An intelligent driving holographic security architecture and method based on field regulation

Abstract

This invention discloses a holographic safety architecture and method for intelligent driving based on field control, belonging to the field of autonomous driving safety technology. The architecture includes: a perception field construction layer, used to fuse data collected by onboard multimodal sensors into a unified holographic field map of the driving environment, wherein the holographic field map contains superimposed information of obstacle potential field, lane constraint potential field, and target path potential field; a decision field control layer, with a built-in field balance controller, which calculates the deviation value of the driving safety situation in real time through a single composite operation consisting of exponential operation, logarithmic operation, and difference operation, and generates control commands; and a control field execution layer, used to uniformly schedule each execution subsystem according to the control commands, so that the vehicle driving state returns to a preset safe steady state. This invention solves the problems of perception and decision-making separation, external safety verification, and conflicting commands from multiple execution subsystems in existing autonomous driving systems through a closed-loop safety architecture of "perception field—decision field—control field," and is applicable to the full-domain safety control of L3 and above autonomous driving systems.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving safety technology, specifically relating to a holographic safety method and architecture that treats the driving environment as a continuous "safety field" and performs dynamic balance control through a three-element closed-loop architecture, applicable to the full-domain safety control of L3 and above autonomous driving systems. Background Technology

[0002] Current autonomous driving safety systems generally adopt a pipeline architecture of "perception-decision-execution," with each stage relatively independent. Safety verification is usually added as an external module to the decision output. This architecture has three fundamental flaws: First, perception and decision-making are disconnected. Multimodal sensors (LiDAR, millimeter-wave radar, cameras, ultrasonic sensors, etc.) process data independently, and the perception results are input into the decision-making module separately. This fragmented processing method lacks a unified field representation of the overall driving environment, making it difficult to comprehensively perceive global risks. When a single sensor fails or data conflicts, the decision-making module often cannot obtain a unified view of the global situation in a timely manner, resulting in a delay in safety response.

[0003] Secondly, there are external safety verification plugins. External safety modules typically use rule engines or classifiers to perform a binary "pass / intercept" judgment on the output of the decision-making module. When the system makes an incorrect decision, the external safety module can only passively intercept or correct it, unable to restore the balance of the entire driving system at a global level. This "adversarial" safety architecture leads to the paradox of "creativity and safety being mutually exclusive"—excessive safety restrictions affect the driving experience, while relaxing safety constraints poses an accident risk.

[0004] Third, control commands are fragmented. Safety commands from various execution subsystems (braking, steering, drive, etc.) are difficult to coordinate uniformly. In emergency situations, safety commands from different subsystems may conflict. For example, if the braking system applies emergency braking while the steering system attempts an emergency lane change, the combination of these actions could lead to loss of vehicle control.

[0005] In 2026, a team from the Chinese Academy of Sciences and other institutions published groundbreaking findings in the journal *Science*, revealing that the primate cerebral cortex is not composed of discrete "functional modules," but rather is defined by a unified "Pr-Al opposing molecular gradient axis." This discovery suggests that the core capability of an efficient safety system does not stem from the control of individual "parts," but rather from the maintenance of dynamic equilibrium in a continuous "field."

[0006] Therefore, a novel intelligent driving safety architecture is urgently needed. Drawing inspiration from the "field" principle of how living organisms regulate complex systems, this invention constructs a closed-loop architecture that unifies perception, decision-making, and control within a continuous safety field, fundamentally addressing the "fragmentation" and "external" shortcomings of existing safety systems. The technical concept of this invention is inspired by ancient Chinese numerology and symbolism, combined with modern autonomous driving safety technology. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of existing technologies by providing a field-controlled intelligent driving holographic safety method and architecture to solve technical problems such as the separation of perception and decision-making in existing autonomous driving systems, external safety verification tools, and conflicting instructions from multiple execution subsystems.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A field-domain controlled holographic safety method for intelligent driving includes the following steps: S1: Perception Field Construction Steps – The perception data collected by the vehicle-mounted multimodal sensors are fused into a unified holographic field map of the driving environment. This holographic field map includes superimposed information of the obstacle potential field, lane constraint potential field, and target path potential field. The weight coefficients of each potential field function are dynamically adjusted according to the driving scenario.

[0009] S2: Decision-making field control step—The holographic field map is input into the field balance controller. The controller calculates the deviation between the current driving situation and the preset safe steady state through a single composite operation consisting of exponential, logarithmic, and difference operations, and generates control commands. The functional form of the single composite operation is: Safety situation value H = e^(risk deviation value) - ln(safety benchmark value).

[0010] S3: Control field execution steps - According to the control instructions, the execution subsystems are uniformly scheduled to bring the vehicle driving state back to the preset safe steady state.

[0011] Furthermore, the control threshold of the safety situation value H is divided into three levels: when H>0.5 and lasts for more than 20 milliseconds, a level one warning is triggered; when H>1.0, a level two intervention is triggered; and when H>2.0, a level three emergency braking is triggered.

[0012] This invention also provides a security architecture for executing the above-described methods, comprising a perception field construction layer, a decision field control layer, and a control field execution layer. The decision field control layer incorporates a single composite computing hardware module. The exponential operation circuit is implemented in hardware using a coordinate rotation digital calculation method, the logarithmic operation circuit is implemented in hardware using a lookup table, and the difference operation circuit is a 32-bit fixed-point operation circuit. The response period of the entire hardware module does not exceed 50 microseconds. Detailed Implementation

[0013] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can refer to and implement it.

[0014] Example 1: System Basic Architecture and Hardware Implementation The perception field construction layer integrates perception data collected by multimodal sensors such as vehicle-mounted LiDAR, millimeter-wave radar, ultrasonic sensors, and surround-view cameras into a unified holographic field map of the driving environment in real time using the artificial potential field method.

[0015] The holographic field diagram is composed of the superposition of three fundamental potential field components: (1) Obstacle potential field: This describes the distribution of obstacles around a vehicle. The potential field strength is inversely proportional to the distance to the obstacle; the closer the distance, the higher the potential field strength. The mathematical expression for the obstacle potential field is: U_obs = k / d², where k is the potential field strength coefficient and d is the Euclidean distance between the vehicle and the obstacle. Different potential field strength coefficients are used for different types of obstacles (pedestrians, vehicles, and fixed obstacles) to reflect different risk levels.

[0016] (2) Lane constraint potential field: used to describe the trend constraint of a vehicle deviating from the lane centerline. The potential field strength increases rapidly as the distance of the vehicle from the lane centerline increases. The lane constraint potential field adopts a parabolic model: U_lane = 0.5 ×k_lane × (Δy)², where k_lane is the lane constraint coefficient and Δy is the lateral distance between the vehicle center and the lane centerline.

[0017] (3) Target path potential field: used to guide vehicles to travel in the direction of the target path. The potential field has a minimum value near the target path and increases when it deviates from the target path. The target path potential field is calculated and updated in real time by the global path planning module.

[0018] The three potential field components are superimposed to form a holographic field map of the current driving environment. This field map is centered on the vehicle's current position and covers a three-dimensional spatial range of 200 meters in front of the vehicle, 50 meters behind it, and three lanes on each side. All sensor data are processed synchronously, and the field map is updated at a frequency of 100Hz, meaning that the entire field map is refreshed every 10 milliseconds.

[0019] The decision-making field control layer incorporates a single composite computing hardware module. This module consists of three cascaded hardware circuits: Level 1: CORDIC exponentiation circuit. It directly calculates e^(risk deviation value) at the hardware level using the Coordinate Rotation Digital Computer method. The CORDIC algorithm requires only addition, subtraction, and shift operations, eliminating the need for multipliers and lookup tables, making it suitable for implementation on FPGAs or application-specific integrated circuits. The calculation precision is 32-bit floating-point precision.

[0020] Level 2: Look-up Table Logarithmic Operation Circuit. The look-up table (LUT) is used to calculate ln (a safety baseline value). The safety baseline value is a preset system constant (typically 1.0), corresponding to ln(1.0) = 0. The LUT stores logarithmic values ​​in intervals of 0.01 within the range of 0.1 to 10.0, with a lookup speed of one clock cycle.

[0021] The third stage: a 32-bit fixed-point difference calculation circuit. Subtracting the output of the second stage (ln(safety reference value)) from the output of the first stage (e^(risk deviation value)) yields the final safety situation value H = e^(risk deviation value) - ln(safety reference value).

[0022] The entire hardware module has a delay of no more than 50 microseconds from receiving holographic field map data to outputting the H value.

[0023] The control field execution layer receives control commands output from the decision field control layer and uniformly schedules the execution subsystems of the braking, steering, and drive domains. The control commands are sent in real time via the vehicle's high-speed bus (such as CAN FD or vehicle Ethernet), and the response latency of each execution subsystem is controlled within 5 milliseconds.

[0024] Example 2: Typical Safety Scenario - Emergency Braking Taking the example of a vehicle detecting a stationary obstacle suddenly appearing in front of it while cruising at high speed, the closed-loop control process of the present invention is fully demonstrated.

[0025] Initial state: The vehicle is cruising on the highway at a speed of 120 km / h, and the perception field construction layer continuously monitors the road conditions ahead. At this time, the distribution of each potential field in the holographic field map is within the normal range, the risk deviation value is about 0, the H value fluctuates around zero (within ±0.15), and the system is in a steady state.

[0026] Triggering Phase (Time t0): The lidar detects a stationary vehicle ahead at a distance of 200 meters. The potential field strength of the obstacle increases instantaneously, and the potential field strength in the area ahead in the holographic field map rises sharply. The perception field construction layer refreshes the field map within 10 milliseconds and sends the updated holographic field map data to the decision field control layer.

[0027] H-value calculation phase (time t0+10ms to t0+60μs): The decision field control layer receives the updated field map data. The risk deviation value jumps from 0 to 3.0 (corresponding to the risk level of a stationary obstacle suddenly appearing 200 meters ahead). The CORDIC exponent calculation circuit calculates e³≈20.085 in approximately 20 microseconds. The lookup table logarithmic operation circuit outputs ln(1.0)=0 within one clock cycle. The difference operation circuit outputs H = 20.085 - 0 = 20.085.

[0028] Since H > 2.0, a Level 3 emergency braking command is directly triggered.

[0029] Control execution phase (starting from time t0+50μs): The field balance controller immediately generates the highest priority control command after detecting H>2.0. The command is simultaneously sent to the braking domain, steering domain, and drive domain. The braking domain performs an emergency braking operation (brake pressure increases to its maximum value at the maximum rate); the steering domain assesses the feasibility of changing lanes to the left or right, and if safe, preloads steering torque to prepare for lane changing; the drive domain forcibly reduces the output torque to zero.

[0030] Closed-loop feedback phase (starting from t0+10ms): The perception field construction layer continuously updates the holographic field map at a frequency of 100Hz. The distance between the vehicle and the obstacle gradually decreases from 200 meters, but the relative speed drops rapidly due to braking. The obstacle potential field strength increases as the distance decreases, but the risk deviation value gradually decreases due to effective braking control. During braking, the H value gradually decreases from 20.085.

[0031] Safe stopping phase: The vehicle comes to a complete stop approximately 20 meters in front of the obstacle. At this point, although the obstacle's potential field strength is high (due to the short distance), the vehicle's speed is zero, and the overall safety situation has returned to a steady state. The H value drops back to near zero, and all execution subsystems return to normal standby status.

[0032] Example 3: Multi-level security response The security situation value H control of this invention adopts a three-level graded response mechanism to adapt to security threats of different urgency levels: Level 1 Warning (H>0.5, lasting longer than 20 milliseconds): The field balance controller generates a Level 1 warning command. The onboard system alerts the driver to the current risk through visual, auditory, or tactile means, while the control field execution layer reserves additional braking and steering capabilities, but does not actively intervene in vehicle movement. Level 1 warnings are suitable for scenarios where the risk is slowly escalating and controllable, such as the vehicle in front gradually decelerating or a vehicle in an adjacent lane slightly crossing the line.

[0033] Level 2 Intervention (H>1.0): The field balance controller generates a level 2 intervention command. The control field execution layer actively adjusts the vehicle's motion state: limiting the maximum speed to below 80% of the current speed; increasing the minimum following distance from the default 2 seconds to 3 seconds; preloading the braking system to shorten emergency braking response time; and adjusting the steering system's power steering characteristics to a more conservative mode. Level 2 intervention is suitable for high-risk scenarios that can still be avoided by adjusting driving parameters, such as moderately congested roads and driving in inclement weather.

[0034] Level 3 Emergency Braking (H>2.0): The field balance controller generates a Level 3 emergency braking command. All execution subsystems respond with the highest priority: the braking domain executes full emergency braking, stopping the vehicle with maximum braking force with the assistance of ABS (Anti-lock Braking System); the steering domain evaluates and executes the optimal avoidance path (performing an emergency lane change if necessary); the drive domain forcibly reduces output torque to zero, cutting off power transmission. Level 3 emergency braking is suitable for extremely dangerous scenarios where a collision is imminent or cannot be avoided by adjusting driving parameters.

[0035] Example 4: Comparison of the improvements of this architecture to existing technologies Compared with the traditional "perception-decision-execution" pipeline architecture, the ternary collaborative closed-loop security architecture of this invention achieves significant improvements in the following key performance indicators: Safety Response Delay: The end-to-end delay from sensor data acquisition to actuator action in this invention is no more than 50 microseconds (calculated by the field balance controller hardware) + 10 milliseconds (field map refresh cycle) = approximately 10.05 milliseconds. In traditional architectures, multimodal sensor data needs to be input separately into the perception fusion module, scene understanding module, and decision planning module, undergoing layer-by-layer processing before outputting control commands, with end-to-end delays typically ranging from 50 to 200 milliseconds. This invention reduces latency by approximately 80%.

[0036] Emergency braking response: Upon detecting a collision risk (H>2.0), the field balance controller generates a braking command within 50 microseconds. Traditional architectures require target identification (approximately 50 milliseconds), risk assessment (approximately 20 milliseconds), and decision selection (approximately 30 milliseconds), totaling approximately 100 milliseconds. This represents a 2000-fold improvement in response speed.

[0037] Multi-subsystem coordination: This invention coordinates the braking, steering, and drive domains simultaneously through a unified control command, avoiding subsystem command conflicts. In traditional architectures, each subsystem is controlled independently; during emergency braking, steering and braking may conflict, leading to loss of vehicle control. This invention eliminates this potential hazard. Beneficial effects

[0038] This is the first time that the "field control" principle has been applied to the field of intelligent driving safety, constructing a unified safety field covering perception, decision-making, and control, fundamentally solving the problem of fragmented decision-making and execution in existing autonomous driving systems.

[0039] A single composite operation implemented in hardware can complete safety situation deviation detection and instruction generation in microseconds, meeting the stringent requirements of autonomous driving for response latency.

[0040] By implementing unified safety command scheduling at the field level, the problem of command conflicts between multiple execution subsystems such as braking, steering, and drive is completely resolved, ensuring vehicle safety in emergency situations.

[0041] Through a three-tiered response mechanism, a smooth transition from early warning to intervention and then to emergency braking is achieved, avoiding the crude intervention of "all or nothing" in traditional safety systems.

[0042] This architecture does not depend on specific sensor configurations or execution subsystem types, and has good platform compatibility and application scalability. Attached Figure Description

[0043] Figure 1 Intelligent Driving Holographic Safety Architecture Diagram Figure 2 Schematic diagram of holographic field diagram structure Figure 3 Security situation control flowchart.

Claims

1. A holographic safety method for intelligent driving based on field control, characterized in that, Includes the following steps: S1: Perception Field Construction Steps - The perception data collected by the vehicle-mounted multimodal sensors are fused into a unified holographic field map of the driving environment. The holographic field map contains superimposed information of obstacle potential field, lane constraint potential field and target path potential field. S2: Decision Field Control Steps - The holographic field map is input into the field balance controller. The controller calculates the deviation between the current driving situation and the preset safe steady state through a single composite operation consisting of exponential operation, logarithmic operation and difference operation, and generates control instructions. S3: Control field execution steps - According to the control instructions, the execution subsystems are uniformly scheduled to bring the vehicle driving state back to the preset safe steady state.

2. The method according to claim 1, characterized in that, The single composite operation in step S2 uses the following function: safety status value H = e^(risk deviation value) - ln(safety benchmark value), where the risk deviation value is the difference between the current perceived risk level and the preset safety level, and the safety benchmark value is the preset steady-state benchmark constant.

3. The method according to claim 2, characterized in that, When the safety status value H exceeds a preset threshold, the field balance controller generates an enhanced suppression command to trigger active safety intervention. The active safety intervention includes at least one of reducing vehicle speed, increasing following distance, and triggering emergency braking.

4. The method according to claim 3, characterized in that, The control threshold of the safety situation value H is divided into three levels: when H > 0.5 and lasts for more than 20 milliseconds, a level 1 warning is triggered to alert the driver; when H > 1.0, a level 2 intervention is triggered to actively limit the vehicle speed and increase the following distance; when H > 2.0, a level 3 emergency braking is triggered, and all execution subsystems respond with the highest priority.

5. The method according to claim 1, characterized in that, The superposition information of obstacle potential field, lane constraint potential field and target path potential field mentioned in step S1 is calculated in real time by artificial potential field method, and the weight coefficient of each potential field function is dynamically adjusted according to the driving scenario.

6. A field-domain controlled intelligent driving holographic safety architecture, used to execute the method of any one of claims 1 to 5, characterized in that, include: The perception field construction layer is used to fuse the perception data collected by the vehicle's multimodal sensors into a unified holographic field map of the driving environment. The decision field control layer is connected to the perception field construction layer. It has a built-in single composite computing hardware module consisting of an exponential operation circuit, a logarithmic operation circuit, and a difference operation circuit, which is used to calculate the security situation deviation value in real time and generate control commands. The control field execution layer is connected to the decision field regulation layer and is used to uniformly schedule each execution subsystem according to the regulation instructions.

7. The security architecture according to claim 6, characterized in that, In the single composite computing hardware module of the decision field control layer, the exponential operation circuit is implemented in hardware using a coordinate rotation digital calculation method, the logarithmic operation circuit is implemented in hardware using a lookup table, and the difference operation circuit is a 32-bit fixed-point operation circuit; the response period of the entire hardware module does not exceed 50 microseconds.

Images