An intelligent driving SR path jump processing method based on multi-hypothesis trajectory rendering

Abstract

The present application relates to a kind of intelligent driving SR path jump processing method based on multi-hypothesis trajectory rendering, it is applied to automatic driving vehicle, the method includes: step 1, obtains K candidate trajectories of ego vehicle and corresponding confidence;Step 2, according to confidence, map K trajectories to different confidence buckets, and render in different style in SR interface;Step 3, determine transition duration, when the confidence of a certain trajectory crosses the bucket boundary, SR interface changes its style smoothly in transition duration by easing interpolation;Step 4, when a certain trajectory is mapped into the bucket with the highest confidence, start timing, and the timing duration is Tlead=k*v+b;Step 5, if timing ends and user does not take over, then lock the trajectory, issue execution request, so that the vehicle travels according to the locked trajectory.The present application solves the problem that the SR interface in the prior art only displays a single determined path, resulting in no warning when the path jumps, poor user experience and large safety hazards.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, and in particular to an intelligent driving SR path transition processing method based on multi-hypothesis trajectory rendering. Background Technology

[0002] Current mainstream intelligent driving systems (such as ACC, LCC, and city / highway NOA) typically include a perception layer, a prediction layer, a planning and control layer, and an execution layer. Surrounding Reality (SR) display technology: Currently, in-vehicle SR interfaces are primarily responsible for visualizing the perceived environment (such as lane lines, vehicles, and pedestrians) and the unique, definite trajectory output by the planning and control layer. This is usually done using a "light carpet" format to display the vehicle's upcoming path. In practical applications, due to the unpredictable behavior of complex traffic participants (such as sudden braking by the vehicle in front, cut-in by the vehicle beside, and irregularly shaped obstacles), the prediction and planning and control layers frequently correct the trajectory within milliseconds. Summary of the Invention

[0003] In view of the above problems, the present invention provides a method for intelligent driving SR path transition processing based on multi-hypothesis trajectory rendering, so as to solve the problem that the SR interface in the prior art only displays a single determined path, resulting in no warning when the path transitions, poor user experience and great safety risks.

[0004] This invention provides a method for handling path transitions in intelligent driving SR (Survey Response) based on multi-hypothesis trajectory rendering. The method includes: Step 1, obtaining K candidate trajectories for autonomous vehicles and their corresponding confidence levels; Step 2, mapping the K trajectories to different confidence level buckets based on their confidence levels, and rendering them in different styles on the SR interface; Step 3, determining the transition duration, and when the confidence level of a trajectory crosses the bucket boundary, the SR interface smoothly changes its style within the transition duration through easing interpolation; Step 4, when a trajectory is mapped to the bucket with the highest confidence level, starting a timer for a duration of T. lead =k*v+b, where v is the vehicle speed, k is the calibration coefficient, and b is the lower limit of the transition time; Step 5, if the timing ends and the user has not taken over, then lock the trajectory and send an execution request to make the vehicle travel according to the locked trajectory.

[0005] Furthermore, K is 3-5, k=0.008s / (km / h), b=0.3s.

[0006] Furthermore, step 1 includes: step 11, based on the current environmental perception results, outputting K future driving trajectories of the vehicle with the highest probability through forward inference of a neural network; step 12, acquiring perception data and vehicle status, and calculating the posterior probability of each candidate trajectory through a Softmax classification layer or regression head; step 13, quantizing the posterior probability into a confidence value of 0-100%, and mapping the geometric coordinate point sequence of the K trajectories to the confidence value for later decision-making and reality.

[0007] Furthermore, step 2 includes: step 21, pre-setting several confidence buckets and dynamically adjusting the bucket boundaries through remote wireless upgrade technology; step 22, storing the interval boundaries of the confidence buckets and the SR rendering style parameters bound to each bucket in a configurable parameter table on the vehicle side; step 23, dynamically distributing and updating through remote wireless upgrade technology.

[0008] Furthermore, the method for determining the transition duration includes: step 31, determining the lower limit of the transition duration based on the human peripheral vision perception threshold and the SR rendering frame rate; step 32, determining the upper limit of the transition duration based on the confidence refresh cycle; and step 33, dynamically selecting the transition duration between the upper and lower limits based on the vehicle model and the current scene.

[0009] Furthermore, the lower limit of the transition duration is 0.3s, and the upper limit of the transition duration is 1.2s.

[0010] Furthermore, the method for the SR interface to smoothly change its style within the transition time by easing interpolation when the confidence of a certain trajectory crosses the bucket boundary includes: Step 34, when the confidence increases, proceed to Step 35, when the confidence decreases, proceed to Step 36; Step 35, according to the determined transition time, the ease-in-out micro-motion effect of the SR interface is realized through the easing function to enhance the user's sense of confirmation; Step 36, after the confidence falls below the bucket boundary and is maintained for more than 3 consecutive frames, the ease-out micro-motion effect of the SR interface is realized through the easing function according to the determined transition time, supplemented by a brief flash to indicate the cancellation of the schematic diagram, to avoid the residual "credible state" of the light carpet misleading the user.

[0011] Furthermore, the easing function employs easeOutCubic cubic easing to simulate the "asymptotic stability" process of human cognition.

[0012] Furthermore, step 5 also includes: if the timer has not ended, and the user presses the pedal or applies steering torque, then proceed to step 1.

[0013] Furthermore, step 5 also includes: if the timing has not ended, and the trajectory confidence falls outside the boundary of the bucket with the highest confidence and is lower than the preset hysteresis window, then proceed to step 1.

[0014] This invention provides a method for handling path transitions in intelligent driving based on multi-hypothesis trajectory rendering. It primarily addresses existing technologies that suffer from a lack of transition during visual transitions, leading to strong startle reflexes and reduced trust in the intelligent driving system. Furthermore, because the SR is strongly bound to the actuator, visual transitions are often accompanied by actual sudden braking or sharp steering, resulting in abrupt execution that causes not only occupant discomfort but can also lead to rear-end collisions on highways. Additionally, users cannot distinguish between "system is hesitating" and "system has made a decision" on the interface, resulting in extremely high cognitive load during human-machine co-driving. Attached Figure Description

[0015] Figure 1 A flowchart of an intelligent driving SR path jump processing method based on multi-hypothesis trajectory rendering provided by the present invention; Figure 2 This invention provides a flowchart of a method for obtaining K candidate trajectories of a vehicle and their corresponding confidence levels. Figure 3 This is a flowchart of a method for mapping K trajectories to buckets with different confidence levels, provided by the present invention. Figure 4 This is a flowchart of a method for determining transition duration and smoothly changing style provided by the present invention. Detailed Implementation

[0016] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0017] Example 1: This application provides a method for handling path transitions in intelligent driving SR based on multi-hypothesis trajectory rendering, such as Figure 1 As shown, the method includes: Step 1: Obtain K candidate trajectories for autonomous vehicles and their corresponding confidence levels; Step 2: Map the K trajectories to different confidence buckets based on their confidence levels, and render them in the SR interface in different styles; Step 3: Determine the transition duration. When the confidence of a certain trajectory crosses the bucket boundary, the SR interface smoothly changes its style within the transition duration through easing interpolation. Step 4: When a trajectory is mapped to the bucket with the highest confidence, start timing for a duration of T. lead =k*v+b, where v is the vehicle speed, k is the calibration coefficient, and b is the lower limit of the transition time; Step 5: If the timer ends and the user has not taken over, lock the trajectory and send an execution request to make the vehicle travel along the locked trajectory.

[0018] This application provides a method for handling path jumps in intelligent driving (SR) based on multi-hypothesis trajectory rendering. It addresses sudden path jumps caused by uncertainty in perception and control in intelligent driving (AD / ADAS / NOA) scenarios, improving driving safety and user experience through multi-hypothesis trajectory visualization and intent buffering mechanisms. This method solves the problems of "visual impact of path jumps" and "abrupt execution" not covered by existing technologies by combining a combination of "multi-hypothesis access at the prediction layer + confidence-based bucket rendering + easing interpolation + execution delay".

[0019] Example 2: This application provides a method for handling path transitions in intelligent driving SR based on multi-hypothesis trajectory rendering, such as Figure 1 As shown, the method includes: Step 1: Obtain K candidate trajectories for autonomous vehicles and their corresponding confidence levels; like Figure 2 As shown, step 1 includes: Step 11: Based on the current environmental perception results, output the K most probable future driving trajectories of the vehicle through forward inference of the neural network; Based on the current environmental perception results, the system uses neural network forward inference to output a set of K highest-probability (optimal) future driving trajectories for the vehicle. In this scheme, K takes values ​​from 3 to 5 (preferably K=3), representing the primary trajectory (highest probability), the secondary trajectory (second highest probability), and the third alternative trajectory, respectively. The prediction layer does not only provide the control with one trajectory, but also passes ≥3 candidate trajectories for the vehicle and their confidence levels separately to the SR rendering layer.

[0020] Step 12: Obtain perception data and vehicle status, and calculate the posterior probability of each candidate trajectory through a Softmax classification layer or regression head. Step 13: Quantize the posterior probability into a confidence level value of 0-100%, and map the geometric coordinates of the K trajectories to the confidence level values ​​for later decision-making and reality.

[0021] The confidence score is typically output directly from the prediction layer model end-to-end. The specific process is as follows: Model input: fusing perception data (obstacle positions, lane line structures, road edge information) and vehicle status (vehicle speed, acceleration, steering wheel angle); Probability calculation: the model internally calculates the posterior probability of each candidate trajectory through a Softmax classification layer or regression head; Numerical mapping: multiplying the probability by 100 to quantize it into a confidence score value of 0-100%; Synchronous transmission: the prediction layer sends the geometric coordinate sequence of these K trajectories and their corresponding confidence scores in parallel to the planning and control module (for decision-making) and the SR rendering module (for display).

[0022] Step 2: Map the K trajectories to different confidence buckets based on their confidence levels, and render them in the SR interface in different styles; The trajectory is mapped to different "confidence buckets" based on the confidence level, and rendered in the SR interface with different styles (dashed / solid / color / transparency), such as... Figure 3 As shown, step 2 includes: Step 21: Pre-set several confidence buckets and dynamically adjust the bucket boundaries using remote wireless upgrade technology; Typically, four confidence buckets can be preset, and the bucket boundaries can be dynamically adjusted via OTA (Over-the-Air, which allows device manufacturers and software developers to update and manage device software online without requiring physical contact with the user), as shown in Table 1.

[0023]

[0024] Table 1 Confidence Bucketing and Style Mapping Table Step 22: The interval boundaries of the confidence buckets and the SR rendering style parameters bound to each bucket are stored in the vehicle-side configurable parameter table. Step 23: Dynamically distribute updates using remote wireless upgrade technology.

[0025] The interval boundaries of the confidence buckets and the SR rendering style parameters (color, line type, transparency, light carpet width, interpolation duration, etc.) bound to each bucket are stored in a configurable parameter table on the vehicle side. These parameters can be dynamically updated from the cloud via OTA (Over-the-Air) without requiring a system firmware upgrade. For example, if the confidence distribution shifts to the right after prediction model iteration, the lower edge of the high-confidence bucket can be lowered from 70% to 60% via OTA, ensuring that SR rendering still presents a "high-confidence style" for trajectories of the same level, maintaining a consistent user experience.

[0026] Step 3: Determine the transition duration. When the confidence of a certain trajectory crosses the bucket boundary, the SR interface smoothly changes its style within the transition duration through easing interpolation. Step 4: When a trajectory is mapped to the bucket with the highest confidence, start timing for a duration of T. lead =k*v+b, where v is the vehicle speed, k is the calibration coefficient, and b is the lower limit of the transition time; Step 5: If the timer ends and the user has not taken over, lock the trajectory and send an execution request to make the vehicle travel along the locked trajectory.

[0027] This application provides a method for handling path jumps in intelligent driving (SR) based on multi-hypothesis trajectory rendering. It addresses sudden path jumps caused by uncertainty in perception and control in intelligent driving (AD / ADAS / NOA) scenarios, improving driving safety and user experience through multi-hypothesis trajectory visualization and intent buffering mechanisms. This method solves the problems of "visual impact of path jumps" and "abrupt execution" not covered by existing technologies by combining a combination of "multi-hypothesis access at the prediction layer + confidence-based bucket rendering + easing interpolation + execution delay".

[0028] Example 3: This application provides a method for handling path transitions in intelligent driving SR based on multi-hypothesis trajectory rendering, such as Figure 1 As shown, the method includes: Step 1: Obtain K candidate trajectories for autonomous vehicles and their corresponding confidence levels; Step 2: Map the K trajectories to different confidence buckets based on their confidence levels, and render them in the SR interface in different styles; Step 3: Determine the transition duration. When the confidence of a certain trajectory crosses the bucket boundary, the SR interface smoothly changes its style within the transition duration through easing interpolation.

[0029] The easing function employs easeOutCubic cubic easing to simulate the "asymptotic stability" process of human cognition. When the confidence level of a trajectory crosses the bucket boundary (e.g., from 30% to 85%), the SR interface smoothly changes its style over a transition period (300-1200ms) through easing interpolation, rather than abruptly changing it. Figure 4 As shown, the method for determining the transition duration includes: Step 31: Determine the lower limit of the transition duration based on the human peripheral vision perception threshold and the SR rendering frame rate; Based on the human eye's peripheral vision perception threshold and the SR rendering frame rate (≥30fps), the lower limit of the transition time is determined to be 300ms. Below this value, the easing curve cannot unfold, and the user will still perceive it as a jump.

[0030] Step 32: Determine the upper limit of the transition time based on the confidence refresh cycle; The upper limit of the transition time is determined based on the prediction layer confidence refresh cycle (10Hz level). If it is higher than this value, the previous interpolation will not be completed when the next confidence update arrives, which may easily cause style jitter.

[0031] Step 33: Based on the vehicle model and the current scenario, dynamically select the transition duration between the upper and lower limits.

[0032] The specific value of the transition time can be dynamically adjusted according to the vehicle model and the current scenario: 600–1200ms for urban scenarios (low speed, stable trajectory), and 300–500ms for highway scenarios (high speed, frequent trajectory switching). As mentioned above, the transition time range is 300ms-1200ms (smaller value for high speed, larger value for low speed). The specific correspondence between the transition time value and speed is shown in Table 2. The transition time T and the vehicle speed v satisfy: T(v)=T_max−k·v, where T_max=1200ms, k is a calibration coefficient (preferably k=5~8ms / (km / h)), and T(v) is clamped in [T_min,T_max], T_min=300ms. For example: v=60km / h, k=7→T=1200−7×60=780ms, falling into urban mode; v=100→T=500ms, falling into highway mode. Interpolation attributes: Color: Linear interpolation of RGB values ​​(e.g., blue → red); Transparency: Linear interpolation of Alpha values; Line style: The proportion of dashed line segments gradually shortens within τ until it becomes a solid line.

[0033]

[0034] Table 2. Correspondence between transition duration and speed In the aforementioned "easing interpolation," the rate of change of interpolation variables (RGB color values, transparency α, light carpet width, etc.) over time is controlled by an easing function, preferably an ease-in-out curve, to create a non-linear "slow-fast-slow" transition in the style, avoiding the visual abruptness caused by linear interpolation. In this scheme, the ease-in-out function can be implemented using a cubic Bézier curve (0.42,0,0.58,1) or an equivalent quadratic function; that is, the easing function uses an easeOutCubic cubic easing, simulating the "asymptotic stability" process of human cognition. Figure 4 As shown, the method by which the SR interface smoothly changes its style within the transition time by using easing interpolation when the confidence of a certain trajectory crosses the bucket boundary includes: Step 34: When the confidence level increases, proceed to step 35; when the confidence level decreases, proceed to step 36. When the trajectory confidence crosses the bucket boundary (e.g., from bucket ③ to bucket ④), the SR interface does not immediately switch styles, but instead performs interpolation calculations for the transition duration. It should be noted that the interpolation strategies for confidence increases (low bucket → high bucket) and confidence decreases (high bucket → low bucket) can be configured asymmetrically.

[0035] Step 35: Based on the determined transition duration, use an easing function to implement the ease-in-out micro-animation effect on the SR interface to enhance the user's sense of confirmation. As confidence increases (from low bucket to high bucket), the interpolation duration is taken near the upper limit (e.g., 600–800ms). Easing (an easing function is a mathematical function used in animation to control changes in motion speed; it changes the animation from uniform speed to non-linear motion such as acceleration, deceleration, or bouncing, thus simulating realistic physics and improving visual naturalness) uses ease-in-out (ease-in-out is a transition effect in CSS animations, where the animation is slower at the beginning and end, and faster in the middle, resulting in a natural and smooth visual experience), combined with the SR interface's "ready state" micro-animation effect, to strengthen the user's sense of confirmation. Step 36: After the confidence level falls below the bucket boundary and is maintained for more than 3 consecutive frames, the ease-out micro-motion effect of the SR interface is realized through the easing function according to the determined transition duration. A brief flash is used to indicate the cancellation of the schematic diagram, so as to avoid the residual "confidence state" of the light carpet misleading the user.

[0036] When the confidence level decreases (from high bucket to low bucket), a jitter maintenance criterion is first introduced (interpolation is triggered only if the confidence level is maintained for ≥3 consecutive frames after falling below the threshold). The interpolation duration is taken near the lower limit (e.g., 200–400ms). Easing adopts ease-out (eask-out specifies a transition effect that ends slowly, which is equal to cubic-bezier(0,0,0.58,1). Compared to uniform speed, it starts fast and ends slowly). A brief flash of 80–120ms can be added to indicate the cancellation of the diagram and avoid the residual "credible state" of the light carpet misleading the user.

[0037] Step 4: When a trajectory is mapped to the bucket with the highest confidence, start timing for a duration of T. lead =k*v+b, where v is the vehicle speed, k is the calibration coefficient, and b is the lower limit of the transition time; When the confidence level reaches the execution threshold (e.g., >85%), the system enters T. leadBuffer period. During this time, a pre-fill command is issued to the chassis actuators—the braking / steering actuators build up pressure or torque in advance, but the execution effect is not released outwards yet; during this stage, the SR rendering side can supplement with "ready state" micro-motion effects (such as low-frequency pulses of light carpet) to convey the signal that "the vehicle has intervened and is ready" to the user. If no user takes over after the buffer period, the action is executed. For example, when the trajectory enters bucket ④ (locked), Intent Hold is triggered: 1. Timed start: T lead =k*v+b. Where v is the vehicle speed, k≈0.008s / (km / h), b≈0.3s. Example: Vehicle speed 60km / h, Tlead≈0.78s; Vehicle speed 120km / h, Tlead≈1.26s. 2. Actuator behavior: Braking: Only master cylinder pre-fill is performed, no decompression is established, and EPA is not triggered. Steering: Only a slight hand torque is applied, and the steering wheel is not turned.

[0038] Step 5: If the timer ends and the user has not taken over, lock the trajectory and send an execution request to make the vehicle travel along the locked trajectory.

[0039] This step, in addition to providing the normal execution conditions, also provides the cancellation conditions (Hysteresis): Condition A: User T lead Depress the pedal or apply steering torque. Condition B: When the trajectory confidence falls outside the boundary of the bucket with the highest confidence and below the preset hysteresis window, for example, falling back below 80% (set a 5% hysteresis window, set an 80%–85% hysteresis window to prevent critical jitter). Once the retraction is triggered, the light carpet pattern returns to bucket ③, and execution is canceled.

[0040] As described above, the technical solution provided by this invention involves the prediction layer simultaneously outputting data to both the planning control layer and the SR rendering layer. The planning control layer outputs the optimal trajectory to the Intent arbitration module, and the SR rendering layer outputs a visual light carpet to the display interface. The Intent arbitration module intercepts the planning control command within the Tlead time, controlling the actuator to only perform pre-filling. For example, scenario: A car in front on a highway suddenly crosses the solid line and cuts in. At time T0: This car is going straight (bucket ④, green solid line). The prediction layer outputs: going straight (70%), emergency braking (15%), lane change (15%). SR only displays the green solid line. At time T1 (+0.2s): The car in front intrudes. The prediction layer outputs: going straight (20%), emergency braking (60%), lane change (20%). SR rendering: The emergency braking trajectory enters bucket ③, displayed as a red pulsating solid line, located above the green solid line. At time T2 (+0.3s): The car in front completely cuts in. The prediction layer outputs: emergency braking (92%). SR rendering: The emergency braking trajectory enters bucket ④. The red light carpet transitions from pulsating to constant illumination and begins to extend forward, indicating the braking point. Intent Hold is initiated: Tlead = 0.9s. The master cylinder is pre-filled. At T3 (+0.4s to +1.2s): the user sees the red light carpet and moves their foot to the brake pedal in preparation for intervention. At T4 (+1.2s): if the user does not intervene, the braking system formally issues a braking request, and the vehicle comes to a smooth stop.

[0041] This application provides a method for handling path jumps in intelligent driving (SR) based on multi-hypothesis trajectory rendering. It addresses sudden path jumps caused by uncertainty in perception and control in intelligent driving (AD / ADAS / NOA) scenarios, improving driving safety and user experience through multi-hypothesis trajectory visualization and intent buffering mechanisms. This method solves the problems of "visual impact of path jumps" and "abrupt execution" not covered by existing technologies by combining a combination of "multi-hypothesis access at the prediction layer + confidence-based bucket rendering + easing interpolation + execution delay".

[0042] In summary, this application provides a method for intelligent driving SR path transition processing based on multi-hypothesis trajectory rendering. This technical solution proposes a mechanism of "multi-hypothesis pre-rendering + intent buffer execution," utilizing the "candidate trajectories" originally discarded by the control layer, visualizing them in layers on the SR interface, and setting a "thinking buffer period" before executing the action. This technology effectively eliminates visual impact: the gradual transition from dashed to solid lines eliminates the "popping" feeling of the image; reduces cognitive load: users can intuitively see the system's "hesitation" and "decision-making process," establishing psychological expectations; and enhances safety redundancy: the IntentHold mechanism effectively prevents false braking caused by sensor noise, giving the driver a final veto right.

[0043] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for handling path transitions in autonomous driving based on multi-hypothesis trajectory rendering, applied to autonomous vehicles, characterized in that, The method includes: Step 1: Obtain K candidate trajectories for autonomous vehicles and their corresponding confidence levels; Step 2: Map the K trajectories to different confidence buckets based on their confidence levels, and render them in the SR interface in different styles; Step 3: Determine the transition duration. When the confidence of a certain trajectory crosses the bucket boundary, the SR interface smoothly changes its style within the transition duration through easing interpolation. Step 4: When a trajectory is mapped to the bucket with the highest confidence, start timing for a duration of T. lead =k*v+b, where v is the vehicle speed, k is the calibration coefficient, and b is the lower limit of the transition time; Step 5: If the timer ends and the user has not taken over, lock the trajectory and send an execution request to make the vehicle travel along the locked trajectory.

2. The intelligent driving SR path transition processing method based on multi-hypothesis trajectory rendering according to claim 1, characterized in that, K is 3-5, k=0.008s / (km / h), b=0.3s.

3. The intelligent driving SR path transition processing method based on multi-hypothesis trajectory rendering according to claim 1, characterized in that, Step 1 includes: Step 11: Based on the current environmental perception results, output the K most probable future driving trajectories of the vehicle through forward inference of the neural network; Step 12: Obtain perception data and vehicle status, and calculate the posterior probability of each candidate trajectory through a Softmax classification layer or regression head. Step 13: Quantize the posterior probability into a confidence level value of 0-100%, and map the geometric coordinates of the K trajectories to the confidence level values ​​for later decision-making and reality.

4. The intelligent driving SR path transition processing method based on multi-hypothesis trajectory rendering according to claim 1, characterized in that, Step 2 includes: Step 21: Pre-set several confidence buckets and dynamically adjust the bucket boundaries using remote wireless upgrade technology; Step 22: The interval boundaries of the confidence buckets and the SR rendering style parameters bound to each bucket are stored in the vehicle-side configurable parameter table. Step 23: Dynamically distribute updates using remote wireless upgrade technology.

5. The intelligent driving SR path transition processing method based on multi-hypothesis trajectory rendering according to claim 1, characterized in that, The method for determining the transition duration includes: Step 31: Determine the lower limit of the transition duration based on the human peripheral vision perception threshold and the SR rendering frame rate; Step 32: Determine the upper limit of the transition time based on the confidence refresh cycle; Step 33: Based on the vehicle model and the current scenario, dynamically select the transition duration between the upper and lower limits.

6. The intelligent driving SR path transition processing method based on multi-hypothesis trajectory rendering according to claim 5, characterized in that, The lower limit of the transition duration is 0.3s, and the upper limit of the transition duration is 1.2s.

7. The intelligent driving SR path transition processing method based on multi-hypothesis trajectory rendering according to claim 1, characterized in that, The method for the SR interface to smoothly change its style during the transition time when the confidence of a certain trajectory crosses the bucket boundary includes: Step 34: When the confidence level increases, proceed to step 35; when the confidence level decreases, proceed to step 36. Step 35: Based on the determined transition duration, use an easing function to implement the ease-in-out micro-animation effect on the SR interface to enhance the user's sense of confirmation. Step 36: After the confidence level falls below the bucket boundary and is maintained for more than 3 consecutive frames, the ease-out micro-motion effect of the SR interface is realized through the easing function according to the determined transition duration. A brief flash is used to indicate the cancellation of the schematic diagram, so as to avoid the residual "confidence state" of the light carpet misleading the user.

8. The intelligent driving SR path transition processing method based on multi-hypothesis trajectory rendering according to claim 7, characterized in that, The easing function uses the easeOutCubic cubic easing to simulate the "asymptotic stability" process of human cognition.

9. The intelligent driving SR path transition processing method based on multi-hypothesis trajectory rendering according to claim 1, characterized in that, Step 5 further includes: if the timer has not ended, and the user presses the pedal or applies steering torque, then proceed to step 1.

10. The intelligent driving SR path transition processing method based on multi-hypothesis trajectory rendering according to claim 1, characterized in that, Step 5 also includes: if the timing has not ended, and the trajectory confidence falls outside the boundary of the bucket with the highest confidence and is lower than the preset hysteresis window, then proceed to step 1.

Images