Multi-sensor time synchronization method for autonomous driving vehicles based on C-ATS algorithm
By adopting the time synchronization method based on the C-ATS algorithm in autonomous vehicles, and using improved weight coefficients to synchronize sensor clocks, the problem of insufficient time synchronization accuracy of multi-sensors is solved, high-precision time synchronization is achieved, and autonomous driving is ensured. Vehicle information fusion accuracy and traffic safety.
Patent Information
- Application Number
- CN202311124983.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-03
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2043-09-03
AI Technical Summary
Multi-sensor time synchronization in autonomous driving vehicles has insufficient time synchronization accuracy due to asymmetric random communication time delay, which cannot meet the requirements of autonomous driving vehicles for sensor time synchronization, which may lead to incorrect information fusion and driving decisions, affecting traffic safety.
A multi-sensor time synchronization method for autonomous driving vehicles based on C-ATS algorithm was designed. Through the bidirectional time synchronization mechanism between the time server and the sensor, the improved weight coefficient is used to synchronize the sensor clock to overcome asymmetric random communication. The impact of time delay on time synchronization accuracy.
It realizes high-precision sensor time synchronization, meets the requirements of autonomous driving vehicles for time synchronization, and through simulation verification, it shows that the C-ATS algorithm can achieve higher time synchronization accuracy under asymmetric network delay conditions, reducing time deviation and relative time variance of deviation.
Smart Images

Figure CN117155506B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of sensor time synchronization, and is a method for sensor time synchronization for realizing information perception of an autonomous driving vehicle, namely, a multi-sensor time synchronization method for an autonomous driving vehicle based on a C-ATS algorithm. Background Art
[0002] With the development of current assisted autonomous driving and future unmanned driving technology, autonomous driving vehicles have entered the stage of commercial pilot implementation, and the technical requirements of autonomous driving vehicles are constantly improving. The working process of autonomous driving vehicles can be divided into four parts: environmental perception, behavioral decision-making, path planning and automatic control. Among them, environmental perception is a prerequisite and a crucial link for achieving safe and stable driving of autonomous driving. Autonomous driving vehicles integrate multiple sensors accurately through the complementary advantages and mutual fusion of multiple laser radars, monocular (binocular) cameras and millimeter wave radars, so as to achieve stable information acquisition and analysis of the autonomous driving system. However, if the domain controller and sensors in autonomous driving use built-in hardware clocks, there will also be deviations in different hardware times. If the time of each sensor cannot be synchronized, information fusion cannot be carried out at all. For example, when the time of the laser radar and the camera is not synchronized, the point cloud data in front of the laser radar is incorrectly fused with the image data of the camera, and the decision unit will make a wrong judgment, which will affect the current driving status, and further serious traffic accidents may occur.
[0003] At present, scholars at home and abroad have paid attention to the time synchronization problem of sensors in autonomous vehicles and studied the time synchronization scheme of sensors through time interaction. However, since multiple sensors such as lidar, monocular (binocular) cameras and millimeter-wave radar are located in the local area network of the car, there will be uncontrollable random network delays when each sensor is communicating synchronously, resulting in asymmetric network delays in the sensor sending and receiving time information links, causing time synchronization divergence and large time deviation, which cannot meet the requirements of autonomous vehicles for sensor time synchronization accuracy. Summary of the invention
[0004] The technical problem to be solved by the present invention is to provide a multi-sensor time synchronization method for autonomous driving vehicles based on the C-ATS algorithm in response to the above-mentioned deficiencies in the prior art. The present invention designs a C-ATS algorithm based on ATS improved weights to coordinately synchronize the time server and the sensor clock of the autonomous driving vehicle, overcome the influence of asymmetric random communication time delay on the time synchronization accuracy, and achieve higher time synchronization accuracy for multiple sensors.
[0005] In order to achieve the above technical objectives, the technical solution adopted by the present invention is:
[0006] A multi-sensor time synchronization method for autonomous driving vehicles based on the C-ATS algorithm, wherein the autonomous driving vehicle is provided with a time server for providing a reference time and a plurality of sensors; the time server and each sensor interact with each other by periodically sending messages;
[0007] The time synchronization methods specifically include:
[0008] Step 1. Sensor c i Broadcast a data packet in the kth round of interaction, i≤n-1, n∈N + , sensor c i The time to send this packet is The time the time server receives the data packet is After receiving the data packet, the time server will Passed to sensor c i , sensor c i The time when the data packet sent by the time server is received is
[0009] Step 2, sensor c i Record the local clock of the time server sending the current broadcast message and sensor c i The time when the data packet sent by the time server is received
[0010] Step 3: When k=1, the preset drift correction coefficient and offset correction factor Value versus time Make corrections and get
[0011] Step 4: For any ε>0, if Then the time synchronization is completed, otherwise, go to step 5;
[0012] Step 5: Assign to
[0013] Step 6, k=k+1; execute step 1 and step 2; calculate the filter proportional factor according to the C-ATS algorithm, and calculate the drift correction coefficient through the filter proportional factor and offset correction factor
[0014] Step 7: Use the drift correction factor and offset correction factor Local time for the sensor Make corrections and get
[0015] Step 8. For any ε>0, if Then the time synchronization is completed, otherwise, return to step 5.
[0016] As a further improved technical solution of the present invention, in step 3,
[0017] As a further improved technical solution of the present invention, the filtering scale factor for:
[0018]
[0019] Among them, ρ η is the weight coefficient of the filter scale factor.
[0020] As a further improved technical solution of the present invention, the drift correction coefficient for:
[0021]
[0022] Among them, ρ α For sensor c i Corrects the drift weight of the local time estimate.
[0023] As a further improved technical solution of the present invention, the offset correction coefficient for:
[0024]
[0025] Among them, ρ o For sensor c i The offset weight factor used to correct the local time estimate.
[0026] As a further improved technical solution of the present invention, by using the drift correction coefficient and offset correction factor Local time for the sensor Make corrections and get Specifically:
[0027] The beneficial effects of the present invention are:
[0028] Based on the sender-receiver bidirectional time synchronization mechanism, the present invention establishes a time interaction synchronization model between the time server and the sensor, designs a C-ATS algorithm based on the ATS improved weight, and coordinates the synchronization of the time server and the sensor clock of the autonomous driving vehicle, thereby overcoming the influence of asymmetric random communication time delay on the time synchronization accuracy and meeting the requirements of the autonomous driving vehicle for the sensor time synchronization accuracy. And through simulation experiments, it is proved that the C-ATS algorithm can achieve higher time synchronization accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] Figure 1 It is a diagram of collaborative interaction between time server and sensor.
[0030] Figure 2 It is a time synchronization model diagram of time server and sensor.
[0031] Figure 3 It is a sender-receiver two-way time synchronization diagram.
[0032] Figure 4 It is the time deviation graph of ATS algorithm.
[0033] Figure 5 It is the time deviation graph of the WMTS algorithm.
[0034] Figure 6 It is the time deviation graph of LSTS and C-ATS algorithms.
[0035] Figure 7 It is the relative time deviation variance graph of the ATS algorithm.
[0036] Figure 8 It is the relative time deviation variance map of the WMTS algorithm.
[0037] Fig. 9 It is the relative time deviation variance plot of LSTS and C-ATS algorithms.
[0038] Fig.10 It is a flow chart of the synchronization method of the present invention. DETAILED DESCRIPTION
[0039] The specific embodiments of the present invention are further described below according to the accompanying drawings:
[0040] This embodiment provides a multi-sensor time synchronization method for autonomous driving vehicles based on the C-ATS algorithm. This embodiment is based on the sender-receiver bidirectional time synchronization mechanism, establishes a time interaction synchronization model between the time server and the sensor, designs a C-ATS algorithm based on the ATS improved weight, and coordinates the synchronization of the time server and the autonomous driving vehicle sensor clock to overcome the impact of asymmetric random communication time delay on the time synchronization accuracy. In addition, the time synchronization convergence characteristics of the C-ATS algorithm in the presence of asymmetric random communication time delay are mathematically derived, and simulation experiments are used to prove that the C-ATS algorithm can achieve higher time synchronization accuracy. The sensors on the autonomous driving vehicles of this embodiment are intelligent sensors, which have the functions of collecting data, processing data, exchanging data with the outside world, and changing the operation of the sensors through software control according to actual needs, thereby realizing intelligence and networking.
[0041] 1. Model establishment:
[0042] There is a time server and several sensors on an autonomous vehicle. The time server can provide a reference time for time synchronization so that the sensors and the time server can synchronize time. At the same time, the time server and the sensor periodically send messages through UDP broadcast to complete the interaction. The message contains the local time of the sending node. Figure 1 As shown, G = (V, E) is used to describe the network topology between the time server and the sensor, where V = {c i |i≤n-1,n∈N +} represents all sensors within the coverage area of the time server, and the edge set E = {(i, j)|i≤n,j≤n,i,j∈N +} represents the communication between nodes in the network.
[0043] This paper adopts the typical linear clock model (Formula (1)), defines t as real time, and uses sensor c as i For example, its local clock reading τ i The relationship with the real time t can be expressed as:
[0044] η i =α i t+β i (1);
[0045] where α i is the local clock drift, β i is the local clock deviation. Since the time synchronization of sensors on autonomous vehicles only needs to achieve relative synchronization, rather than absolute synchronization, the local time of the time server can be considered as absolute time. The relationship between the clock reading of the time server and the real time is τ j =t.
[0046] The time model of the time server and sensor is as follows Figure 2 As shown in the figure, β represents the relative clock offset of the sensor, and α represents the relative clock frequency drift of the two nodes. Clock synchronization is to estimate α and β so that the clocks of the sensor and the time server are consistent. If the clocks of the two nodes are completely consistent, β = 0 and α = 1. Since clock offset and drift are an objective situation that always exists in actual situations, it is usually necessary to periodically synchronize the clocks of sensors to ensure their time consistency with the time server.
[0047] The time server and the sensor are based on a sender-receiver bidirectional time synchronization mechanism such as Figure 3 As shown, the straight line c in the figure i is the sensor's clock, the time server O's clock is a ray starting from the origin, and the slope of the line is 1. Assume that the clock update rate of the sensor is α i , and the sensor has β with the clock of the time server at the initial time i The initial deviation is expressed as t k represents the real clock of each round, k is the round of interaction between the sensor and the time server, then the clock of the sensor in each round can be expressed as the formula: The clock of the time server can represent the real clock, that is, each round of update is 1 second, which can be expressed as Assume that the sensor broadcasts a data packet about its own local time in the kth round of interaction, and the time when the sensor sends the data packet is Due to random network delays, assume that the time the time server receives the packet is After receiving the data packet, the time server will Passed to the sensor, due to random network delays, the time when the sensor receives the data packet sent by the time server This completes a complete interaction process.
[0048] Each sensor periodically broadcasts messages to interact with the time server and estimates and adjusts its own local time based on the local time readings received from other nodes. i There is formula (2):
[0049]
[0050] in For sensor c i The logical clock reading in the kth broadcast cycle (i.e. the final time after synchronization), is the sensor c in the kth broadcast cycle i Correction factor for the drift rate of its own local clock, is the sensor c in the kth broadcast cycle i Correction factor for the deviation of the local clock.
[0051] For any ε>0, and the sensor and time server nodes i and j can undergo a finite number of interaction iterations (the number of interaction iterations is less than or equal to the bounded value ) and then satisfy formula (3), the time synchronization of the autonomous driving vehicle sensors is considered to be completed:
[0052]
[0053] 2. C-ATS algorithm:
[0054] Considering that the ATS algorithm cannot achieve accurate time synchronization in the presence of asymmetric communication delay, this paper proposes a time synchronization algorithm suitable for asymmetric communication networks. In the process of executing time synchronization between the time server and the sensor, suppose that in the kth round of interaction, the time server node j receives the time from the sensor node c. i The time server first sends the relevant sensor node c in the message i The time data information is The form is stored in the cache, where: The kth round of time server node j receives the data from sensor node c i The local time estimate at the time of the message; is the kth round sensor node c i The local time estimate when sending the message. After receiving the k+1 round message, the time server uses the new data received in the current k+1 round Replace the data received during the kth interaction, and cycle in sequence. Sensor c i Update its own local time period Local time update period with time server For comparison, use the filter scale factor To characterize the estimated ratio of the deviation between each other's local clocks after each round, where ρ η represents the weight coefficient. After that, the sensor c i The filtering ratio factor will be based on each broadcast cycle The value is based on the local logical clock The correction factor of the local logical clock is adjusted, including the drift rate correction factor of the local logical clock and deviation correction factor The specific synchronization algorithm is expressed as follows, where ρ α For sensor c iCorrect the drift weight coefficient of the local time estimate, ρ o For sensor c i The offset weight factor used to correct the local time estimate.
[0055] like Fig.10 As shown, the C-ATS time synchronization algorithm:
[0056] Step 1. Sensor c i In the kth round of interaction, a data packet is broadcast via UDP, i≤n-1,n∈N + , sensor c i The time to send this packet is The time the time server receives the data packet is After receiving the data packet, the time server will Through UDP broadcast, it is then transmitted to the sensor c i , sensor c i The time when the data packet sent by the time server is received is
[0057] Step 2, sensor c i exist The time server is received at The broadcast message sent at the moment; extract the timestamp from the broadcast message and record the local clock of the time server sending the current broadcast message and sensor c i The time when the data packet sent by the time server is received
[0058] Step 3: When k=1, the preset drift correction coefficient and offset correction factor Value versus time Make corrections and get The correction formula is:
[0059] Step 4: For any ε>0, if Then the time synchronization is completed, otherwise, go to step 5;
[0060] Step 5: Assign to
[0061] Step 6, k=k+1; execute step 1 and step 2;
[0062] Calculate the filter scaling factor based on the C-ATS algorithm
[0063]
[0064] Calculate the drift correction factor by filtering the proportional factor and offset correction factor
[0065]
[0066]
[0067] Step 7: Use the drift correction factor and offset correction factor Local time for the sensor Make corrections and get
[0068] Step 8. For any ε>0, if Then the multi-sensor time synchronization is completed, otherwise, return to step 5.
[0069] After completing time synchronization, the final Denote as the estimated time after synchronization. are all preset empirical values. In this embodiment, Take 1, Take 1, Take 0.
[0070] 3. Synchronization performance analysis of C-ATS algorithm:
[0071] In order to prove that the C-ATS algorithm can suppress the time error divergence caused by asymmetric network delay, the present invention uses the logical clock deviation expression in the kth broadcast cycle: Convergence is achieved for sensor c i The proof of precise time synchronization. Decompose it into the sum of two sub-terms. By proving that both sub-terms are bounded or convergent, we can get the expression This embodiment can prove that is convergent, we can deduce that sensor c i The local time estimate meets the convergence requirement.
[0072] 4. Simulation analysis:
[0073] In order to analyze and verify the influence of different time synchronization algorithms on the convergence and divergence of time synchronization errors under asymmetric link delay, Matlab is used to simulate and verify the time synchronization algorithm of the sensor. The model is set up as a distributed network including a time server node and a sensor node. The nodes send time messages in data frames through UDP broadcast with a period of 100 milliseconds. For any sensor node c i , its actual sending cycle can be expressed as Where T represents the time update period in absolute time.
[0074] The present invention selects the ATS algorithm, WMTS algorithm, LSTS algorithm and C-ATS algorithm, and firstly simulates and compares the synchronization performance of different time synchronization algorithms, and the parameter settings are shown in Table 1.
[0075] Table 1, time synchronization algorithm parameter table:
[0076] ATS WMTS LSTS C-ATS <![CDATA[ρ η ]]> 0.5 1 / (1+k) <![CDATA[1 / (1+…+k 2 )]]> 1 / k <![CDATA[ρ α ]]> 0.5 1 <![CDATA[1 / (1+k) 0.6 ]]> 1 / k <![CDATA[ρ o ]]> 0.5 1 0.4 0.5
[0077] Taking into account various factors such as the manufacturing process of the vehicle and environmental impact, the typical error of the sensor node clock crystal oscillator is [10,100]ppm, that is, the clock offset generated by the clock in every million seconds, and the actual sensor node clock crystal oscillator is usually different due to the manufacturing process. In the simulation, the initial value of the hardware clock frequency deviation of each node is randomly selected in the interval [0.9999,1.0001], and the initial value of the phase deviation is selected in the interval [0,0.0002]. Therefore, the simulation part of the present invention sets the sensor clock drift slope α i =0.9999, clock offset deviation o i =0.0002, and a random communication delay in the interval (0, 0.0005) that obeys a uniform distribution is selected for the experiment.
[0078] To measure the time synchronization effect between the time server and the sensor, the time deviation d is defined k and the relative time deviation variance R k The meaning of each indicator is as follows:
[0079] Time deviation k : The size of the adjusted logical time deviation between the sensor and the time server after any round of communication, which is used to measure the effect of clock synchronization between the sensor and the time server. k The smaller it is, the better the synchronization effect is. It can be expressed as follows: Indicates the logical time of the time server, which is different from the local clock Consistent, Indicates the logic time after sensor adjustment.
[0080] Relative time deviation variance R k : Logic time after sensor adjustment With absolute time t k The change trend of the difference between k The smaller the value, the closer the logic time is to the sensor adjustment. With absolute time t k The smaller the difference between the two, the faster the sensor's logic clock The more stable it is, the more it tends to converge. It can be expressed as follows:
[0081]
[0082] The ATS, WMTS, LSTS and C-ATS algorithms were used to simulate the sensor and time server for 50 seconds, that is, 500 interactions. The simulation results are as follows:
[0083] Depend on Figure 4 and Figure 5 It can be seen that when using the ATS algorithm and the WMTS algorithm, the time deviation cannot converge when there is an asymmetric time delay. When using the ATS algorithm, after the interaction begins, as the number of interactions increases, the time deviation gradually increases and diverges. In 500 interactions, the maximum time deviation is 55.57ms; when using the WMTS algorithm, the time deviation continues to oscillate in the range of (-0.6ms to 0.4ms). Figure 6 It can be seen that in the presence of asymmetric random communication time delay, the LSTS algorithm and the C-ATS algorithm can achieve time synchronization, and the time deviation does not diverge. Both the LSTS algorithm and the C-ATS algorithm converge quickly after the start of the simulation. The LSTS algorithm converges after about 40 interactions, and the time deviation stabilizes at 0.34ms. The time deviation of the C-ATS algorithm decreases rapidly within 20 interactions, and there is a small vibration within 100 interactions, but the maximum time deviation remains within 0.20ms, and then the time deviation converges to 0.19ms. It can be seen that the C-ATS algorithm can achieve the convergence of synchronization time deviation compared with the ATS algorithm and the WMTS algorithm, and in the most ideal case in the simulation environment set in this article, the time synchronization accuracy is improved by 0.15ms compared with the LSTS algorithm, and the time deviation is reduced by 44.12%.
[0084] Depend on Figure 7 It can be seen that in the presence of asymmetric random communication time delay, using the ATS algorithm, the relative time deviation variance gradually increases in the form of oscillation divergence. When the interaction iteration is 460-470 times, the relative time deviation variance reaches a maximum value of 733.61ms. 2 ;Depend on Figure 8 It can be seen that when there is an asymmetric random communication time delay, the relative time deviation variance does not diverge like when the ATS algorithm is used, but it is still in (0~0.08ms 2 ) range. Fig. 9 It can be seen that when using the LSTS and C-ATS algorithms, the relative time deviation variance does not diverge and converges to a certain value, which is 0.034ms respectively. 2 and 0.0099ms 2Therefore, it can be concluded that using the C-ATS algorithm for time synchronization can make the sensor's logical clock more stable, and in the most ideal case in the simulation environment set in this article, compared with the LSTS algorithm, the stability of the logical clock is improved by 70.88%.
[0085] The protection scope of the present invention includes but is not limited to the above embodiments. The protection scope of the present invention shall be based on the claims. Any replacement, deformation, and improvement of the technology that can be easily thought of by technicians in this field shall fall within the protection scope of the present invention.
Claims
1. Multi-sensor time synchronization method for autonomous driving vehicles based on C-ATS algorithm, It is characterized in that The autonomous driving vehicle is equipped with a time server and several sensors for providing reference time; the time server interacts with each sensor by periodically sending messages; The time synchronization methods specifically include: Step 1. Sensor In the interactive broadcast a data packet, ,sensor The time to send this packet is , the time server receives the data packet is After receiving the data packet, the time server will change its local time Passed to sensor ,sensor The time when the data packet sent by the time server is received is ; Step 2: Sensor Record the local clock of the time server sending the current broadcast message and sensors The time when the data packet sent by the time server is received ; Step 3: When k=1, the preset drift correction coefficient and offset correction factor Value versus time Make corrections and get ; Step 4: For any ,like , then time synchronization is completed, otherwise, go to step 5; Step 5: Assign to ; Step 6: Reassign the value of k+1 to k; execute steps 1 and 2; calculate the filter proportional factor according to the C-ATS algorithm, and calculate the drift correction coefficient through the filter proportional factor and offset correction factor ; Step 7: Use the drift correction factor and offset correction factor Local time for the sensor Make corrections and get ; Step 8: For any ,like , then time synchronization is completed, otherwise, return to step 5.
2. The multi-sensor time synchronization method for an autonomous driving vehicle based on the C-ATS algorithm according to claim 1, Features: In the step 3, .
3. The multi-sensor time synchronization method for an autonomous driving vehicle based on the C-ATS algorithm according to claim 1, Features: The filter scaling factor for: , ; in, is the weight coefficient of the filter scale factor.
4. The multi-sensor time synchronization method for an autonomous driving vehicle based on the C-ATS algorithm according to claim 3, Features: The drift correction factor for: , ; in, For sensor Corrects the drift weight of the local time estimate.
5. The multi-sensor time synchronization method for an autonomous driving vehicle based on the C-ATS algorithm according to claim 4, Features: The offset correction factor for: ; in, For sensor The offset weight factor used to correct the local time estimate.
6. The multi-sensor time synchronization method for an autonomous driving vehicle based on the C-ATS algorithm according to claim 1, Features: By using the drift correction factor and offset correction factor Local time for the sensor Make corrections and get , specifically: .
