.

·  This case mainly uses NVIDIA Jetson TX2.


Beijing Fanhua Intelligent Technology Co., Ltd. ( gi) was established in 2015. The generalized intelligence team covers experts and scholars in many fields such as machine learning, SLAM, drones, etc.; and has a number of leading patent technologies at home and abroad. The goal of generalized intelligence is to upgrade drones from flying cameras to robots that can use 3D space, so as to accelerate the arrival of various applications of drones and UAM air traffic.

GAAS (Generalized Autonomy Aviation System) is an open source drone autonomous flight software platform. GAAS is currently one of the fastest growing aviation open source projects in the world, with developers from more than 35 countries and regions. As a project protected by the BSD protocol, any company, researcher, or drone enthusiast can legally and compliantly modify our code to meet their customized needs. GAAS can provide drones with autonomous flight functions including autonomous flight without GPS signal and external communication, landing in complex scenes, global perception, global tracking, target recognition, 3D restoration and reconstruction, and 3D path planning/obstacle avoidance navigation.

Now open source on GitHub:


Although drones are called "unmanned" drones, they are actually just no one in the sky, not requiring no one. On the contrary, drones rely heavily on human operations. In the United States, an industrial-grade drone requires a service team of five people: two pilots, a maintenance engineer, a ground station engineer, and a path planner. In China, an industrial-grade UAV also requires a team of 3 – 5 people for service. This does not include the large amount of manpower required to process the data collected by the drone.

As the UAV hardware has become more mature, the problem of UAV's dependence on human operations has gradually emerged. In the past ten years, the main development direction of drones is how to make people fly without problems (commonly known as bombers). Since 2008, with the continuous development of various open source flight controllers, the difficulty of drone operations has been simplified and the stability of drones has been enhanced. When letting the pilot fly the drone, there is no need to worry about the sudden problem of the aircraft itself. But with the maturity of flight controllers, the industry has gradually realized that dependence on manpower is a new bottleneck for drones. It is estimated that by the end of this year, the number of industrial drones in my country will reach 460,000, but as of the end of 2018, the cumulative number of people with drone pilot licenses in the country was only 44,573. The pilot gap is huge.

At the same time, even if there are pilots, the success of drone operations cannot be guaranteed. For example, Airbus patrols passenger planes through drones, requiring each flight to have an error within 10cm, which is an accuracy that cannot be achieved by pilots. Moreover, the cooperation between pilots and drones needs to rely on GPS and other GNSS geographic location information systems, so in many scenarios, drones cannot be used. For example, there is no GNSS signal under the bridge inspection bridge; or in strong interference environments such as substations, drones cannot be used. Even these scenarios are just needed for drone inspection.

The autopilot function of traditional UAVs is only flying based on GPS waypoints, which can no longer meet the needs of autonomous flight of the next generation UAVs. Because the UAV itself has limited endurance and load capacity, high-performance and low-power airborne processors are required for edge computing for autonomous flight. Moreover, a large number of images collected by drones also require GPU operations for deep learning image recognition.


·  TX2 helps drones achieve high-performance edge computing. The autonomous flight of UAVs requires high computing equipment, but due to the endurance and load capacity, UAVs can only use lightweight processors. In the past, it was difficult for developers to find a suitable choice—computing devices with sufficient performance could not meet the weight and power requirements of drones; processors with appropriate weight and power consumption were insufficient. TX2 allows drone developers to find the right choice for the first time, with strong performance and suitable power consumption. And equipped with GPU can better process image information through neural network to help drones fly autonomously.

·  Detailed documentation and support services greatly facilitate the work of developers. In the past, drone developers could only develop on specific proprietary chips. Incomplete documentation and poor support have always been a headache for developers. TX2 has detailed development documents, and there are global developers who can help us answer questions. Accelerated the speed of UAV development.

At present, through TX2, GAAS can deploy a series of algorithms such as SLAM, path planning, autonomous landing, and target tracking on drones without the need to customize chips and reduce functions. GAAS uses TX2 to achieve end-to-end wireless for drones. Open source framework for autonomous flight of man and machine.


Using NVIDIA Jetson TX2, GAAS enables drones to be able to achieve a level equivalent to the level 4 of vehicle autopilot-except in emergency situations, without human intervention. This accelerates the expansion and utilization of new drone scenarios by drone companies and developers.