Self-adaptive path planning and obstacle avoidance control system for mining heavy-load transport vehicle

Abstract

The invention relates to the field of vehicle automation control, and discloses a self-adaptive path planning and obstacle avoidance control system for a mining heavy-load transport vehicle, which comprises an environment sensing module configured to sense the surrounding environment of the vehicle in real time and perform semantic attribute calibration on sensed obstacles, a state field construction module configured to construct a state field, the system comprises a semantic attribute calibration module configured to establish a multi-dimensional posture field representing trafficability based on a semantic attribute calibration result, a risk memory management module configured to monitor the multi-dimensional posture field to identify a high-risk event and generate a risk memory potential field with a time decay characteristic based on the high-risk event, and a track generation control module configured to generate a track memory potential field with a time decay characteristic based on the high-risk event. And the superposition module is configured to superpose the multi-dimensional potential field and the risk memory potential field to form a total potential energy field. Through refined environment perception, risk avoidance with time sequence memory and stable planning control execution, the planning intelligence and operation safety of the mining vehicle in a complex dynamic environment are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of vehicle automation control, in particular to a self-adaptive path planning and obstacle avoidance control system for mine heavy-load transport vehicles. BACKGROUND

[0002] For path planning and obstacle avoidance of mine heavy-load transport vehicles, the artificial potential field method is a basic technical solution. This method is applied due to its relatively simple algorithm structure and high calculation efficiency. The core of this method is to construct a potential field space representing the environmental passability through mathematical functions. In this space, an attractive potential field with increasing potential value with increasing distance is constructed around the preset target point. Meanwhile, a repulsive potential field with sharply increasing potential value with decreasing distance is constructed around the obstacles in the environment. The planned path of the vehicle is determined according to the negative gradient direction of the total potential field obtained by superimposing the attractive potential field and the repulsive potential field.

[0003] However, in the scene of mine area with complex operation process, mixed driving of people and vehicles, and dynamically changing environment, the traditional artificial potential field method has some inherent technical problems in actual application. First, the method has a relatively single treatment for environmental obstacles, which are usually regarded as homogeneous repulsive sources. It cannot effectively distinguish between transport vehicles driving along fixed routes with predictable behavior and on-site workers with random motion trajectories and strong behavior suddenness. This non-differentiated treatment leads to overly conservative obstacle avoidance strategies of the system, affecting the traffic efficiency, or insufficient risk margin, causing safety hazards. In addition, the decision-making process of the traditional method lacks time sequence correlation, and the path selection only depends on the environmental information at the current time. For sudden high-risk events occurring during driving, such as close-range emergency obstacle avoidance, the system cannot form records or memories, which leads to the inability of the vehicle to adjust the subsequent traffic strategy according to historical risk events. When entering the same dangerous area again, the same unoptimized obstacle avoidance behavior is still used. Finally, the conversion process from the potential field negative gradient vector generated by the planning layer to the vehicle bottom layer physical control instruction is not a direct mapping. Mine heavy-load vehicles have large inertia and control delay characteristics. If the expected motion vector cannot be effectively mapped to smooth and stable longitudinal acceleration and front wheel steering angle instructions, it is easy to cause oscillation of the driving trajectory or overshoot of the control amount, thereby affecting the stability and safety of the operation. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides a self-adaptive path planning and obstacle avoidance control system for mine heavy-load transport vehicles, which solves the problems of the prior art that the path planning and obstacle avoidance of heavy-load transport vehicles cannot differentiate obstacles, the obstacle avoidance strategy lacks pertinence, the risk memory mechanism is lacking, potential dangers cannot be avoided according to historical experience, and driving oscillation is easily caused due to unstable conversion of planning instructions to physical control.

[0005] To achieve the above object, the present application is implemented by the following technical solutions: a self-adaptive path planning and obstacle avoidance control system of a mine heavy-load transport vehicle, comprising an environment perception module, a potential field construction module, a risk memory management module and a trajectory generation control module.

[0006] The environment perception module is configured to obtain information of the vehicle driving environment in real time and process the information, and specifically comprises a data acquisition unit and a semantic labeling unit. The data acquisition unit is used to acquire multi-modal data of the environment around the vehicle, and the semantic labeling unit receives the multi-modal data and labels the semantic attributes of the obstacles identified in the data to determine the category attributes of the obstacles. Specifically, the semantic labeling unit determines the category attributes of the obstacles as one of static obstacles, predictable dynamic obstacles and unpredictable dynamic obstacles.

[0007] The potential field construction module is configured to establish a multi-dimensional potential field for describing the passability of the vehicle driving environment according to the labeling results output by the environment perception module, and the module comprises a target attractive potential field generation unit, a static repulsive potential field generation unit, a dynamic interference potential field generation unit and a rule constraint potential field generation unit.

[0008] The target attractive potential field generation unit is used to generate a target attractive potential field pointing to a preset target point, which provides a global driving direction for the vehicle.

[0009] The static repulsive potential field generation unit is used to generate a static physical repulsive potential field generated by the static obstacles according to the position information of the static obstacles.

[0010] The dynamic interference potential field generation unit is used to generate a dynamic interference potential field according to the position, speed and category attribute information of the dynamic obstacles. For the objects labeled as predictable dynamic obstacles, the unit generates an anisotropic potential field distribution extending along the predicted trajectory direction of the objects. For the objects labeled as unpredictable dynamic obstacles, the unit generates an isotropic potential field distribution with the current position of the objects as the center.

[0011] The rule constraint potential field generation unit is used to generate a rule and constraint potential field for constraining the behavior of the vehicle according to the road boundary and no-passing area rule information.

[0012] The risk memory management module is configured to identify high-risk events in the historical driving process and generate local risk area information with timeliness, and the module comprises a risk event triggering unit and a memory potential field generation unit.

[0013] The risk event triggering unit continuously monitors the multidimensional potential field generated by the potential field construction module, and determines the occurrence of a risk event when the potential energy value of the area where the vehicle is located exceeds the preset risk threshold.

[0014] The memory potential field generation unit generates a risk memory potential field at the corresponding position according to the position and occurrence time information of the risk event determined by the risk event triggering unit, and sets the potential energy value of the risk memory potential field as a time function with exponential decay characteristics, so that the potential energy value decreases over time.

[0015] The trajectory generation control module is configured to integrate all potential field information and generate specific control instructions for the vehicle bottom layer, and the module includes a total potential field superposition unit, a potential gradient calculation unit and an expected state generation unit.

[0016] The total potential field superposition unit superimposes the multidimensional potential field generated by the potential field construction module and the risk memory potential field generated by the risk memory management module to form a total potential field.

[0017] The potential gradient calculation unit is used to calculate the potential energy negative gradient of the current position of the vehicle in the total potential field, which is determined as the direction with the fastest potential energy decrease in the total potential field, representing the optimal local passing direction.

[0018] The expected state generation unit generates the expected motion state of the vehicle based on the calculated potential energy negative gradient, and specifically, the unit decouples the expected motion state into two directly executable control quantities, i.e., the expected longitudinal acceleration and the expected front wheel steering angle.

[0019] The present application provides a kind of self-adapting path planning and obstacle avoidance control system of heavy load transport vehicle in mine.

[0020] With the following beneficial effects:

[0021] 1、The environment perception module of the present application labels the semantic attributes of obstacles, and distinguishes obstacles into three categories: static, predictable dynamic and unpredictable dynamic, and the potential field construction module generates potential fields of different shapes for obstacles of different categories, achieving fine modeling and differentiated response to dynamic complex environment. Compared with the technology of treating all obstacles equally, the present application can more predictably plan an obstacle avoidance trajectory, improving the passing efficiency and smoothness of the vehicle in a dynamic scene under the premise of ensuring safety.

[0022] 2、The application sets the risk memory management module, the module can monitor potential field to identify high-risk events, and generate risk memory potential field with time decay characteristics at the event occurrence position, thereby giving the system the ability to learn and adapt from short-term historical driving experience, when the vehicle approaches the area where the high-risk event such as emergency obstacle avoidance has occurred again, the temporary memory potential field will guide the vehicle to take more conservative and safe strategy, effectively reduce the probability of accidents in specific scenarios with repeatability or potential danger, and improve the overall safety of the system.

[0023] 3、The application generates a total potential field by the trajectory generation control module, which superimposes all influencing factors such as target attraction, multi-type obstacle avoidance, rule constraint and risk memory, and based on the negative gradient of the total potential field, finally decouples the expected motion state into expected longitudinal acceleration and expected front wheel steering angle, builds a complete and clear closed-loop system from environment perception to vehicle bottom layer physical control, directly maps the abstract planning algorithm into executable control instructions, enhances the engineering practicability and reliability of the whole system, and ensures that the planning decision can be accurately and stably executed by the vehicle. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 The flow framework diagram of the system of the application is shown in the figure;

[0025] Figure 2 The flow framework diagram of the environment perception module of the application is shown in the figure;

[0026] Figure 3 The flow framework diagram of the potential field construction module of the application is shown in the figure;

[0027] Figure 4 The flow framework diagram of the risk memory management module of the application is shown in the figure;

[0028] Figure 5 The flow framework diagram of the trajectory generation control module of the application is shown in the figure. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the specification of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the application.

[0030] Please refer to the drawings in the specification of the application Figure 1 - the drawings in the specification of the application Figure 5The embodiment of the present application provides a kind of mine heavy load transport vehicle's adaptive path planning and obstacle avoidance control system, including environment perception module, situation field construction module, risk memory management module and trajectory generation control module.These modules work cooperatively to realize the autonomous, safe travel of vehicle in complex mine environment.

[0031] The function of environment perception module is to obtain the state information of the environment around the vehicle in real time, and to structure the key information, to provide input for subsequent path planning and decision-making.The module is specifically composed of data acquisition unit and semantic calibration unit.

[0032] Data acquisition unit: this unit obtains environmental data through a multi-type sensor array on the vehicle, for example, laser radar can be used to obtain high-precision three-dimensional point cloud data for accurate ranging and obstacle contour construction;Visual camera is used to obtain image data for identifying the color and texture information of objects;Millimeter wave radar is used to detect long-distance targets and their speed, especially in bad weather conditions such as rain, snow and fog, millimeter wave radar has good penetration, and by fusing these multi-modal data from different sources, a more complete and robust perception of the surrounding environment can be formed.

[0033] Semantic calibration unit: this unit is responsible for processing the fused data output by the data acquisition unit, and its core function is to detect, track and classify obstacles in the environment.First, a target detection algorithm (such as a convolutional neural network based on deep learning) is used to identify independent obstacle entities from image or point cloud data, and then a multi-target tracking algorithm (such as Kalman filter or its variants) is used to continuously track these obstacles to obtain their kinematic information such as position, velocity and heading in consecutive time frames.

[0034] Based on these kinematic information, the semantic calibration unit labels each tracked obstacle with a class attribute, and the labeling process can follow the following rules:

[0035] Static obstacle: if the position of the obstacle does not change or the change is less than a preset small threshold in consecutive multiple observation periods, it is labeled as a static obstacle, for example, rocks, parked engineering vehicles and buildings in the mine area.

[0036] Predictable dynamic obstacle: if the obstacle shows a stable and continuous motion law (e.g., a constant speed and heading) in consecutive observation periods, it is labeled as a predictable dynamic obstacle, and the future trajectory of such obstacle can be predicted in a short time through a kinematic model, for example, other transport vehicles normally driving on the mine road.

[0037] Unpredictable dynamic obstacle: If the motion trajectory of an obstacle shows high randomness and uncertainty, which cannot be effectively predicted by simple kinematic models, or its own attributes are of suddenness, it will be marked as an unpredictable dynamic obstacle, for example, a worker walking freely in the work area, an animal crossing the road, or a excavator in loading and unloading operation.

[0038] The module finally outputs a list of structured environment information, which contains the position, speed, geometric size and marked semantic category attribute of each obstacle.

[0039] The potential field construction module receives the output from the environment perception module and constructs a multi-dimensional potential field based on it, which can comprehensively describe the trafficability of the environment. The potential field is composed of multiple sub-potential fields of different properties, including:

[0040] Target attractive potential field generation unit: This unit generates a global attractive potential field, which guides the vehicle to move towards the preset destination or path point. This potential field can be designed as a function proportional to the distance from the vehicle to the target point. The farther the distance, the higher the potential value, thus generating an attractive force pointing to the target point at the current position of the vehicle.

[0041] Static repulsive potential field generation unit: For objects marked as "static obstacles", this unit generates a physical repulsive potential field around them. The potential value of this field decreases from the center of the obstacle to the periphery. Its influence range and strength can be set according to the size of the obstacle and the safety distance requirement. When the vehicle enters this area, it will be affected by a repulsive force, thus achieving avoidance of static obstacles.

[0042] Dynamic interference potential field generation unit: This is the core unit for handling dynamic obstacles. It generates different forms of potential fields according to the semantic category attributes of the obstacles:

[0043] For "predictable dynamic obstacles", this unit generates an anisotropic interference potential field based on their current speed and predicted trajectory. This potential field is stretched and strengthened in the predicted forward direction of the obstacle, while it is relatively weak in the lateral and rear directions. This form of potential field can make an early judgment and avoidance of the space that the obstacle will occupy, so that the vehicle can avoid in a smoother and more predictable way, rather than making a sharp reaction when it is close.

[0044] For "unpredictable dynamic obstacles", due to the randomness of its motion, this unit generates an isotropic circular or spherical interference potential field with the current position of the obstacle as the center. The intensity and range of this potential field are the same in all directions, forming a strict, directionless warning area, ensuring that the vehicle maintains a sufficient safety margin from this high-uncertainty risk source.

[0045] Rule-constrained potential field generation unit: This unit converts the traffic rules and geographical constraints of the mining area into a potential field, thereby strictly limiting the planned path of the vehicle within the drivable area.

[0046] The risk memory management module gives the system the ability to learn from high-risk events, and its workflow is as follows:

[0047] Risk event triggering unit: This unit monitors the total potential energy value (formed by the superposition of various potential fields in the potential state field construction module) of the vehicle's current location in real time. During the vehicle's journey, if it is forced to avoid a sudden danger and comes very close to an obstacle, the repulsive potential energy of this area will increase sharply. When the potential energy value exceeds the pre-set risk threshold, the system determines that a high-risk event has occurred.

[0048] Memory potential field generation unit: Once a risk event is triggered, this unit will immediately generate an additional, temporary repulsive potential field at the geographical location of the triggering event, i.e., the risk memory potential field. The function of this potential field is to mark the area as a potential danger point in a short period of time. The potential energy value of this risk memory potential field is set as a function that decays over time, specifically an exponential decay model. This means that as time passes, the impact of the event will gradually diminish until it disappears. This mechanism allows the vehicle to exhibit a cautious behavior pattern after experiencing a close call with an obstacle, by preferring a wider detour or a lower speed through the area if it needs to pass through again shortly after the event.

[0049] Trajectory generation control module is the final decision and execution link of the system, responsible for converting abstract potential field information into specific vehicle control instructions.

[0050] Total potential field superposition unit: This unit linearly superimposes the target attractive potential field, static repulsive potential field, dynamic interference potential field, rule-constrained potential field generated by the potential state field construction module, and the risk memory potential field generated by the risk memory management module. Through this process, all factors affecting vehicle decision-making, including targets, obstacles, rules, and historical risks, are unified into a single total potential field.

[0051] Potential gradient calculation unit: At the vehicle's current location, this unit calculates the gradient of the total potential field. The opposite direction of the gradient, i.e., the negative gradient of the potential energy, points to the direction in which the potential energy decreases most quickly. In a physical sense, this direction is the optimal combined force direction that the vehicle experiences in the current environment, balancing the attractive force towards the target and the repulsive force to avoid various obstacles.

[0052] Desired State Generation Unit: The calculated negative potential gradient is a two-dimensional or three-dimensional vector representing the desired motion trend. This unit decomposes (or decouples) this vector into components parallel to and perpendicular to the vehicle's current orientation.

[0053] The magnitude and direction of the parallel component determine whether the vehicle should accelerate or decelerate, and the desired longitudinal acceleration command is generated accordingly.

[0054] The magnitude and direction of the vertical component determine whether the vehicle should turn left or right and the magnitude of the turn, thereby generating the desired front wheel steering angle command.

[0055] Ultimately, these two decoupled control commands are sent to the vehicle's underlying actuators (such as the engine, braking system, and steering system) to achieve closed-loop control of the vehicle's trajectory.

[0056] In one specific embodiment, the environmental perception module in the system is the perception foundation of the entire system, responsible for providing accurate and reliable structured environmental information for the subsequent planning and control modules.

[0057] The environmental perception module mainly consists of a data acquisition unit and a semantic labeling unit.

[0058] The data acquisition unit is used to capture raw information about the vehicle's driving environment in all directions and all weather conditions. To ensure the robustness and redundancy of perception, the unit preferably adopts a multi-sensor fusion technology solution. Specifically, lidar, millimeter-wave radar and vision cameras can be deployed around the vehicle.

[0059] LiDAR (Light Detection and Ranging) generates high-density 3D point cloud data by emitting laser beams and receiving reflected signals, thereby accurately acquiring the geometric contours, dimensions, and positional information of surrounding obstacles. It has high ranging accuracy and is a core sensor for 3D environmental modeling.

[0060] Visual cameras are used to capture rich two-dimensional image data. Through image processing technology, they can identify the color, texture, and type information of objects, providing an important basis for subsequent semantic understanding.

[0061] Millimeter-wave radar has strong anti-interference capabilities and can still effectively detect targets even in adverse weather conditions such as dust, rain, snow, and fog common in mining areas. It can also directly measure the radial velocity of targets, providing direct input for judging the behavior of dynamic targets.

[0062] These different modalities of data are synchronized on timestamps and unified into the same vehicle coordinate system through coordinate transformation, forming a multi-dimensional data stream that includes point clouds, images, and radar target points, providing a comprehensive data foundation for subsequent processing.

[0063] The semantic labeling unit is the core of the environmental perception part of this system. It receives multi-dimensional data streams from the data acquisition unit and performs in-depth processing and parsing. Its ultimate goal is to output a list of obstacles with clear semantic category attributes. The workflow of this unit can be decomposed into three consecutive steps: target detection, multi-target tracking, and semantic attribute labeling.

[0064] During the target detection phase, the system processes the fused data to identify independent obstacle entities in the environment. For example, a deep learning-based algorithm can be used to perform 3D target detection on the LiDAR point cloud, directly outputting the 3D bounding box and initial position of the obstacle.

[0065] In the multi-target tracking phase, the system needs to associate obstacles detected within consecutive time frames to determine their unique identities and estimate their motion states. Preferably, extended Kalman filtering or unscented Kalman filtering methods can be used to track the state of each detected target, and the target's motion state vector x j At time k, it can be represented as a set of position and velocity information, and its state prediction process can be described by the following state-space equation:

[0066] x j =Fx k-1 +Bux k =Fx k-1 +Bu k-1 +w k-1 ;

[0067] Where F is the state transition matrix, describing how the motion state evolves from time k-1 to time k, x k-1 Let B be the state vector from the previous time step, and let B be the control input matrix. k-1 To control the input, w k-1 The process noise represents the uncertainty of the model. Through this tracking process, the system can obtain the accurate trajectory and velocity information of each obstacle in continuous time.

[0068] In the semantic attribute labeling step, after obtaining the stable motion trajectory of the obstacle, the unit assigns a semantic category attribute to each tracked obstacle according to the preset classification rules.

[0069] For the calibration of static obstacles, the system analyzes their estimated velocity values ​​within a time window ΔT over a past period. If the average velocity value is consistently lower than a preset low-speed threshold, the system will determine the obstacle's speed. v If the obstacle is marked as a static obstacle, this applies to roadside rocks, parked vehicles, and equipment.

[0070] For dynamic obstacles, the system will further analyze the stability and predictability of their movement.

[0071] For predictable dynamic obstacles, the system evaluates the consistency of their motion model. For example, within the framework of Kalman filtering, the system calculates the residual between the predicted state and the actual observation. If the motion residual of the dynamic obstacle remains within a small range over multiple consecutive cycles, and the variance of its higher-order motion quantities such as acceleration and angular velocity is low, it indicates that its motion mode is stable and follows Newton's laws of motion. In this case, it can be labeled as a predictable dynamic obstacle, which is applicable to other vehicles traveling on conventional routes on mining roads.

[0072] For unpredictable dynamic obstacles, the criterion is the high degree of uncertainty in their motion, which can be quantified in the state estimation covariance matrix P. k Regarding the magnitude of the covariance matrix, it describes the degree of uncertainty in the state estimate, and its update process can be expressed as:

[0073] P k =(IK k H)P k∣k-1 ;

[0074] Among them, P k∣k-1 To predict covariance, K k Let H be the Kalman gain and H be the observation matrix. When the target's trajectory undergoes frequent abrupt changes (such as rapid acceleration or sharp turns), it will lead to an increase in the covariance P of its state estimation. k If the trace Tr(P) of the matrix increases, k The uncertainty threshold θ is exceeded. unc If an obstacle is identified as an unpredictable dynamic obstacle, or if its movement pattern does not conform to any predefined motion model, then the obstacle is identified as an unpredictable dynamic obstacle. Such obstacles include on-site workers and engineering machinery with irregular behavior.

[0075] Finally, the environment perception module will output a processed structured environment model. This model not only contains the precise location, velocity, and size physical information of all obstacles, but more importantly, it adds a semantic label to each obstacle, such as "static," "predictable dynamic," or "unpredictable dynamic." This semantically informative environment model enables the subsequent potential field construction module to build differentiated potential fields that are more in line with the actual interaction scenario for obstacles with different attributes, thus providing a foundation for achieving more intelligent and safer path planning and obstacle avoidance control.

[0076] In one specific embodiment, the potential field construction module is the decision-making core of the entire system. It receives structured environmental information with semantic tags from the environmental perception module and constructs a potential field that can comprehensively and multidimensionally characterize the passability of the vehicle's surrounding environment based on this information. This potential field is ultimately used to guide the vehicle's path decision.

[0077] The potential field construction module is specifically composed of a target attraction potential field generation unit, a static repulsion potential field generation unit, a dynamic interference potential field generation unit, and a rule constraint potential field generation unit.

[0078] The target attraction potential field generation unit is designed to provide the vehicle with a global traction force directed toward the final target. In this embodiment, the unit generates a target attraction potential field whose potential energy is related to the distance from the vehicle's current position to the preset target point.

[0079] Preferably, the attractive potential field U att It can be constructed as a quadratic function, and its mathematical expression is as follows:

[0080]

[0081] Where q represents the vehicle's current position coordinates, q goal k represents the preset target point coordinates. att It is a positive constant and serves as the gain coefficient of the attractive potential field, used to adjust the magnitude of gravity.

[0082] Under the influence of this potential field, the farther the vehicle is from the target point, the higher its potential energy value, and thus it will be subject to an attractive force pointing towards the target point, the magnitude of which is proportional to the distance. This constitutes the basic driving force for the vehicle's movement.

[0083] The static repulsive potential field generation unit enables the vehicle to effectively avoid stationary, fixed obstacles in the environment. This unit receives object information identified as "static obstacles" by the environmental perception module and establishes corresponding repulsive potential fields around these obstacles. The characteristic of this potential field is that the potential energy value is high in the vicinity of the obstacle and gradually decreases with increasing distance from the obstacle, until it decays to zero outside a certain range of influence. The preferred repulsive potential field U... rep The implementation method is as follows:

[0084]

[0085] Where, k rep ρ(q,q) is a positive constant and serves as the gain coefficient for the repulsive potential field. obs ) represents the vehicle's current position q and the obstacle q. obs The shortest distance between them, d0 is the preset radius of influence of the obstacle. When the distance between the vehicle and the obstacle is greater than this radius, the obstacle will no longer have a repulsive effect on the vehicle. This potential field construction method ensures that the vehicle will naturally maintain a safe distance from static obstacles when planning its path.

[0086] The dynamic interference potential field generation unit constructs differentiated interference potential fields based on the different semantic attributes of dynamic obstacles, so as to achieve a more refined and predictive response to dynamic environments.

[0087] For objects identified as "predictable dynamic obstacles" by the environmental perception module, considering the stability and predictability of their motion state, the unit constructs an anisotropic interference potential field. This potential field is no longer a simple circular distribution centered on the obstacle, but is stretched along the direction of its predicted driving trajectory, while being correspondingly shortened to its sides and rear. This asymmetric potential field shape takes into account the space area that the obstacle will occupy in the near future, allowing the vehicle to avoid it smoothly and in advance, rather than reacting when it is close to it. This potential field is established by introducing the relative velocity vector between the vehicle and the dynamic obstacle when calculating the effective action distance between them, so that the repulsive force is not only related to the relative position, but also closely related to its relative motion trend.

[0088] In contrast, for objects labeled as "unpredictable dynamic obstacles," given the high degree of randomness and uncertainty in their motion, the unit adopts a more conservative strategy. It generates a larger and stronger isotropic repulsive potential field centered on the obstacle's current position. This potential field is uniformly distributed in all directions. Its mathematical model can follow the formula for the aforementioned static repulsive potential field, but it employs a larger radius of influence d0 and a larger gain coefficient k. rep The purpose of this approach is to establish a larger safety buffer zone for such high-risk targets, ensuring that vehicles maintain sufficient safety margins against them under any circumstances to cope with sudden movements in any direction.

[0089] The rule-constrained potential field generation unit is responsible for transforming the physical boundaries and traffic rules within the mining area into a potential field model. For example, based on high-precision map data, the edges of roads and prohibited work areas are set as "potential energy walls" with extremely high potential energy values. When a vehicle approaches these boundaries, the repulsive force it experiences will increase dramatically, thus preventing vehicles from deviating from the drivable area or entering the restricted area at the source of path planning, ensuring the compliance and safety of vehicle driving.

[0090] Through the collaborative work of the above units, the potential field construction module integrates multi-source information from environmental perception into a unified and structured comprehensive potential field model. This model not only includes the attraction to the target, but also finely distinguishes the repulsive effect of obstacles with different attributes, and incorporates the constraints of driving rules, providing a rich and computationally feasible decision basis for the subsequent trajectory generation and control module.

[0091] In one specific embodiment, the core function of the risk memory management module is to give the system the ability to learn from historical high-risk events and adapt in the short term, thereby exhibiting a more cautious and safer behavior pattern in repetitive or potentially dangerous scenarios.

[0092] Logically, this module consists of a risk event triggering unit and a memory potential field generation unit.

[0093] The risk event triggering unit is designed to continuously monitor the vehicle's driving status and accurately identify high-risk events that need to be remembered. In this embodiment, the unit achieves this function by monitoring the potential energy value of the vehicle's current position in the total potential energy field in real time. The total potential energy field is the superposition of various sub-potential fields generated by the aforementioned potential field construction module.

[0094] During the vehicle's operation, the total potential energy value U total It is a comprehensive quantitative indicator of the risk level of the current environment in which the vehicle is located. Under normal driving conditions, this value will remain within a relatively stable range. However, when the vehicle enters the influence zone of an obstacle very close to it due to a sudden situation (such as emergency avoidance of a suddenly appearing obstacle at close range), the repulsive potential energy of that area will increase sharply and instantaneously, resulting in a peak value of the total potential energy.

[0095] The risk event triggering unit triggers the event by comparing the real-time total potential energy value with a pre-set risk threshold η. risk The system makes a judgment based on comparisons. A high-risk event is determined to have occurred when the following conditions are met:

[0096] U total (q vehicle ,t)>η risk ;

[0097] Among them, U total It is the total potential energy value, q vehicle Let t be the current position of the vehicle, t be the current time, and η be the risk threshold. risk These are parameters that can be adjusted according to different work scenarios and safety level requirements.

[0098] Once a high-risk event is triggered, the unit will immediately record key information about the event, including the geographical location q where the event occurred. risk and trigger time t event This information is then passed to the memory potential field generation unit as the basis for generating risk memories.

[0099] The memory potential field generation unit, after receiving event information from the risk event triggering unit, has its core task of determining the location q where the event occurred. risk At this point, a temporary, repulsive risk memory potential field U is generated.mem .

[0100] The construction of this risk memory potential field has two dimensional characteristics: spatial distribution characteristics and time decay characteristics.

[0101] In terms of spatial distribution, this potential field is similar to the aforementioned static repulsive potential field, with the risk event occurring at point q. risk Centered on the risk point, its potential energy decreases outwards, ensuring local coverage of the risk point and affecting vehicles passing through the vicinity.

[0102] In the time decay characteristic, the potential energy value of the risk memory potential field is preferably set as a time function with exponential decay characteristics. This means that the strength of the memory potential field will not exist permanently, but will gradually weaken and eventually disappear over time. The purpose of this design is to simulate a short-term memory, which can keep the vehicle alert to the area for a period of time after the event, and avoid unnecessary permanent interference to the subsequent normal path planning due to outdated risk information.

[0103] The risk memory potential field U mem The mathematical expression can be concretized as follows:

[0104]

[0105] Where: t is the current time, and t≥t event A0 represents the risk memory potential field at the initial time t. event The peak intensity of f(||qq) can be correlated with the peak potential energy at the triggering event. τ is a positive constant, a time decay constant used to control the rate of memory forgetting. The larger the value of τ, the longer the memory potential field lasts. risk ||) is a function describing the spatial distribution of the potential field, which varies with the vehicle's current position q and the risk point q. risk It decreases as the distance between them increases.

[0106] The risk memory potential field generated in this way is dynamically and temporarily added to the calculation of the total potential field. When the vehicle needs to approach or pass through the risk area again in a short period of time, this additional U... mem This will increase the overall potential energy of the area, causing vehicles to naturally tend to choose wider detours or pass through cautiously at lower speeds when making route decisions, exhibiting risk-avoidance behavior.

[0107] In a specific embodiment, the trajectory generation control module, as the final execution link of the entire adaptive path planning and obstacle avoidance control system, has the core task of transforming the abstract potential field information constructed by the previous module into specific physical control commands that can be understood and executed by the underlying controller of the mining heavy-duty transport vehicle.

[0108] In its specific implementation, this module can be composed of a total potential energy field superposition unit, a potential energy gradient calculation unit, and a desired state generation unit.

[0109] The function of the total potential field superposition unit is to integrate all factors that affect vehicle driving decisions and form a single, unified decision basis. Specifically, this unit will algebraically superimpose the various sub-potential fields generated by the potential field construction module and the temporary risk memory potential field generated by the risk memory management module.

[0110] The resulting total potential energy field U total It can be represented as:

[0111]

[0112] Where q represents the vehicle's position in the planned space, and U att Attracting potential fields to the target The sum of the repulsive potential fields generated by all static obstacles. U is the sum of the interference potential fields generated by all dynamic obstacles. rule For the rule-constrained potential field, and U mem It is a risk memory potential field that only exists when a risky event is triggered.

[0113] Through this superposition process, all information about the vehicle's environment, including targets, various obstacles, driving rules, and historical risks, is uniformly mapped into the total potential energy field. This total potential energy field constitutes a complete potential energy map that describes the drivability of the environment.

[0114] After obtaining the total potential energy field, the potential energy gradient calculation unit is responsible for determining the optimal instantaneous motion direction for the vehicle at its current position. In this embodiment, the optimal direction is determined to be the direction in which the potential energy decreases the fastest in the total potential energy field.

[0115] In mathematics, the direction in which potential energy decreases the most is the direction of the negative gradient of the potential energy field. Therefore, this unit calculates the negative gradient of the total potential energy field at the vehicle's current position qq to obtain the virtual resultant force vector F that drives the vehicle's motion. des (q), its calculation formula is as follows:

[0116]

[0117] in, It's the gradient operator, and the calculated vector F des (q) Its direction points to the optimal driving direction after considering all attractive and repulsive effects, and its magnitude represents the "urgency" of driving in that direction. q represents the vehicle's current position in the planned space, U total (q) represents the total potential energy of the vehicle at position q. Gradient vector Component representation in a two-dimensional Cartesian coordinate system.

[0118] The desired state generation unit connects the upper-level planning and the lower-level control. It receives the virtual resultant force vector F output by the potential energy gradient calculation unit. des (q) and converts it into control commands that are physically executable by the vehicle. In a preferred embodiment, the unit achieves this conversion by decoupling the desired motion state.

[0119] Specifically, this unit first converts the two-dimensional virtual resultant force vector F des (q) Projected onto the vehicle's own coordinate system, which is usually based on the vehicle's center of mass as the origin, with the direction of the vehicle's front as the longitudinal axis and the direction perpendicular to the vehicle's front as the transverse axis.

[0120] The longitudinal component obtained after projection directly reflects the vehicle's expected motion trend in the forward direction in terms of its magnitude and sign. If the component is positive, it means that the vehicle should accelerate forward; if it is negative, it means that it should decelerate or even brake. Based on the magnitude of this component and the vehicle's current speed, the unit calculates the expected longitudinal acceleration.

[0121] Meanwhile, the magnitude and sign of the lateral component obtained after projection determine the vehicle's steering requirements. If the component points to the left of the vehicle, it means that the vehicle needs to turn left; otherwise, it needs to turn right. The magnitude of the component is proportional to the required steering angle. Based on this lateral component and considering the vehicle's kinematic or dynamic model (such as the Ackermann steering geometry model), the unit calculates the desired front wheel steering angle.

[0122] Through this decoupling method, the motion trend vector can be clearly converted into two independent, orthogonal control quantities: the desired longitudinal acceleration and the desired front wheel steering angle. These two control quantities are instructions that the vehicle's underlying actuators (such as the engine management system, braking system, and steering system) can directly receive and execute, thus ultimately realizing complete closed-loop control from environmental perception to vehicle physical motion.

[0123] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An adaptive path planning and obstacle avoidance control system for heavy-duty mining transport vehicles, characterized in that, include: The environmental perception module is configured to perceive the vehicle's surrounding environment in real time and to perform semantic attribute labeling on the perceived obstacles. The potential field construction module is configured to construct a multi-dimensional potential field representing accessibility based on the semantic attribute calibration results. The risk memory management module is configured to monitor the multi-dimensional potential field to identify high-risk events and generate a risk memory potential field with time decay characteristics based on the high-risk events. The trajectory generation control module is configured to superimpose the multi-dimensional potential field with the risk memory potential field to form a total potential energy field, and generate vehicle control commands based on the total potential energy field.

2. The adaptive path planning and obstacle avoidance control system for a heavy-duty mining transport vehicle according to claim 1, characterized in that, The environment sensing module includes: The data acquisition unit is used to collect multimodal data of the environment surrounding the vehicle. The semantic labeling unit is used to label the semantic attributes of obstacles in the multimodal data in order to determine the category attributes of the obstacles.

3. The adaptive path planning and obstacle avoidance control system for a heavy-duty mining transport vehicle according to claim 1, characterized in that, The potential field construction module includes: The target attraction potential field generation unit is used to generate a target attraction potential field pointing to a preset target point; The static repulsive potential field generation unit is used to generate a static physical repulsive potential field generated by a static obstacle. The dynamic interference potential field generation unit is used to generate a dynamic interference potential field generated by dynamic obstacles. The rule-constrained potential field generation unit is used to generate the rule and constraint potential fields generated by the road boundaries.

4. The adaptive path planning and obstacle avoidance control system for a heavy-duty mining transport vehicle according to claim 1, characterized in that, The risk memory management module includes: A risk event triggering unit is used to monitor the multi-dimensional potential field and determine a risk event when the potential energy value of the multi-dimensional potential field exceeds a preset risk threshold. The memory potential field generation unit is used to generate a risk memory potential field whose potential value decays over time based on the location and time information of the risk event.

5. The adaptive path planning and obstacle avoidance control system for a heavy-duty mining transport vehicle according to claim 1, characterized in that, The trajectory generation control module includes: The total potential energy field superposition unit is used to superimpose the multi-dimensional potential field with the risk memory potential field to form the total potential energy field; Potential energy gradient calculation unit, used to calculate the negative potential energy gradient of the vehicle's current position in the total potential energy field; The desired state generation unit is used to generate the desired motion state of the vehicle based on the negative potential gradient.

6. The adaptive path planning and obstacle avoidance control system for a heavy-duty mining transport vehicle according to claim 2, characterized in that, The semantic labeling unit is used to determine the category attribute of the obstacle as one of: static obstacle, predictable dynamic obstacle, and unpredictable dynamic obstacle.

7. The adaptive path planning and obstacle avoidance control system for a heavy-duty mining transport vehicle according to claim 3, characterized in that, The dynamic interference potential field generation unit is also used for: For the predictable dynamic obstacle, an anisotropic potential field distribution extending along the predicted trajectory direction of the predictable dynamic obstacle is generated; For the unpredictable dynamic obstacle, an isotropic potential field distribution centered on the current position of the unpredictable dynamic obstacle is generated.

8. The adaptive path planning and obstacle avoidance control system for a heavy-duty mining transport vehicle according to claim 4, characterized in that, The memory potential field generation unit is used to set the potential energy value of the risk memory potential field as a time function with exponential decay characteristics.

9. The adaptive path planning and obstacle avoidance control system for a heavy-duty mining transport vehicle according to claim 5, characterized in that, The desired state generation unit is used to decouple the desired motion state into desired longitudinal acceleration and desired front wheel steering angle.

10. The adaptive path planning and obstacle avoidance control system for a heavy-duty mining transport vehicle according to claim 1, characterized in that, The potential energy gradient calculation unit is used to determine the negative potential energy gradient as the direction in which the potential energy decreases the fastest in the total potential energy field.

Images