Parking lot inspection robot path planning method

Abstract

The invention relates to the technical field of intelligent robot navigation and control, and discloses a parking lot inspection robot path planning method which comprises the following steps: step 1, acquiring parking lot environment information and constructing a multi-dimensional heterogeneous potential site map which comprises a physical barrier layer, a communication field intensity distribution layer and a risk probability distribution layer, meanwhile, global control parameters of path planning are reversely deduced and set on the basis of preset operation indexes, and the global control parameters at least comprise opportunity constraint confidence coefficients and stable gating threshold values. A global cost function of nonlinear coupling of risk revenue and communication quality is adopted, and multi-dimensional heterogeneous potential field map construction is combined, so that the technical effects of dynamically and preferentially planning a high-risk area and giving consideration to signal coverage are achieved, and quick response and value maximization to a temperature rise or smoke dangerous case in a finite time window are realized; the defect that high-risk hidden danger leak detection is caused by lack of elasticity of an existing static inspection route is overcome.

Description

A Path Planning Method for Parking Lot Inspection Robots Technical Field

[0001] This invention relates to the field of intelligent robot navigation and control technology, specifically to a path planning method for a parking lot inspection robot. Background Technology

[0002] Due to accelerated urbanization and the continuous growth of motor vehicle ownership, large underground parking lots have become an indispensable supporting facility for modern urban buildings. At the same time, the widespread adoption of new energy vehicles has gradually transformed the function of parking lots from simple vehicle parking to energy replenishment centers. However, underground parking lots suffer from enclosed environments, complex spatial structures, poor air circulation, and high-density vehicle parking, which significantly increases the risks of fire hazards, toxic gas accumulation, and vehicle scratch disputes. In particular, the potential for battery thermal runaway during the charging process of new energy vehicles places extremely high demands on the timeliness of parking lot safety management.

[0003] To ensure parking lot operational safety, reduce the labor intensity of manual inspections, and mitigate the health risks of harsh environments for security personnel, the use of autonomous mobile robots for 24 / 7 inspections has become a mainstream trend in the industry. Inspection robots are typically equipped with LiDAR, vision cameras, thermal imagers, and gas sensors, responsible for environmental monitoring, parking space inventory, and anomaly warnings. Efficient and reliable path planning is the core technology for intelligent operation of inspection robots, directly determining their response speed to emergencies and system stability during long-term operation.

[0004] Existing route planning technologies can solve basic point-to-point navigation problems, but conventional technical solutions still have significant limitations when facing the specific high-risk, high-dynamic, and weak communication environment of underground parking lots, and are not easy to meet the needs of refined operation.

[0005] First, existing inspection robots typically use fixed routes for traversal, making it impossible to dynamically adjust the inspection order in response to sudden, high-risk events. Under the constraints of limited power and time, static planning leads to delayed responses to abnormal temperature rises or smoke hazards, resulting in missed inspections of high-risk locations and failing to maximize the value of the inspection task.

[0006] Second, the charging area has narrow passageways and vehicles often park temporarily, causing dynamic obstacles to frequently block the path. Conventional path planning algorithms are prone to getting stuck in local optima when facing dynamic environmental changes, leading to repeated path recalculations and oscillations. This causes the robot to wander in place, reducing passage efficiency and increasing the risk of collisions with vehicles.

[0007] Third, underground parking lots have communication blind spots and lack dynamic reserve of remaining power. Existing solutions do not take signal quality and energy redeployment into planning considerations. When robots enter signal blind spots, alarm information cannot be transmitted in real time, or they may be unable to return to the charging station due to power depletion, resulting in systemic failures such as loss of inspection data and abnormal interruption of tasks. Summary of the Invention

[0008] This invention proposes a path planning method for parking lot inspection robots. By constructing a multi-dimensional potential field coupled with risk communication, and applying stable gating and energy withdrawal budget, it achieves priority inspection of high-risk blind spots, smooth path anti-shaking, and communication endurance safety.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a path planning method for a parking lot inspection robot, comprising:

[0010] Step 1: Obtain parking lot environmental information and construct a multidimensional heterogeneous potential field map. The multidimensional heterogeneous potential field map includes a physical obstacle layer, a communication field strength distribution layer, and a risk probability distribution layer. At the same time, based on preset operation indicators, reverse the calculation and set global control parameters for path planning. The global control parameters include at least chance constraint confidence and stability gate threshold.

[0011] Step 2: Construct a path planning model based on a multidimensional heterogeneous potential field map. The path planning model selects a set of feasible paths that meet the probability safety threshold based on the confidence level of the chance constraint. Within the set of feasible paths, a global cost function with nonlinear coupling characteristics is constructed. The global cost function establishes a coupling relationship between the revenue value of the risk probability distribution layer and the availability coefficient of the communication field strength distribution layer.

[0012] Step 3: Based on the global cost function, search for and generate an initial inspection path in the set of feasible paths, and decompose the global constraints and pass them down to the path nodes to generate a constrained state node sequence containing time window constraints and energy withdrawal budget.

[0013] Step 4: Control the robot to perform inspection tasks along the initial inspection path, and use multi-dimensional sensors to perform real-time monitoring of multi-source consistency. When a sudden environmental change is detected that meets the multi-source logic triggering conditions, a replanning request is generated.

[0014] Step 5: In response to the replanning request, generate candidate replanning paths and perform stable gating decisions. Calculate the cost optimization ratio between the candidate replanning path and the remaining segments of the current path. When the cost optimization ratio exceeds the stable gating threshold, switch the current path to the candidate replanning path.

[0015] Step 6: Monitor the robot's remaining battery power in real time during path execution, compare the remaining battery power with the energy withdrawal budget corresponding to the current position, and when the remaining battery power reaches the energy withdrawal budget, forcibly interrupt the current task and plan a safe path back to the charging station.

[0016] Preferably, step one further includes:

[0017] Sub-step Using LiDAR and odometer data, the parking lot environment is scanned to construct a two-dimensional gridded physical obstacle layer map. The parking lot space is divided into sets of grid cells of the same size. For each grid cell in the set, the occupancy probability value of that grid cell is calculated. Furthermore, the physical state of the grid cell is determined using a binarization decision formula. The binarization determination formula is as follows:

[0018] ,

[0019] in, The coordinates of the grid cells, This represents the occupancy probability at that coordinate. The preset obstacle detection threshold, It is an indicator function;

[0020] when When the value is 0, it indicates that the grid is a physical obstacle that cannot be passed through;

[0021] when When the time is right, it indicates that the grid is a free area;

[0022] Sub-step Based on the two-dimensional rasterized physical obstacle layer map, each raster cell marked as a free region is traversed, and the communication signal strength and data packet loss rate at that location are collected. The communication field strength availability coefficient of that raster cell is calculated using a linear weighted normalization algorithm. A communication field strength distribution layer is generated, and the formula for the linear weighted normalization algorithm is as follows:

[0023] ,

[0024] in, This is the received signal strength indication value at this coordinate. and The preset lower and upper limits of signal strength. This represents the packet loss rate at that coordinate. and These are the signal strength weighting coefficient and the packet loss rate weighting coefficient;

[0025] Sub-step Based on grid cell coordinates, the frequency of historical fault records at that coordinate is correlated with real-time thermal imaging temperature data, utilizing... The activation function is used to construct a nonlinear risk assessment model to calculate the risk probability and return value of the grid cell. A risk probability distribution layer is generated, and the calculation formula of the nonlinear risk assessment model is as follows:

[0026] ,

[0027] in, This is the normalized real-time temperature rise data at this coordinate. This represents the frequency of historical occupancy disputes or equipment malfunctions at this coordinate location. The weighting is based on the impact of temperature rise. Weighting for the impact of historical faults. For model bias terms, The base of the natural logarithm; when calculating the risk probability return value of this grid cell. The closer This indicates that the inspection value of that location is higher;

[0028] Sub-step Based on the communication field strength distribution layer and the risk probability distribution layer, the global control parameters for path planning are inferred from the preset parking lot operation indicators, and the confidence level is constrained by the missed detection rate constraint formula. Furthermore, the stable gating threshold is calculated using the switching frequency constraint formula. The formula for the missed detection rate constraint is: ≤ ,

[0029] in, The maximum allowable missed detection rate for the parking lot. This refers to the average number of high-risk nodes covered in a single inspection mission.

[0030] The switching frequency constraint formula is as follows: ,

[0031] in, This represents the maximum number of path replanning attempts allowed per unit of time. This represents the average number of reprogramming iterations measured so far. To minimize the change in effective cost, This represents the historical average total path cost.

[0032] Preferably, step two further includes:

[0033] Sub-step Based on the probability constraint confidence level And the aforementioned two-dimensional rasterized physical obstacle layer map, introducing a dynamic obstacle prediction model, for any candidate path within the planning space The system identifies dynamic obstacles within the area traversed by the path, establishes the probability density function of the future positions of the dynamic obstacles using a Gaussian distribution function, and calculates candidate paths. Cumulative collision risk probability within the future time window ;Will Meet the conditions ≤ Path retention, removal > High-risk paths are selected, and a set of feasible paths that meet the probability safety threshold is identified. ;

[0034] Sub-step In the set of feasible paths Within, define adjacent path nodes. and Edge cost function The edge cost function Distance and time cost Energy consumption cost item and risk communication coupling cost term The edge cost function is constructed using a weighted summation method. The calculation formula is:

[0035] ,

[0036] in, , and These are distance weights, energy consumption weights, and coupling weights.

[0037] Sub-step For the risk communication coupling cost term in the edge cost function Risk probability return value based on the risk probability distribution layer and the communication field strength availability coefficient of the communication field strength distribution layer A nonlinear inverse proportional coupling model is established, and the calculation formula for the nonlinear inverse proportional coupling model is as follows:

[0038] ,

[0039] in, For nodes to The average risk-reward value on the path segment. This is the average communication field strength availability coefficient on this path segment. For communication sensitivity index, To prevent extremely small positive numbers with a denominator of zero;

[0040] when When it approaches zero, the risk communication coupling cost term The absolute value of approaches infinitesimal, causing the total cost of the path segment to increase dramatically, thereby suppressing the generation of high-risk but communication-free areas in the global cost function.

[0041] Preferably, step three further includes:

[0042] Sub-step Based on the set of feasible paths and edge cost function A heuristic search algorithm is used to traverse and solve the set of feasible paths to obtain the node sequence that minimizes the cumulative total generation value as the initial inspection path. The formula for calculating the cumulative total cost is as follows: ,

[0043] in, This represents the total number of path nodes. and For adjacent nodes on the path, The edge cost function value between adjacent path nodes;

[0044] Sub-step Based on the initial inspection path Perform forward time extrapolation to generate time window constraints, and use time recursion formulas to calculate the estimated arrival time of each node on the path. And a time tolerance is set based on the expected arrival time. Define the time window constraint for this node. The time recursion formula and the time window constraint are defined as follows:

[0045] ,

[0046] ,

[0047] in, This represents the arrival time of the previous node. This represents the Euclidean distance between adjacent nodes. To improve the robot's planned inspection speed, This refers to the time the robot spends performing its task at the previous node. It is a closed interval;

[0048] Sub-step Based on the initial inspection path The energy budget is calculated by using the charging pile coordinates recorded in the physical obstacle layer map, performing reverse energy budget decomposition to generate an energy withdrawal budget, and then using a safe withdrawal model to calculate the minimum retained power required for each node to return to the charging pile, which is defined as the energy withdrawal budget. The calculation formula for the safety pullback model is as follows:

[0049] ,

[0050] in, For the current node To the Taiwan charging pile The shortest path distance, This represents the average energy consumption rate of the robot per unit distance. Energy safety redundancy factor;

[0051] Sub-step Constrain the time window With energy drawdown budget Mapped to the initial inspection path The corresponding nodes are combined to generate a sequence of constrained state nodes. Each state node in the sequence The data structure is expressed as:

[0052] ,

[0053] in, These are the spatial coordinates of the node.

[0054] Preferably, step four further includes:

[0055] Sub-step Analyze the constrained state node sequence The robot is controlled to move sequentially to each state node in the sequence using a trajectory tracking algorithm. spatial coordinates Furthermore, infrared thermal imaging data is collected simultaneously during the driving process. Visible light visual images and lidar point cloud data Construct a real-time multi-dimensional sensing set for the current moment. The expression for the real-time multidimensional sensing set is:

[0056] ,

[0057] in, The current time;

[0058] Sub-step Based on the real-time multidimensional sensing set Perform multi-source consistent real-time monitoring and extract regional temperature rise characteristic values ​​from infrared thermal imaging data. and confidence level of smoke texture in visible light visual images A multi-source consistency determination formula is constructed using logical AND operations to calculate the triggering identifier of high-risk events. The multi-source consistency determination formula is as follows:

[0059] ,

[0060] in, The preset abnormal temperature rise threshold, The preset confidence threshold for smoke recognition is set when both abnormal temperature rise and smoke characteristics simultaneously meet the threshold condition. The value is This indicates that it has been confirmed as a genuine high-risk event; otherwise, it is... This is used to filter out false alarms caused by noise from a single sensor;

[0061] Sub-step Based on the lidar point cloud data in the real-time multi-dimensional sensing set Detect dynamic obstacles located ahead of the current driving path. Combined with the time window constraint Calculate the probability of removing the dynamic obstacle within the remaining available time window. Furthermore, the channel blocking trigger flag is determined using the blocking determination formula. The blocking determination formula is as follows:

[0062] ,

[0063] in, For the current target node The upper limit of the time window, The remaining available passage time, The parameters are Poisson distribution parameters fitted based on the historical movement speed of the obstacle. This is a preset minimum passage probability threshold; when the calculated removal probability is lower than this threshold, Set as This indicates that the obstacle cannot be removed within the specified time window, and the path is in a state of substantial blockage.

[0064] Sub-step Real-time monitoring of the high-risk event trigger identifiers With the channel blocking trigger identifier Generate replanning requests using replanning trigger logic. The replanning triggering logic is as follows:

[0065] ,

[0066] in, For logical OR operation; when replanning is requested. If true, immediately pause the current trajectory tracking task and send the robot's current state and trigger type to the decision layer.

[0067] Preferably, step five further includes:

[0068] Sub-step Based on the robot's current state included in the replanning request, determine the robot's current position coordinates. and the set of remaining target nodes in the current path that have not yet been visited. Based on the global cost function, for Perform a heuristic path search to generate a connection. Candidate replanning paths with all remaining target nodes The cumulative cost of the candidate replanning path is calculated and defined as the cost of the new path. The calculation formula is:

[0069] , in, Represents the edges on the path. Let $\frac{ ...

[0070] Sub-step Extract the currently executing path from the path containing the... Unexecuted segments of the old path leading to the destination are defined as the remaining segments of the old path. The cumulative cost of the road segment is calculated using the global cost function and defined as the remaining cost of the old path. The calculation formula is:

[0071] ,

[0072] in, The edges are on the remaining segment of the old path. To account for the additional blocking penalty that may occur due to adhering to the original path, when the trigger type is channel blocking. Take the preset positive value, otherwise take the default positive value. ;

[0073] Sub-step Obtain the stable gating threshold. Utilizing the cost of the new path With the remaining cost of the old path Construct a hysteresis comparison inequality and calculate the cost optimization ratio of path switching. The logic for determining the hysteresis comparison inequality is as follows:

[0074] ≥ ,

[0075] in, The new path represents the increase in revenue compared to the old path; when this inequality holds, it is determined that the replanning path has a significant advantage, and a path switching instruction is generated. If the inequality does not hold, generate a command to maintain the original path. To suppress path jitter caused by small gains;

[0076] Sub-step In response to the path switching command Using the candidate replanning path Replace the current path, and based on The topology is recursively recalculated to determine the estimated arrival time and energy consumption of each node, updating the time window constraints and energy withdrawal budget in the constrained state node sequence, and generating an updated state node sequence. For robots to perform.

[0077] Preferably, step six further includes:

[0078] Sub-step The battery management system on the robot chassis reads the current terminal voltage and discharge current of the battery in real time, and uses a Kalman filter algorithm combining the ampere-hour integral method and the open-circuit voltage method to estimate the remaining power at the current moment. At the same time, obtain the robot's current positioning coordinates. In the constrained state node sequence Perform a nearest neighbor search to match the current best associated node. The formulas for estimating the remaining power and matching the optimal associated node are as follows:

[0079] ,

[0080] ,

[0081] in, This is the initial fully charged capacity. For battery discharge efficiency, for Instantaneous current at a given moment These are the coordinates of each node in the sequence;

[0082] Sub-step The optimal associated node Extract the corresponding energy drawdown budget from the data structure. Define the energy safety margin factor Construct an energy meltdown determination inequality to calculate energy crisis indicators in real time. The logic of the energy melting determination inequality is as follows:

[0083] ,

[0084] in, The percentage reserved to withstand sudden high energy consumption during the return trip;

[0085] When the inequality holds Set as This indicates that the current battery level has reached the physical limit of the charging station, and the system enters an irreversible forced recharging state.

[0086] Sub-step , in response to determination In this state, immediately trigger the highest priority task interruption command, freeze the current inspection task progress, and traverse all available charging pile sets in the physical obstacle layer map. Calculate the actual travel cost from the current location to each charging station, and select the target charging station with the lowest cost. And plan a safe return-to-charge path The selection formula for the target charging station is as follows:

[0087] ,

[0088] in, From current position to number The distance between charging stations This represents the probability of the charging station being currently occupied. and The penalty weights are distance and occupancy status; the robot follows the safe recharge path. Drive until docking is complete.

[0089] Preferably, the sub-step In this context, the Gaussian distribution function is constructed based on the center position and moving velocity vector of the dynamic obstacle, and is used to describe the dynamic obstacle at future moments. Appearing in location probability density The calculation formula is:

[0090] ,

[0091] in, For a moment The predicted center coordinates of a dynamic obstacle are obtained by linear extrapolation based on the obstacle's current velocity. Calculated; and Let be the standard deviation of the positional uncertainty along the coordinate axis, and given the prediction time. The increase is monotonically increasing; The positional correlation coefficient; the cumulative collision risk probability For path All nodes corresponding to The sum of the integrals of the values.

[0092] Preferably, the sub-step In the context of task execution dwell time It is not a fixed constant, but a variable dynamically calculated based on the data in the risk probability distribution layer. The calculation formula is as follows:

[0093] ,

[0094] in, Minimum dwell time for basic inspections The maximum permitted time for key inspections, For nodes Risk probability and return value at the location, This is a risk sensitivity index; the formula stipulates that for high-risk areas, the robot automatically extends its stay time to conduct in-depth scanning, while for low-risk areas, it maintains the basic passage time to optimize the efficiency of time window allocation.

[0095] Preferably, the sub-step In this context, the heuristic search algorithm employs an improved method. Algorithm, the heuristic function used in the algorithm Introducing communication availability constraints to guide the search direction and avoid communication blind spots, the heuristic function The calculation formula is:

[0096] ,

[0097] in, For the current node To the target node Euclidean distance, This represents the communication field strength availability coefficient obtained by the current node from the communication field strength distribution layer. For communication guidance weighting coefficients, To prevent extremely small positive numbers with a denominator of zero; when the search path enters a communication dead zone, Reduced Increasing the number of nodes allows the algorithm to prioritize expanding nodes with good communication quality.

[0098] The advantage of step one is that by integrating historical data from lidar, communication signals, and thermal imaging, a multi-dimensional digital foundation is constructed that includes physical obstacles, communication blind spots, and accident hotspots. Furthermore, abstract parking lot operation indicators are transformed into specific control parameters, enabling path planning to avoid physical obstacles and to quantitatively evaluate the weight of communication quality and inspection value from a global perspective.

[0099] The benefits of step two are: establishing an evaluation system that non-linearly couples risk and reward with communication quality; forcibly eliminating high-risk but non-signal-covered invalid areas when selecting paths; ensuring that the planned paths can guide robots to prioritize accident-prone locations for investigation; and ensuring that there are communication links at the inspection locations that can transmit alarm images in real time. This effectively avoids the operational risks of robots losing contact in blind spots or failing to upload important monitoring data due to the pursuit of high inspection coverage.

[0100] The benefits of step three are as follows: by using spatiotemporal constraint decomposition technology, the macroscopic global path is transformed into a microscopic executable state sequence, and the allowable time window and the mandatory reserved power limit are precisely allocated to each discrete node on the path. This allows the robot to arrive at the key point strictly according to the preset rhythm to meet business needs. Moreover, the minimum energy reserve required for return is locked in advance during the planning stage, which fundamentally eliminates the risk of task failure due to time delay or energy overdraft caused by planning negligence.

[0101] The benefits of step four are: using multi-dimensional sensor data fusion technology to perform dual logic verification, triggering a replanning request when both infrared thermal features and visual smoke features meet threshold conditions or when there is a substantial blockage in the physical channel, significantly reducing the false alarm frequency caused by noise from a single sensor or changes in ambient light and shadow, ensuring that the robot maintains the continuity of its motion trajectory during task execution, and avoiding frequent interruptions of normal inspection tasks for ineffective calculation adjustments due to system oversensitivity.

[0102] The benefits of step five are: introducing a stable gating decision mechanism based on the principle of hysteresis comparison, calculating the comprehensive improvement of the new path relative to the original path when facing dynamic environmental changes, and allowing path switching when the optimization effect significantly exceeds the preset threshold, effectively suppressing path oscillation and repeated jumps caused by the small displacement of dynamic obstacles in narrow spaces, ensuring the smoothness of robot motion control and significantly reducing unnecessary computing power consumption and machine wear.

[0103] The benefit of step six is ​​that it builds a location-aware real-time energy failure protection mechanism. During the robot's operation, it continuously compares the remaining power with the dynamic budget required to return to the charging station at the current location. Once the power is detected to be below the safety threshold, it forcibly interrupts the current task and takes over control to perform the recharging operation, forming an insurmountable physical safety barrier and preventing systemic failures caused by abnormal energy consumption due to environmental factors, which could lead to the robot running out of power and becoming paralyzed in the parking lot.

[0104] This invention provides a path planning method for a parking lot inspection robot. It has the following beneficial effects:

[0105] 1. This invention employs a global cost function that nonlinearly couples risk and reward with communication quality, combined with the construction of a multidimensional heterogeneous potential field map, to achieve the technical effect of dynamically prioritizing the planning of high-risk areas while taking into account signal coverage. This enables rapid response and value maximization to temperature rise or smoke hazards within a limited time window, solving the shortcomings of existing static inspection routes that lack flexibility and lead to missed detection of high-risk hazards.

[0106] 2. This invention adopts a stable gating decision-making mechanism based on the hysteresis comparison principle. By calculating the cost optimization ratio of the new and old paths, it filters replanning requests with small benefits. This achieves the technical effect of effectively suppressing repeated path calculations and oscillations in dynamic obstacle-dense environments, realizing the smoothness and stability of the robot's operation trajectory, and solving the shortcomings of path jitter and collision risks caused by small environmental disturbances in narrow channels.

[0107] 3. This invention adopts a global state node transmission mechanism that includes energy withdrawal budget and time window constraints, combined with a real-time power circuit breaker watchdog strategy, to achieve the technical effect of avoiding communication blind spots throughout the process and ensuring that the remaining power always meets the bottom line for safe return. It realizes the integrity of inspection data and the closed-loop reliability of tasks, and solves the shortcomings of system paralysis caused by data loss due to communication blind spots and inability to recharge due to power depletion. Attached Figure Description

[0108] Figure 1 is a flowchart of the parking lot inspection robot path planning method provided by the present invention. Detailed Implementation

[0109] To enable those skilled in the art to understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0110] The present invention will now be described in detail with reference to the accompanying drawings:

[0111] Example 1: High-risk inspection and emergency fire response in new energy charging areas

[0112] Please refer to Figure 1, which illustrates a second-level underground parking garage in a large commercial complex according to an embodiment of the present invention. This area contains densely packed columns and unevenly distributed... Signal coverage and dedicated charging areas for new energy vehicles.

[0113] Implementation process:

[0114] Step 1: The system uses LiDAR scanning to construct a 2D grid map, marking pillars and walls as impassable areas. The system traverses the free area and finds weak signal in the corners of the charging zone, while the main channel has a strong signal. Simultaneously, historical data reveals three instances of equipment overheating in the charging zone. Therefore, a nonlinear risk assessment model is used to calculate the risk probability-benefit value for this area. The regular parking area is The system sets opportunity constraints with a confidence level based on operational indicators. The stable gate threshold is .

[0115] Step Two: Constructing the Global Cost Function in the Planning Model. While traversing weak signal corners can shorten path distances, the global cost function includes a risk-communication coupling term, and weak signals can lead to a significant increase in cost. Therefore, the model generates a feasible path covering high-risk charging pile areas and deliberately detouring to the side of the main channel with stronger signals, ensuring sufficient communication bandwidth for real-time transmission of high-definition video during inspections of high-risk locations.

[0116] Step 3: The system generates an initial inspection path and decomposes constraints downwards. For high-risk nodes in the charging area, the system automatically adjusts the task dwell time from the standard based on the risk sensitivity index. seconds extended to Seconds are needed to perform a deep thermal imaging scan. Simultaneously, the energy required for the return trip from the furthest node to the charging station is calculated. The total power is set as an insurmountable energy drawdown budget.

[0117] Step Four: The robot performs its task. Upon reaching the charging area, the infrared thermal imager detects an abnormal temperature rise in a certain parking space. The temperature was set to [degrees Celsius], and simultaneously, a visual camera captured texture features resembling smoke. Since the logical AND operation between the two results in a true value, the multi-source consistency determination formula confirms that a "fire risk" has been triggered, generating a replanning request.

[0118] Step 5: In response to the request, the system generates the shortest reconnaissance path directly to the fire source. Calculations show that the new path, compared to continuing along the original path, achieves a cost optimization ratio of [percentage missing] for this emergency. far exceeding The system establishes a stable gating threshold. It immediately switches paths, controlling the robot to approach the anomaly point, and uses the preferred communication link to transmit alarm information back to the central control room in real time.

[0119] Step Six: After the alarm is completed, the robot's remaining battery power is... Higher than the current position With the energy budget withdrawn, the robot continues to perform the remaining tasks.

[0120] Implementation Effectiveness Verification: This embodiment verifies the effectiveness of the invention in handling conflicts between high-value targets and communication blind spots. Through a nonlinear coupling cost function, the robot is successfully guided to avoid communication blind spots; through multi-source consistent monitoring, real fire situations are accurately identified and high-priority replanning is triggered, achieving zero missed detections and real-time early warning.

[0121] Example 2: Dynamic anti-shake and forced recharge when the battery is low for narrow channels

[0122] Please refer to Figure 1. The embodiment of this invention depicts a residential underground parking garage during late-night hours, where the passageways are narrow and temporary vehicles are entering and exiting. The robot is in a low-battery inspection state.

[0123] Implementation process:

[0124] Step 1: The system loads a multidimensional heterogeneous potential field map. Considering the relatively small dynamic changes in the nighttime environment but the narrow passageways, to prevent the robot from frequently avoiding obstacles and causing "snake-like" movement, the system sets a relatively high stability gating threshold. This means that switching is only allowed when a new path can bring significant benefits.

[0125] Steps two and three: Generate a standard "bow"-shaped inspection path covering the entire site. The system performs reverse energy budget decomposition to calculate the minimum energy consumption required to return to the charging station from the current furthest point. The battery capacity, plus For safety redundancy, the energy drawdown budget is dynamically set to... .

[0126] Step 4: During the robot's movement, in front... A private car is reversing into a parking space, temporarily blocking the passage. The lidar detects the dynamic obstacle and calculates the probability of its removal. Due to the slow movement of the vehicle, the obstruction determination formula is triggered, and the system generates a replanning request.

[0127] Step 5: The system attempts to generate a candidate path to bypass the vehicle. Calculations show that the bypass path requires reversing and traversing a speed bump, resulting in a higher overall cost than waiting for the original path to be restored. . This The optimization ratio and the preset The system compares the request against a stable gating threshold; if the inequality does not hold, the system determines the replanning request as "invalid jitter," refuses to switch paths, and controls the robot to slow down and wait in place. Seconds later, the vehicle is in the parking lot, the road is clear, and the robot continues to travel along the original path, avoiding unnecessary detours.

[0128] Step Six: The robot continues its inspection, but the slippery ground increases motor power consumption. When the robot reaches the end of the garage, the battery management system reading shows that the remaining power has decreased. At this point, the system's real-time comparison revealed that the remaining power had reached the energy withdrawal budget for that location. The energy crisis indicator was immediately set to [value missing]. The system forcibly interrupts the currently unfinished inspection task, ignores all subsequent target points, plans the shortest safe path from the current location to the nearest available charging station, and controls the robot to successfully dock and charge via this path.

[0129] Implementation Results Verification: This embodiment verifies the robustness and system security of the invention in dynamic environments. The stable gating mechanism successfully filters out minor replanning gains caused by temporary blockages, preventing the robot from oscillating in narrow passages; the energy withdrawal budget mechanism accurately triggers forced recharging at the critical point of battery depletion, ensuring that the robot does not become paralyzed deep in the garage due to battery exhaustion, verifying the closed-loop reliability of the system.

[0130] Embodiments of the present invention have been presented and described. It will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to the 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. A path planning method for a parking lot inspection robot, characterized in that, include: Step 1: Acquire parking lot environmental information and construct a multidimensional heterogeneous potential field map. This map includes a physical obstacle layer, a communication field strength distribution layer, and a risk probability distribution layer. Simultaneously, based on preset operational indicators, global control parameters for path planning are deduced and set. These global control parameters include at least chance constraint confidence and a stability gating threshold. Step 2: Construct a path planning model based on the multidimensional heterogeneous potential field map. This model selects a set of feasible paths that meet the probability safety threshold based on the chance constraint confidence. Within this set of feasible paths, a global cost function with non-linear coupling characteristics is constructed. This global cost function establishes a coupling relationship between the revenue value of the risk probability distribution layer and the availability coefficient of the communication field strength distribution layer. Step 3: Based on the global cost function, search and generate an initial inspection path from the set of feasible paths. Then, decompose the global constraints and backpropagate the path. The process is as follows: Step 4: The robot is controlled to perform inspection tasks along the initial inspection path, and multi-source consistency monitoring is performed using multi-dimensional sensors. When a sudden environmental change is detected that meets the multi-source logic triggering condition, a replanning request is generated. Step 5: In response to the replanning request, candidate replanning paths are generated and stable gating decisions are performed. The cost optimization ratio between the candidate replanning path and the remaining segment of the current path is calculated. When the cost optimization ratio exceeds the stable gating threshold, the current path is switched to the candidate replanning path. Step 6: During the path execution, the remaining battery power of the robot is monitored in real time. The remaining battery power is compared with the energy withdrawal budget corresponding to the current position. When the remaining battery power reaches the energy withdrawal budget, the current task is forcibly interrupted and a safe path back to the charging pile is planned.

2. The path planning method for a parking lot inspection robot according to claim 1, characterized in that, Step one further includes: sub-steps Using LiDAR and odometer data, the parking lot environment is scanned to construct a two-dimensional gridded physical obstacle layer map. The parking lot space is divided into sets of grid cells of the same size. For each grid cell in the set, the occupancy probability value of that grid cell is calculated. Furthermore, the physical state of the grid cell is determined using a binarization decision formula. The binarization determination formula is as follows: ,in, The coordinates of the grid cells, This represents the occupancy probability at that coordinate. The preset obstacle detection threshold, For indicator functions; when When the grid is in a certain state, it indicates that the grid is a physical obstacle that cannot be passed; when... When this occurs, it indicates that the grid is a free region; sub-step Based on the two-dimensional rasterized physical obstacle layer map, each raster cell marked as a free region is traversed, and the communication signal strength and data packet loss rate at that location are collected. The communication field strength availability coefficient of that raster cell is calculated using a linear weighted normalization algorithm. A communication field strength distribution layer is generated, and the formula for the linear weighted normalization algorithm is as follows: ,in, This is the received signal strength indication value at this coordinate. and The preset lower and upper limits of signal strength. This represents the packet loss rate at that coordinate. and Signal strength weighting coefficient and packet loss rate weighting coefficient; sub-step Based on grid cell coordinates, the frequency of historical fault records at that coordinate is correlated with real-time thermal imaging temperature data, utilizing... The activation function is used to construct a nonlinear risk assessment model to calculate the risk probability and return value of the grid cell. A risk probability distribution layer is generated, and the calculation formula of the nonlinear risk assessment model is as follows: ,in, This is the normalized real-time temperature rise data at this coordinate. This represents the frequency of historical occupancy disputes or equipment malfunctions at this coordinate location. The weighting is based on the impact of temperature rise. Weighting for the impact of historical faults. For model bias terms, The base of the natural logarithm; when calculating the risk probability return value of this grid cell. The closer This indicates that the higher the inspection value of that location; sub-step Based on the communication field strength distribution layer and the risk probability distribution layer, the global control parameters for path planning are inferred from the preset parking lot operation indicators, and the confidence level is constrained by the missed detection rate constraint formula. Furthermore, the stable gating threshold is calculated using the switching frequency constraint formula. The formula for the missed detection rate constraint is: ≤ ,in, The maximum allowable missed detection rate for the parking lot. The average number of high-risk nodes covered in a single inspection task; the switching frequency constraint formula is: ,in, This represents the maximum number of path replanning attempts allowed per unit of time. This represents the average number of reprogramming iterations measured so far. To minimize the change in effective cost, This represents the historical average total path cost.

3. The path planning method for a parking lot inspection robot according to claim 2, characterized in that, Step two further includes: sub-steps Based on the probability constraint confidence level And the aforementioned two-dimensional rasterized physical obstacle layer map, introducing a dynamic obstacle prediction model, for any candidate path within the planning space The system identifies dynamic obstacles within the area traversed by the path, establishes the probability density function of the future positions of the dynamic obstacles using a Gaussian distribution function, and calculates candidate paths. Cumulative collision risk probability within the future time window ;Will Meet the conditions ≤ Path retention, removal ＞ High-risk paths are selected, and a set of feasible paths that meet the probability safety threshold is identified. Sub-step In the set of feasible paths Within, define adjacent path nodes. and Edge cost function The edge cost function Distance and time cost Energy consumption cost item and risk communication coupling cost term The edge cost function is constructed using a weighted summation method. The calculation formula is: ,in, 、 and The weights are distance, energy consumption, and coupling; sub-steps. For the risk communication coupling cost term in the edge cost function Risk probability return value based on the risk probability distribution layer and the communication field strength availability coefficient of the communication field strength distribution layer A nonlinear inverse proportional coupling model is established, and the calculation formula for the nonlinear inverse proportional coupling model is as follows: ,in, For nodes to The average risk-reward value on the path segment. This is the average communication field strength availability coefficient on this path segment. For communication sensitivity index, To prevent extremely small positive numbers with a denominator of zero; when When it approaches zero, the risk communication coupling cost term The absolute value of approaches infinitesimal, causing the total cost of the path segment to increase dramatically, thereby suppressing the generation of high-risk but communication-free areas in the global cost function.

4. The path planning method for a parking lot inspection robot according to claim 3, characterized in that, Step three further includes: sub-steps Based on the set of feasible paths and edge cost function A heuristic search algorithm is used to traverse and solve the set of feasible paths to obtain the node sequence that minimizes the cumulative total generation value as the initial inspection path. The formula for calculating the cumulative total cost is as follows: ,in, This represents the total number of path nodes. and For adjacent nodes on the path, The edge cost function value between adjacent path nodes; sub-step Based on the initial inspection path Perform forward time extrapolation to generate time window constraints, and use time recursion formulas to calculate the estimated arrival time of each node on the path. And a time tolerance is set based on the expected arrival time. Define the time window constraint for this node. The time recursion formula and the time window constraint are defined as follows: ， ,in, This represents the arrival time of the previous node. This represents the Euclidean distance between adjacent nodes. To improve the robot's planned inspection speed, This refers to the time the robot spends performing its task at the previous node. For a closed interval; sub-step Based on the initial inspection path The energy budget is calculated by using the charging pile coordinates recorded in the physical obstacle layer map, performing reverse energy budget decomposition to generate an energy withdrawal budget, and then using a safe withdrawal model to calculate the minimum retained power required for each node to return to the charging pile, which is defined as the energy withdrawal budget. The calculation formula for the safety pullback model is as follows: ,in, For the current node To the Taiwan charging pile The shortest path distance, This represents the average energy consumption rate of the robot per unit distance. Energy safety redundancy factor; sub-step Constrain the time window With energy drawdown budget Mapped to the initial inspection path The corresponding nodes are combined to generate a sequence of constrained state nodes. Each state node in the sequence The data structure is expressed as: ,in, These are the spatial coordinates of the node.

5. The path planning method for a parking lot inspection robot according to claim 4, characterized in that, Step four further includes: sub-steps Analyze the constrained state node sequence The robot is controlled to move sequentially to each state node in the sequence using a trajectory tracking algorithm. spatial coordinates Furthermore, infrared thermal imaging data is collected simultaneously during the driving process. Visible light visual images and lidar point cloud data Construct a real-time multi-dimensional sensing set for the current moment. The expression for the real-time multidimensional sensing set is: ,in, Current time; sub-step Based on the real-time multidimensional sensing set Perform multi-source consistent real-time monitoring and extract regional temperature rise characteristic values ​​from infrared thermal imaging data. and confidence level of smoke texture in visible light visual images A multi-source consistency determination formula is constructed using logical AND operations to calculate the triggering identifier of high-risk events. The multi-source consistency determination formula is as follows: ,in, The preset abnormal temperature rise threshold, The preset confidence threshold for smoke recognition is set when both abnormal temperature rise and smoke characteristics simultaneously meet the threshold condition. The value is This indicates that it has been confirmed as a genuine high-risk event; otherwise, it is... This is used to filter out false alarms caused by noise from a single sensor; sub-step Based on the lidar point cloud data in the real-time multi-dimensional sensing set Detect dynamic obstacles located ahead of the current driving path. Combined with the time window constraint Calculate the probability of removing the dynamic obstacle within the remaining available time window. Furthermore, the channel blocking trigger flag is determined using the blocking determination formula. The blocking determination formula is as follows: ,in, For the current target node The upper limit of the time window, The remaining available passage time, The parameters are Poisson distribution parameters fitted based on the historical movement speed of the obstacle. This is a preset minimum passage probability threshold; when the calculated removal probability is lower than this threshold, Value This indicates that the obstacle cannot be removed within the specified time window, and the path is substantially blocked; sub-step Real-time monitoring of the high-risk event trigger identifiers With the channel blocking trigger identifier Generate replanning requests using replanning trigger logic. The replanning triggering logic is as follows: ,in, For logical OR operation; when replanning is requested. If true, immediately pause the current trajectory tracking task and send the robot's current state and trigger type to the decision layer.

6. The path planning method for a parking lot inspection robot according to claim 5, characterized in that, Step five further includes: sub-steps Based on the robot's current state included in the replanning request, determine the robot's current position coordinates. and the set of remaining target nodes in the current path that have not yet been visited. Based on the global cost function, for Perform a heuristic path search to generate a connection. Candidate replanning paths with all remaining target nodes The cumulative cost of the candidate replanning path is calculated and defined as the cost of the new path. The calculation formula is: , in, Represents the edges on the path. The cost function value for the edge; sub-step Extract the currently executing path from the path containing the... Unexecuted segments of the old path leading to the destination are defined as the remaining segments of the old path. The cumulative cost of the road segment is calculated using the global cost function and defined as the remaining cost of the old path. The calculation formula is: ,in, The edges are on the remaining segment of the old path. To account for the additional blocking penalty that may occur due to adhering to the original path, when the trigger type is channel blocking. Take the preset positive value, otherwise take the default positive value. Sub-step Obtain the stable gating threshold. Utilizing the cost of the new path With the remaining cost of the old path Construct a hysteresis comparison inequality and calculate the cost optimization ratio of path switching. The logic for determining the hysteresis comparison inequality is as follows: ≥ ,in, The new path represents the increase in revenue compared to the old path; when this inequality holds, it is determined that the replanning path has a significant advantage, and a path switching instruction is generated. If the inequality does not hold, generate a command to maintain the original path. To suppress path jitter caused by small gains; sub-step In response to the path switching command Using the candidate replanning path Replace the current path, and based on The topology is recursively recalculated to determine the estimated arrival time and energy consumption of each node, updating the time window constraints and energy withdrawal budget in the constrained state node sequence, and generating an updated state node sequence. For robots to perform.

7. The path planning method for a parking lot inspection robot according to claim 6, characterized in that, Step six further includes: sub-steps The battery management system on the robot chassis reads the current terminal voltage and discharge current of the battery in real time, and uses a Kalman filter algorithm combining the ampere-hour integral method and the open-circuit voltage method to estimate the remaining power at the current moment. At the same time, obtain the robot's current positioning coordinates. In the constrained state node sequence Perform a nearest neighbor search to match the current best associated node. The formulas for estimating the remaining power and matching the optimal associated node are as follows: ， ,in, This is the initial fully charged capacity. For battery discharge efficiency, for Instantaneous current at a given moment The coordinates of each node in the sequence; sub-step The optimal associated node Extract the corresponding energy drawdown budget from the data structure. Define the energy safety margin factor Construct an energy meltdown determination inequality to calculate energy crisis indicators in real time. The logic of the energy melting determination inequality is as follows: ,in, This is a reserve percentage to withstand sudden high energy consumption on the return trip; when the inequality holds, Set as This indicates that the current battery level has reached the physical limit of the charging station, and the system enters an irreversible forced recharging state; sub-step , in response to determination In this state, immediately trigger the highest priority task interruption command, freeze the current inspection task progress, and traverse all available charging pile sets in the physical obstacle layer map. Calculate the actual travel cost from the current location to each charging station, and select the target charging station with the lowest cost. And plan a safe recharge path The selection formula for the target charging station is as follows: ,in, From current position to number The distance between charging stations This represents the probability of the charging station being currently occupied. and The penalty weights are distance and occupancy status; the robot follows the safe recharge path. Drive until docking is complete.

8. The path planning method for a parking lot inspection robot according to claim 3, characterized in that, The sub-step In this context, the Gaussian distribution function is constructed based on the center position and moving velocity vector of the dynamic obstacle, and is used to describe the dynamic obstacle at future moments. Appearing in location probability density The calculation formula is: ,in, For a moment The predicted center coordinates of a dynamic obstacle are obtained by linear extrapolation based on the obstacle's current velocity. Calculated; and Let be the standard deviation of the positional uncertainty along the coordinate axis, and given the prediction time. The increase is monotonically increasing; The positional correlation coefficient; the cumulative collision risk probability For path All nodes corresponding to The sum of the integrals of the values.

9. A path planning method for a parking lot inspection robot according to claim 4, characterized in that, The sub-step In the context of task execution dwell time It is not a fixed constant, but a variable dynamically calculated based on the data in the risk probability distribution layer. The calculation formula is as follows: ,in, Minimum dwell time for basic inspections The maximum permitted time for key inspections, For nodes Risk probability and return value at the location, This is a risk sensitivity index; the formula stipulates that for high-risk areas, the robot automatically extends its stay time to conduct in-depth scanning, while for low-risk areas, it maintains the basic passage time to optimize the efficiency of time window allocation.

10. A path planning method for a parking lot inspection robot according to claim 4, characterized in that, The sub-step In this context, the heuristic search algorithm employs an improved method. Algorithm, the heuristic function used in the algorithm Introducing communication availability constraints to guide the search direction and avoid communication blind spots, the heuristic function The calculation formula is: ,in, For the current node To the target node Euclidean distance, This represents the communication field strength availability coefficient obtained by the current node from the communication field strength distribution layer. For communication guidance weighting coefficients, To prevent extremely small positive numbers with a denominator of zero; when the search path enters a communication dead zone, Reduced The increase in size allows the algorithm to prioritize expanding nodes with good communication quality.

Images