In the big game of drones, there are already many players, and now more giants are rushing in. Intel and Qualcomm have both entered the core chip field for drones, and their demonstration products were almost released simultaneously; Tencent previously launched the “Kongying” drone; next week, Xiaomi’s drone is also set to be released, challenging DJI, which has long held a firm grip on the consumer drone market.
Although it is still not very clear, it seems that the large space of “drone intelligence” is the decisive factor attracting these manufacturers.
Here, “intelligence” means that not only cars need to drive automatically, but drones also need to fly autonomously. Of course, this article mainly discusses civilian drones.
This issue’s smart insider recommends this report from GF Securities, which can be downloaded in full by replying with the keyword “intelligence” on Zhidx (WeChat public account: zhidxcom).
Market Control
1. The American financial services firm Oppenheimer summarized some predictions from third-party research institutions regarding the current state and future of the drone market. Although there are some contradictions in the above statistics and forecasts, the overall view of this market is mostly optimistic.
2. For the Chinese market, Analysys believes that the market size in China is expected to reach 2.3 billion yuan in 2015, and will maintain a growth rate of over 50% in the coming years, exceeding 10 billion yuan by 2018.
3. Currently, civilian drones have already been commercially applied in some industries. The FAA (Federal Aviation Administration) of the United States has started issuing commercial operation licenses for drones since May 2014. By September 2015, over 1,000 applications for commercial operations had been approved. The first batch of approved industry applications is as follows.
4. The FAA also compiled statistics on the performance of the first batch of approved commercial application drones. It can be seen that the vast majority of these commercial application drones are still lightweight drones with an average weight of 5KG (11.02 pounds) and a flight time of about 30 minutes, which are very similar in characteristics to consumer drones.
5. The FAA also compiled statistics on the manufacturers and models of drones used in the first batch of commercial applications, where DJI leads with a 70% market share, while the second-ranked 3DR has only a 10% market share, and manufacturers like Haoxiang and SenseFly have a market share of less than 2%. DJI’s Phantom 3, Wukong 1, and Phantom 2 occupy the top three positions in the declared model share. From the above statistics, it can be seen that there is no absolute threshold between consumer drones and commercial drones; mid-to-high-end consumer drones can perform a large part of commercial aerial photography tasks.
Why Intelligent?
1. The potential of drones lies in their relatively simple mechanics; if intelligent, they can be viewed as a “flying robot”.
2. The Congressional Research Service of the United States has listed some changes in the commercial field of drones after intelligence, which can open up many limitations of ordinary drones in various application fields.
3. The Tabor Research Institute reported that in 2015, the total amount of venture capital financing for civilian drones worldwide exceeded 4 billion yuan, while Oppenheimer’s total financing was $438 million.
Some well-known financing cases in the drone industry in 2015 include:
On December 31, 2014, EHang announced completion of a $10 million Series A financing from GGV Capital;
On February 26, 2015, 3D Robotics raised $50 million in Series C financing, led by Qualcomm;
On May 6, 2015, DJI Innovations received $75 million from Silicon Valley venture capital Accel Partners;
On August 24, 2015, EHang announced completion of $42 million in Series B financing;
On August 26, 2015, Intel invested $60 million in Yuneec;
How to Achieve Intelligence?
Achieving drone intelligence requires technological upgrades in both hardware and software.
On the hardware side, machine vision hardware technology needs to be introduced;
On the software side, machine vision software technology, artificial intelligence technology, and other necessary software technologies such as OS, flight control, and navigation need to be well-prepared.
1. Machine vision hardware technology mainly includes three categories:
2. Machine vision software technology currently has two types of algorithms:
Optical flow algorithm: The optical flow algorithm is a single-camera vision technology that completely relies on software algorithms to solve motion detection and positioning problems. The optical flow method obtains the optical flow field of different objects in the scene by comparing two adjacent images taken at different times by the same camera (differential operation).
The advantage lies in its reliance on software and computing power to recognize objects and distances, thus requiring less additional optical and acoustic hardware, saving costs. However, its disadvantages include low precision and significant limitations (not suitable for poorly lit indoors, glass, thin wires). Tencent’s “Kongying” drone uses a single-camera optical flow algorithm to complete obstacle detection and relies on ultrasonic assistance to avoid obstacles.
Image segmentation algorithm: Edge detection is a widely used image segmentation algorithm, which works by first deriving the gray function of the image and then detecting the edges of objects based on static or dynamic threshold settings.
All edge detection algorithms share a common feature of high computational requirements. For example, in Qualcomm’s implementation of a binocular depth camera solution, a separate core in the Snapdragon 801 multi-core chip is needed to run dedicated SIMD (Single instruction, multiple data) operations, which can only provide 30Hz real-time computing capability. For quadcopters that can fly at speeds of up to 20m/s, this frequency means that during the two recognition periods, the drone has already flown 0.66m, making it unable to respond in time to suddenly appearing objects in the environment.
3. Artificial intelligence includes image recognition algorithms, facial recognition algorithms, and voice semantic recognition algorithms. This allows drones to perform more functions, such as scanning specific components on electrical equipment; grabbing specific objects (an orange instead of a banana) according to human instructions and flying to another room; distinguishing ordinary animals from suspicious humanoid targets during border patrols and alerting in time; and flying control through the operator’s voice.
Unlike image segmentation algorithms, the computational load of neural network-based image recognition/facial recognition/voice recognition algorithms varies greatly depending on the usage scenario.
4. Intelligent drone software technology also includes: flight control, navigation and path planning, as well as operating systems that support all intelligent software. In addition, intelligent drones need some underlying firmware code to connect hardware and software systems, ensuring normal and efficient operation of communications, sensors, and computing units. For intelligent drone manufacturers, independently developing such complex and diverse software is akin to “reinventing the wheel”; customizing and processing based on open-source software projects can accelerate product iteration and reduce development costs.
The earliest open-source projects in the drone field appeared in the flight control software area, and currently popular open-source drone flight control (including hardware and software) includes.
5. On the chip side: Currently, mainstream manufacturers use ARM architecture MCU (Microcontroller Unit) chips on existing drones, such as STMicroelectronics’ STM32 series (used in DJI Phantom series drones), Atmel’s Mega2560 series, etc. These chips are characterized by low main frequency (STM32’s main frequency is around 200M Hz, Atmel’s MCU as low as 20M Hz), poor computing power, often only supporting flight control functions, and cannot provide the high-speed computing and parallel computing capabilities required for drone intelligence.
Currently, the three major chip giants: Qualcomm/Intel/Nvidia are all entering this industry, launching drone reference designs or kits based on their mobile chips or graphics computing chips; in addition, the Chinese chip design company Unisoc Technology has also developed solutions for intelligent drones in collaboration with Chinese drone manufacturer Zero Degree Intelligent Control.
Qualcomm launched the Snapdragon Flight reference design;
Intel launched the Edison for Arduino open-source hardware flight control reference design board;
Nvidia launched the Jetson TX1 drone and robot chip module solution;
Unisoc launched drone flight control based on LC1860;
6. In the field of open-source tools: There are many open-source software projects and open-source software libraries in areas such as machine vision and artificial intelligence.
Based on the development history of the open-source software industry, we predict that in the future, a batch of software productization secondary development companies will emerge in the field of intelligent drone software, developing based on open-source machine vision and artificial intelligence software, and selling packaged mature and usable intelligent software products and supporting interfaces to downstream drone manufacturers.
DJI has already occupied a large part of the civilian drone market, but the low threshold for drone development has led to many participants in the industry, and competition remains intense.
On the road of innovation, if you do not actively think, there will always be people approaching you from all directions. It’s not that DJI is not trying hard, but the trend of terminal intelligence is evident, and once the time is right, everyone will take action.
Long press to identify the QR code in the image to quickly follow us!
