Analysis of UAV Obstacle Avoidance Technology Unavoided by UAV Company

Unconsciously, I am going to the new year's CES.

CES in 2016? The most popular is the intelligent obstacle avoidance performance of Realense UAV in cooperation with Intel. Then XRO exhibited the Xplorer 2, which introduced laser obstacle avoidance technology, and introduced laser obstacle avoidance technology. Soon, the first double Obstacle-consuming consumer drones? Phantom 4, also officially shipped in March, unveiled the era of automatic obstacle avoidance by drones.

The obstacle avoidance technology suddenly became the key word for the consumer drone sector in 2016.

But at the end of 2016, Intel seems to focus on professional-grade drones, and Ai Faner also reported that Yuneec's latest self-timer drone Breeze also abandoned Intel's RealSense obstacle avoidance. On the other hand, although Danya is still selling big, Phantom 4, which uses automatic obstacle avoidance as a selling point, is far less concerned than the extremely portable Mavic Pro (above).

Consumers seem to be as big as the "automatic obstacle avoidance" interest, and the automatic obstacle avoidance at the beginning of the year suddenly changed from the protagonist to the running dragon.

UAV obstacle avoidance technology overview

What is the obstacle technology of drones?

The original drone, through the gyroscope and accelerator, can sense the smoothness of the drone and stabilize its flight and hovering modes by controlling the output of the four motors. Then, the drone introduces GPS and barometer sensors, combined with gyroscope and accelerator positioning; GPS determines its own plane position, the barometer determines its vertical height, and then controls the motor output to keep the drone in the same plane. And vertical points.

Although the drone has been able to successfully find the three-dimensional "absolute positioning" (GPS plane sitting, vertical air pressure), how to perceive the surrounding environment in the air that does not rely on the four sides, know the "relative positioning" of the drone and neighboring things. (The distance from the four things around) has become the biggest issue in the industry. It is true that the latest batch of unmanned aerial vehicle obstacle avoidance technology is actually in the bones: the sensor is used to judge the distance between the surrounding things and itself, and then by controlling the motor output, stop continuing to approach the obstacles, or even bypass the obstacles.

Initially, the range measurement method commonly used by drones is “TIme of Flight (TOF)”. The drone transmits laser/infrared/ultrasonic waves, then calculates the time difference between the reflection of the wave and the sensor, and estimates the distance between the obstacle and the self. However, this method has a narrow range of perception and distance is not far enough. Ai Faner has reported that a new batch of drones in 2016 was mainly based on computer vision: using parallax (above) through binocular cameras or structured light for instant 3D modeling , determine the distance between the obstacle and the drone.

However, why is the automatic obstacle avoidance of consumer drones advanced, but it still won't fire?

Obstacle avoidance/flight, a contradictory proposition

The reason why obstacle avoidance is important is because of the existence of obstacles; without obstacles, there is no need to avoid obstacles at all.

In addition to being a robot, the drone is a flying robot; it is necessary to shoot the robot to fly to the sky, in order to avoid obstacles on various surfaces to achieve a more direct movement route, or a broader one. Unobstructed view. In short, obstacle avoidance is used to avoid obstacles, but the higher the drone flies, the fewer obstacles there are, and the need for obstacle avoidance is getting lower and lower.

So to a certain extent, the ability of the drone to join the obstacle avoidance is itself a contradiction.

Why is the automatic obstacle avoidance technology of consumer drones in the end? Is it because of the contradiction of “obstacle avoidance/flight”? ─? At present, consumer-grade drones are still dominated by aerial photography. In order to shoot the "God's perspective", the user needs to fly the drone to the sky; but there are not many obstacles to hide in the sky, the higher the aerial drone flight, the more spectacular the scenery, the more spectacular the scenery, the more avoidance The barrier is irrelevant (above).

Even so, the automatic obstacle avoidance technology of the drone is not very useful in aerial photography, but it is still very important in the industrial drone.

For example, the plant protection drone is aimed at the ground environment. It is generally sprayed over 1-2 meters of crops. When the drone is flying high, it is difficult to concentrate the spraying of pesticides. If it is not automatically obstacle avoidance, it will easily hit. Tree. Similarly, professional film drones are also aimed at the shooting scene on the ground. If you fly too high, you will become a small bean. But if you fly too low, the stars are all expensive and not careless. Who is responsible for cutting to the skin? The rest are like the disaster drones that often shuttle the disaster site, or the security drones that are patrolled and monitored.

Is automatic obstacle avoidance technology really useless?

Therefore, the drone avoidance is not useless, but on the consumer-grade drone, it can't be fired.

Consumer-grade drones, do you really need to avoid obstacles? no. Under the circumstances of take-off, landing and automatic return after the disconnection, the aerial drone still needs a certain degree of obstacle avoidance capability; not to mention the users of consumer-grade drones, many of which are novices and small drones. White, more need automatic obstacle avoidance protection. As Ai Faner pointed out, when many novices are manipulating drones, the most common mistakes are to make mistakes in the direction of operation. Simply speaking, it is “pushing forward and left, pushing right after walking” and then hitting the wall (above) .

However, because "visual obstacle avoidance" requires extremely high computing power, it exceeds the level that consumer drones can afford. Even high-ends such as Yuneec Typhoon H, Dajiang Phantom 4 or Mavic Pro can only provide one-way binocular obstacle avoidance under the constraints of cost and endurance (until the Phantom 4 Pro launched at the end of Dajiang) The ability to avoid obstacles in infrared, but the role is still limited). Even if the forward obstacle avoidance may prevent the wall accident when the air is lost, the newbie is still often hit by obstacles from the side or the back due to operational errors, and then the machine is bombed.

More ironically, the self-timer drone is flying shorter, and more need to avoid obstacles automatically? However, it is the self-timer drone with the smallest flying height, closest to the ground, the most obstacles, and the most need for automatic obstacle avoidance. It is more limited by volume, cost and endurance, and it is more difficult to introduce automatic obstacle avoidance. Perhaps to some extent, the self-timer drone lacks automatic obstacle avoidance, which hinders its popularity; while the self-timer drone has not been popularized, it has also made consumers more eager to avoid obstacles.

Why do manufacturers want to develop automatic obstacle avoidance?

If so, why is the drone company so active in developing automatic obstacle avoidance technology? Because it is the future.

To be correct, Odin's “future” does not refer to automatic obstacle avoidance technology, but “computer vision” and “automated operation”. Obstacle avoidance is only their “application scenario”. "Computer vision" and "automated operation" are both explicit in the field of robotics and artificial intelligence in the future, and automatic obstacle avoidance is the application method. For example, Mobileye uses computer vision for assisted driving, and through automatic obstacle avoidance technology, allows the car to walk autonomously to a certain extent.

However, compared to unmanned cars, drones have a greater need for "automatic obstacle avoidance" and are less difficult. For example, it is not easy for us to manually control a certain number of drones and carry out long-term flight missions. Therefore, drones such as plant protection or security have begun to use centralized management of ground stations. , for a long time automated automatic spraying or monitoring tasks (above). Therefore, industry drones must have reliable automatic obstacle avoidance capabilities to allow drones to operate on their own.

Even though automatic obstacle avoidance capabilities have not been effective in consumer drones, the computer vision behind them is also reducing the operational difficulty of consumer drones.

For example, the binocular down-view lens of Dajiang uses a stereoscopic visual odometer (Visual Odometry) technology to make the drone's height determination ability more precise; and its FlightAutonomy five-way sensing technology greatly enhances the indoor positioning of the drone. Capabilities, as well as binocular vision and lateral infrared sidelights, make the drone more stable - not to mention the various visual following and pointing flight functions based on computer vision (above), turning the drone's operation into simpler?

The future of automatic obstacle avoidance?

In general, Odin does not deny the automatic obstacle avoidance capability of consumer drones. At present, it is still tasteless, and it is tasteless. However, although consumers are currently not interested in automatic obstacle avoidance, due to the current potential of computer technology computing, there is still room for development in the consumer market in the future.

Therefore, if the drone company simply does not use automatic obstacle avoidance for the aerial photography market, or because the self-timer drone does not contain the computer vision, it is too short-sighted.