From November 6th to 9th, the 20th Association for Computing Machinery (ACM) Conference on Embedded Networked Sensor Systems (SenSys) was held in Boston, USA. The research group of Associate Professor He Yuan, School of Software, Tsinghua University and the UAV team of Meituan cooperated in the paper "MicNest: Long-Range Instant Acoustic Localization of Drones in Precise Landing" (MicNest: Long-Range Instant Acoustic Localization of Drones in Precise Landing ) won the second place of the Best Paper Award (Best Paper Runner-Up).
ACM Embedded Network Perception System Conference is the flagship academic conference in the field of Internet of Things hosted by ACM. It has been held for 20 consecutive sessions since 2003. This conference received a total of 209 paper submissions, 52 of which were accepted for publication, and finally selected 1 "Best Paper Award" (Best Paper Award) and 1 "Best Paper Runner Up Award" (Best Paper Runner Up Award) ).
The research motivation of this paper comes from the real application scenario of the urban low-altitude logistics network that Meituan is building. High-precision positioning and tracking technology is one of the key technologies to ensure the safe, reliable and accurate take-off and landing of delivery service UAVs. The research proposes a long-distance real-time sound source localization scheme based on ground microphone arrays, which effectively solves the problems of fast signal attenuation, low signal-to-noise ratio, and Doppler nonlinear distortion in complex urban environments. The positioning height is up to 120m, and the relative positioning error is 0.5%.
background
In order to enable UAVs to meet the requirements of distribution operations, the UAV flight control system mainly relies on RTK, vision and other information to realize the positioning of UAVs. However, in urban environments, especially in urban canyon scenes close to the ground, the reflection of buildings near the drone may even block the signals of GPS satellites, resulting in severe multipath effects or non-line-of-sight signal propagation.
The long-term business goal of drone delivery is to achieve all-weather delivery. In order to improve the robustness of drone positioning, this paper proposes a positioning method based on sound waves.
2 overall framework
The paper proposes a novel acoustic positioning system to help UAVs land precisely. As shown in Figure 1 below, the UAV will be equipped with a common speaker, so that the UAV can send acoustic pulse signals to assist positioning. Airports on the ground will deploy multiple microphones as positioning anchors. The ground airport will detect the acoustic pulse from the signals collected by each microphone, and then calculate the relative time delay of the pulse signal to locate the UAV.
This paper needs to solve the following three technical problems:
-
The first challenge is that the signal-to-noise ratio of the acoustic pulse signal is very low. There are four reasons: (1) Since the drones are traveling between cities, the transmission power of the speakers must be limited to avoid disturbing the city dwellers. (2) This system needs to locate high-altitude drones (>100m), and the acoustic pulse signal will suffer a lot of signal attenuation. (3) Also, the background noise in many cities is inherently strong, around 40-75 dB SPL. (4) When flying in the air, the UAV propeller will also produce a lot of acoustic interference, which may be as high as 104 dB SPL.
-
The second challenge is Doppler signal distortion caused by drone motion. For wireless signals, the severity of the Doppler effect is proportional to the speed of the object and inversely proportional to the propagation speed of the wireless signal. Compared with the propagation speed of radio frequency signals (ie, the speed of light), the propagation speed of acoustic signals in air is very slow. Therefore, the acoustic signal will suffer severe Doppler distortion.
-
A third challenge is that signal processing must be efficient. This positioning system is used to guide the moving UAV to land, and the positioning data delay should not be too high, otherwise the high-delay positioning result will introduce serious system instability to the flight control, threatening the safety of the entire landing process.
In a word, the core technical problem to be solved in this proposal is: how to detect the distorted acoustic pulse signal under the condition of low signal-to-noise ratio.
Figure 2 above shows the schematic diagram of the positioning process of this proposal and the corresponding system implementation: the UAV carries a loudspeaker to continuously play the acoustic pulse signal. Four microphones are deployed at the four corners of the landing platform to capture the acoustic pulses. This system locates the drone by locating the position of the speaker.
3 specific plans
3.1 PRN modulation and transmission
In a real scenario, the acoustic signal sent by the UAV needs to meet the following requirements:
-
Acoustics are friendly to the human ear : the sound from the drone cannot cause auditory discomfort to the residents.
-
Support concurrent detection and identification : in the same airspace, there may be multiple drones taking off and landing, that is, multiple drones may transmit acoustic pulses at the same time. This requires the system to be able to detect the pulses of each UAV separately from the conflicting acoustic signals and identify which UAV each detected pulse belongs to.
-
Security : Ability to prevent malicious attackers from forging the drone's pulses to mislead the system.
In order to meet the above requirements, this system uses pseudo-random noise (Pseudo-Random Noise, PRN) modulation to generate the acoustic pulse of the drone. We use each drone's identification number (ID) to set a pseudo-random seed and generate a series of N Gaussian random variables as the pulse signal sent by the drone. In a specific implementation, the code rate is equal to the sampling rate of the speaker, that is, 48 kHz.
3.2 Pulse detection
After debugging and sending the acoustic signal, we need to perform pulse detection on the ground side.
We chose a matched filter for pulse detection. The idea is to template the transmitted pulse and correlate it with the received signal. By streaming the received signal into the matched filter, the matched filter will stream the correlation result. If an obvious correlation peak is found from the output, we judge that the pulse has been detected. However, in order to solve the problem of low signal-to-noise ratio, it is necessary to increase the pulse length, but the existence of Doppler distortion makes it counterproductive to increase the pulse length.
A straightforward and effective solution to this problem is to compensate for Doppler distortion: the Doppler effect scales the duration of the pulse codeword. If the radial velocity of the UAV relative to the microphone is known, we can calculate the actual duration of the codeword and use this parameter to resample the original PRN pulse template to generate a pulse that is synchronized with the received pulse codeword Signal template. It can be expected that using the new template to detect the template of the PRN pulse compensates the interference of the Doppler effect, and then the length of the PRN pulse can be increased as required to overcome the problem of low signal-to-noise ratio.
In practice, however, the radial velocity of the microphone is unknown. Therefore, we use a linear search to traverse the radial velocity of the microphone. For the set of possible UAV speeds we sequentially perform resampling and correlation calculations. When all the speeds in the set complete the above operations, we get the corresponding N correlation functions. From these, we only need to keep the one correlation function with the largest correlation value. This is because the largest correlation value means that the Doppler distortion of the pulse has been compensated to the greatest extent, and the corresponding search speed is also the closest to the real radial speed of the UAV.
3.3 TDoA Estimation and Location
The system deploys four microphones in total, which are recorded as Mic0, Mic1, Mic2, and Mic3 clockwise.
For the sound source of each microphone Mici, we perform the above-mentioned Doppler velocity compensation and pulse detection in sequence, and obtain the correlation function. From the correlation function, the correlation peak is found, and the position of the correlation peak is regarded as the arrival time (Time of Arrial, ToA) of the pulse signal to Mici. For convenience, record the arrival time of the pulse signal to the four microphones Mic0, Mic1, Mic2 and Mic3 as ToA0, ToA1, ToA2 and ToA3 respectively.
Next, we calculate the relative delay of arrival (Time Difference of Arrial, TDoA) of the pulse. In this system, we only calculate the TDoA of the diagonal microphone pair, that is, the microphone pair <Mic0,Mic2> and <Mic1,Mic3>. This is because geometrically, diagonal microphones have the largest microphone separation (i.e., aperture), and therefore the finest spatial granularity.
Two TDoAs will be transmitted to the drone via WiFi. Based on this information, the UAV can establish two systems of hyperboloid equations. According to the hyperboloid equations and the height information of the UAV, the information of the remaining two degrees of freedom of the aircraft can be obtained.
4 innovation points
Compared with the visual positioning scheme, this scheme has the following advantages:
-
Because the propagation of the acoustic signal is not affected by the lighting conditions, the work of the system is not affected by the lighting conditions;
-
The acoustic signal is radiated to the whole space, so the horizontal positioning range of the system is larger;
-
In theory, it also supports multiple drone positioning.
At present, this technical solution has been connected to the flight control system of Meituan UAV, and has passed the test and verification of various complex environments. In the future, it plans to apply it in Meituan's unmanned delivery business.
The booming drone industry provides a more efficient distribution solution for the new instant retail format that is currently poised to explode. The application of cutting-edge technologies in real scenarios can further promote the development of technologies. According to the data of the Shenzhen Municipal Bureau of Transportation, Meituan UAVs have opened 11 normalized test routes in urban scenarios, completed more than 70,000 real orders, and the total delivery flight time has exceeded 10,000 hours, with a total of more than 400,000 sorties tested .
Mao Niannian, head of Meituan’s drone business, said: “On the whole, urban low-altitude terminal logistics has the characteristics of high efficiency, stability, and high quality, which can better meet the new needs of the current market for the development of the real-time logistics industry. The main reason for promoting the rapid landing of related industries in the past two years.”
Related information and instructions
-
MicNest：MicNest
-
The paper was jointly completed by researchers from Tsinghua University, Meituan and Politecnico di Milano, and the research work was supported by Tsinghua University-Meituan Digital Life Joint Research Institute and the National Natural Science Foundation of China.
---------- END ----------
Meituan scientific research cooperation
Meituan's scientific research cooperation is committed to building a bridge and platform for cooperation between Meituan's technical team and universities, scientific research institutions, and think tanks. Relying on Meituan's rich business scenarios, data resources, and real industrial problems, open innovation, gather upward forces, and focus on robots , artificial intelligence, big data, Internet of Things, unmanned driving, operational optimization and other fields, jointly explore cutting-edge technology and industry focus macro issues, promote industry-university-research cooperation and exchange and achievement transformation, and promote the cultivation of outstanding talents. Facing the future, we look forward to cooperating with more teachers and students from universities and research institutes. Teachers and students are welcome to send emails to: [email protected].
maybe you want to see
| ICRA 2022 Best Paper Released: Meituan UAV Team Wins the Only Best Navigation Paper Award
read more
Frontend | Algorithms | Backend | Data
Security | Android | iOS | Operation and Maintenance | Testing