An autonomous navigation scheduling method and system for a parking lot mobile charging robot

Abstract

This invention discloses an autonomous navigation and scheduling method and system for mobile charging robots in parking lots, relating to the field of mobile robot navigation and scheduling technology. The method includes map construction, dynamic cost maintenance, predictive task allocation, dual-cost joint optimization, time-window sequential scheduling, segmented precise navigation, and two-level progressive rescheduling. This invention effectively addresses the common technical pain points of existing parking lot mobile charging systems—namely, the disconnect between scheduling and navigation, a single task allocation dimension, and frequent timing conflicts in multi-robot paths—by constructing a grid-based topology dual-layer map and an exponentially decaying dynamic cost map system, combined with predictive task allocation, spatial and energy dual-cost optimization, and a time-window sequential scheduling mechanism.

Description

Technical Field

[0001] This invention relates to the field of mobile robot navigation and scheduling technology, specifically to an autonomous navigation and scheduling method and system for a parking lot mobile charging robot. Background Technology

[0002] The continuous growth in the number of new energy vehicles has led to an increasingly prominent contradiction between the supply and demand of charging services in urban parking lots. Existing parking lot charging infrastructure primarily consists of fixed charging piles, which have three inherent limitations: First, charging piles are tied to fixed parking spaces, and the charging service area is rigidly constrained by the installation location, making it difficult to match the utilization rate of charging piles with the charging needs of car owners in a spatial dimension. Second, large-scale deployment of fixed charging piles relies on complex underground cable and parking space renovation projects, resulting in high construction costs and low flexibility, which is particularly limiting in existing parking lots. Third, when charging demand is distributed in peak and off-peak periods, the number of fixed charging piles cannot be dynamically expanded or contracted as needed, leading to a situation where queuing and resource waste coexist during peak electricity consumption periods.

[0003] To address the aforementioned issues, mobile charging robots, as a type of equipment capable of autonomously moving to target vehicles and providing charging services, theoretically possess the potential to break free from spatial constraints. However, existing mobile charging robot solutions remain at a relatively low level of technological maturity: on the one hand, most solutions rely on manual guidance or simple track following for movement, lacking truly autonomous navigation capabilities for the complex spatial structures of parking lots; on the other hand, existing solutions are generally designed based on the premise of independent operation of a single robot, failing to establish a system-level scheduling mechanism for collaborative operation of multiple robots. In scenarios with multiple concurrent charging requests, they cannot effectively allocate tasks or avoid path conflicts, fundamentally limiting service efficiency.

[0004] The parking lot environment itself has unique characteristics that distinguish it from open roads and regularized warehousing environments, further exacerbating the aforementioned technical challenges. In parking lots (especially underground parking lots), GPS signals are blocked by the building structure, preventing robots from relying on satellite positioning to determine their location. Lane width is limited by the layout of parking spaces, and robots must navigate unstructured dynamic obstacles such as temporarily parked vehicles and pedestrians. Multi-level parking lots also involve cross-level connections of ramps or elevators, resulting in a hierarchical path topology. These characteristics collectively determine that the autonomous navigation technology requirements for mobile charging scenarios in parking lots cannot simply be adapted from existing solutions for warehousing logistics or open roads.

[0005] At the scheduling level, when multiple mobile charging robots serve the same parking lot, there is a dynamic coupling relationship between the optimality of task allocation and the real-time status of navigation execution. Channel congestion, changes in the robot's own power level, and new charging requests may all cause the existing scheduling scheme to fail. In the existing technology, the scheduling layer and the navigation layer operate independently and lack a closed-loop linkage mechanism, resulting in insufficient robustness and global optimization capability of the overall system in dynamic scenarios.

[0006] In summary, existing technologies have significant shortcomings in the following aspects: they lack autonomous navigation methods applicable to parking lot GPS rejection, dynamic obstacles, and multi-layer topology environments; they lack system-level collaborative scheduling mechanisms for multiple mobile charging robots; and they lack effective information coupling and linkage optimization methods between the navigation and scheduling functional layers. Summary of the Invention

[0007] The purpose of this invention is to provide an autonomous navigation and scheduling method and system for a mobile charging robot in a parking lot, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: an autonomous navigation and scheduling method and system for a mobile charging robot in a parking lot, comprising the following steps: S1. Map Building: Construct a two-layer raster topology map structure. The bottom raster map is used for local obstacle avoidance, and the upper topology navigation map is used for global path planning and scheduling decisions. S2. Dynamic Cost Maintenance: Each robot periodically reports the measured passage cost of the road segment and its own status. The dispatch center uses an exponential decay model to maintain the dynamic cost weight of each edge of the topology graph and generates a global dynamic cost map. S3. Predictive task allocation: After receiving a charging request, the candidate set is traversed. The set includes idle robots and charging robots that meet the pre-allocation threshold. The threshold is determined based on the proportion of remaining charging time. The expected start time of each candidate is calculated based on the predicted release time to initially screen candidate robots. S4. Dual-cost joint optimization: Using the spatial energy dual-cost functional as the scoring criterion, the optimal robot task allocation is selected from the candidate set; the functional integrates the travel time and total energy consumption dimensions, and the weights are dynamically adjusted according to the system's operating status, with energy safety constraints applied during the allocation process; S5. Time Window Sequential Scheduling: After task allocation, the robot declares that the path occupies a time window. The scheduling center treats the time window as a temporal obstacle and performs conflict avoidance in subsequent path planning. S6, Segmented Precise Navigation: Navigation is divided into two stages: global path planning and precise terminal docking. The docking posture is adjusted by combining the orientation of the charging port and the obstruction of neighboring vehicles to complete the charging docking operation. S7, Two-level progressive rescheduling: During on-the-go travel, a two-level judgment mechanism is used to trigger rescheduling. After the ratio of the measured passage cost to the expected passage cost exceeds the threshold, the obstacle persistence attribute is further determined, and a differentiated handling strategy is output. When triggering replanning, the time window is adjusted simultaneously.

[0009] Preferably, in step S1, the topology navigation map uses lane intersections, ramp entrances, and elevator positions as nodes, and lanes connecting adjacent nodes as edges. Each edge records the length, width constraints, and basic passage time of the corresponding road segment. For multi-story parking lots, each floor independently stores a grid map and a topology sub-map, and the floors are connected by cross-floor topology edges. Map switching and pose re-initialization are automatically completed during cross-floor navigation.

[0010] Preferably, in step S2, the method for calculating the measured passage cost of the road segment is as follows: project the dynamic obstacles detected by the laser radar within the preset perception distance in front of the robot onto the grid map, calculate the grid occupancy rate in the length direction of the road segment, and multiply the occupancy rate by the basic passage time of the road segment to obtain the measured passage cost. The exponential decay model satisfies the following: when a measured passage cost is received within a preset update period, the dynamic cost weight of the corresponding topological edge is directly updated with the measured value; when no measured data is updated after the preset update period, the dynamic cost weight is restored to the basic passage cost according to the following formula: ; In the formula: This is the cost of basic traffic flow on the road section; The cost of the most recently reported actual measurement; The most recent reporting time; Preset update cycle; This is the attenuation coefficient.

[0011] Preferably, in step S3, the predicted release time is calculated by the physical model of the charging process: based on the current state of charge of the target vehicle, the requested charging amount and the nominal characteristics of the battery, the charging process is decomposed into a constant current stage and a constant voltage stage, and the duration is calculated separately. The predicted release time is obtained by superimposing the fixed duration of the robotic arm operation, and is corrected in real time as the actual charging parameters deviate. The pre-allocation threshold is the ratio threshold of the remaining charging time to the total charging time. When the proportion of the robot's remaining charging time is less than this ratio threshold, it is included in the candidate set. The calculation rule for the expected start time is as follows: the expected start time of an idle robot is the predicted travel time overlaid on the dynamic cost map at the current time; the expected start time of a charging robot is the larger of the predicted release time and the travel time overlaid on the current time. The task pre-locking involves adding new tasks to the task queue of the corresponding robot. After the robot completes the current task, it automatically executes the next task navigation according to the queue order without having to return to the waiting area.

[0012] Preferably, in step S4, the spatial energy dual-cost functional expression is: ; In the formula: The predicted travel time for candidate robots to reach the target parking space on the dynamic cost map; The predicted energy consumption of the robot throughout the entire task includes the energy consumption of driving to the target parking space, the energy consumption of charging output, and the energy consumption of driving back to the standby node. and As a weighting factor; The weighting factor , The adjustment rule is as follows: when the number of concurrent charging requests in the system exceeds the preset dense threshold, increase... Weights are prioritized to compress response time; when the average energy storage of robots in the system is lower than a preset safety threshold, the weights are increased. Weights are assigned to prioritize reducing system energy consumption; The energy safety constraint is: the current energy storage of the candidate robot minus the predicted energy consumption is greater than the preset safe energy storage lower limit, and the remaining energy storage can support the robot to return from the task completion position to the dedicated charging position.

[0013] Preferably, in step S5, the path occupancy time window declared by the robot includes the expected entry time and expected exit time of each topological edge on the planned path; the scheduling center stores each time window in association with the corresponding topological edge. In subsequent path planning, if the overlap between the candidate path and the stored time window in the corresponding time period exceeds the safe interval, it is determined to be a timing conflict, and the path planning module automatically searches for alternative paths to complete the avoidance. When path replanning is triggered, the time window of the corresponding robot is updated synchronously, and the conflict between the updated time window and the time windows of other robots is checked. The time windows of the affected robots are coordinated in order of task priority until there is no conflict in the entire system.

[0014] Preferably, in step S6, during the global path planning stage, the robot travels to a preset observation stop point in front of the target parking space; during the precise end-point docking stage, the following steps are performed: The infrared camera is activated to scan the rear of the target vehicle. The charging port protective cover area is located through the target detection model. The three-dimensional position and surface normal vector of the charging port are calculated by combining the installation pose calculation of the camera to determine the orientation category of the charging port. Based on the orientation of the charging port and the obstruction of adjacent parking spaces, a local obstacle avoidance path is generated by selecting an unobstructed direction, and the robot is adjusted to a pose that matches the orientation of the charging docking surface and the charging port. The robotic arm uses a combination of vision servo and flexible force control to sequentially open the charging port protective cover and insert the charging gun.

[0015] Preferably, in step S7, the two-level judgment mechanism specifically includes: The first level is the initial judgment of the cost deviation level: calculate the ratio of the measured traffic cost of the road segment to the planned expected traffic cost, compare it with the preset trigger threshold, and if it does not exceed the threshold, it is judged as normal fluctuation; if it exceeds the threshold, it enters the second level of judgment. The second level is obstacle attribute review: It statistically analyzes the percentage of road segments exceeding the threshold over N consecutive reporting periods (time dimension) and extracts the projected occupancy rate of obstacles along the road segment width (spatial dimension). Combining these two dimensions, a three-level handling strategy is output. In the first scenario, when the continuous percentage of the obstacle is less than 30% and the width occupancy rate is less than 30%, it is judged as a momentary obstacle, and a local deceleration and obstacle avoidance command is issued without adjusting the global path and the reserved time window. In the second scenario, if the continuous percentage is higher than 70% and the width occupancy rate is higher than 60%, it is judged as a continuous obstacle, triggering full path replanning and simultaneously correcting the reserved time window; In the third scenario, all other scenarios are judged as suspected persistent obstacles, a short-term observation window is activated, and the road segment is marked as high-risk. The short-term observation window lasts for multiple consecutive reporting cycles. Within the window, a temporary cost weighting is applied to high-risk road sections to guide subsequent detours. If the obstacle disappears within the window, the marker is removed; if it does not disappear by the end of the window, the entire route will be replanned.

[0016] Preferably, the following operational scheduling steps are also included: First, preventative pre-deployment: Divide the day into several equal-length windows, count the historical charging request frequency of each parking area in each time window, and generate a spatiotemporal heat map of charging demand; with a preset lead time before the start of the current time window, pre-deploy idle robots to the standby nodes of the corresponding area according to the demand probability weight; when the actual charging request arrives, prioritize scheduling the pre-deployed robots in the corresponding area, and correct the heat map parameters online when the demand deviates from the prediction. Second, autonomous power replenishment scheduling: continuously monitor the energy storage status of each robot. When the energy storage of a robot is lower than the preset low power threshold, it is removed from the candidate set. After completing the current task, the robot automatically navigates to a dedicated charging position to replenish power. After replenishing power to the preset recovery threshold, it is reinstated into the candidate set.

[0017] An autonomous navigation and scheduling system for mobile charging robots in parking lots includes a scheduling center and multiple mobile charging robots, which interact bidirectionally via a wireless communication link. The dispatch center includes: The dual-layer map maintenance module is used to build and maintain a raster topology dual-layer map structure, and supports switching between multi-layer parking lot maps; The dynamic cost maintenance module is used to receive the measured passage costs of road segments reported by the robot and generate a global dynamic cost map using an exponential decay model. The predictive task allocation module is used to traverse the candidate robot set, filter candidates by predicting the release time based on the charging curve, and pre-lock the task to be performed by the robot during charging. The space energy dual-cost optimization module is used to calculate candidate scores through dual-cost functionals, select the optimal robot to assign tasks, and apply energy safety constraints. The time window management module is used to record the time window occupied by the path declared by the robot, and to coordinate the time window during execution of timing conflict avoidance and rescheduling. The segmented navigation control module is used to control the robot to perform a segmented navigation process of global path planning and precise end-effector docking; The two-level progressive rescheduling module is used to perform two-level obstacle attribute determination, output differentiated handling strategies, and synchronously correct the time window; The operation and scheduling module is used to maintain the spatiotemporal heat map of charging demand and execute robot pre-deployment, as well as robot power management and autonomous power replenishment scheduling; Each of the mobile charging robots is equipped with a mobile chassis, an energy storage unit, a robotic arm system, and a perception and positioning system. The on-board unit runs a multi-sensor fusion positioning, path planning, and charging docking module, which is used to execute the corresponding steps of the autonomous navigation and scheduling method for the mobile charging robot in the parking lot.

[0018] Compared with the prior art, the beneficial effects of the present invention are: 1. In this invention, by constructing a grid topology dual-layer map and an exponentially decaying dynamic cost map system, combined with predictive task allocation, spatial energy dual-cost optimization, and a time window timing scheduling mechanism, the common technical pain points of existing parking lot mobile charging systems—such as the disconnect between scheduling and navigation, a single task allocation dimension, and frequent timing conflicts in multi-robot paths—are effectively addressed. The solution integrates real-time road condition feedback into the entire-link scheduling decision-making process, and uses the charging curve to predict the release time to achieve early pre-locking of robots during charging, avoiding premature idleness of scheduling resources. The dual-cost functional dynamically adjusts weights according to the system state, taking into account both response time and energy safety constraints. The time window mechanism avoids path conflicts from a timing perspective, significantly improving the system throughput and resource utilization of multi-robot concurrent operations, and quantitatively optimizing the overall charging response time and global scheduling efficiency. 2. In this invention, a two-level judgment logic—initial judgment of cost deviation magnitude and secondary judgment of obstacle attributes in both spatiotemporal dimensions—can accurately distinguish between transient and persistent obstacles and output differentiated handling strategies. This significantly reduces invalid replanning and time window chain corrections caused by transient obstacles such as pedestrian crossings and temporary lane occupancy in parking lot scenarios, and significantly reduces the computing load on the scheduling center and the unnecessary detour mileage of robots. Combined with a segmented precise navigation and spatiotemporal heatmap pre-deployment mechanism, it can not only ensure the accuracy and reliability of charging docking in complex parking space obstruction scenarios, but also realize preventive pre-deployment of robots based on historical needs, further compressing response latency during peak hours, and comprehensively improving the robustness and operational service quality of the system in dynamic scenarios. Attached Figure Description

[0019] Figure 1 This invention provides an autonomous navigation and scheduling method for a mobile charging robot in a parking lot, and an overall architecture diagram of the autonomous navigation and scheduling system for the mobile charging robot in a parking lot. Figure 2 This invention provides an autonomous navigation and scheduling method for a mobile charging robot in a parking lot, and a flowchart of the overall process for the autonomous navigation and scheduling method of the mobile charging robot in a parking lot within the system. Figure 3 This invention provides an autonomous navigation scheduling method for a mobile charging robot in a parking lot, and a flowchart of the logic for determining the in-transit two-level progressive rescheduling in the system. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Combined with appendix Figures 1-3 This invention provides an autonomous navigation and scheduling method and system for mobile charging robots in parking lots. Addressing issues such as the fixed availability of charging piles in parking lots, the separation of navigation and scheduling for mobile charging robots, and the lack of obstacle attribute discrimination for in-transit rescheduling, this invention constructs a bidirectional coupled scheduling framework driven by navigation cost feedback. This framework enables predictive optimization of scheduling decisions for future states, real-time feedback correction of scheduling decisions by navigation execution, and a two-level progressive judgment of obstacle persistence for in-transit rescheduling. The system as a whole consists of a scheduling center deployed on an edge computing server in the building where the parking lot is located, multiple mobile charging robots, and a user's mobile client, all working collaboratively.

[0022] The dispatch center employs a 12-core, 24-thread server processor, 64GB of RAM, and 1TB of high-speed storage, running a Linux operating system. It establishes bidirectional communication links with each mobile charging robot via Wi-Fi 6 wireless LAN, using the MQTT over TLS protocol at a frequency of 1Hz. The dispatch center's software architecture includes the following functional modules: dynamic cost map maintenance module, charging curve release time prediction module, space-energy dual-cost functional optimization module, asynchronous task timing chain pre-allocation module, time window reservation module, in-transit two-level progressive rescheduling module, charging demand spatiotemporal heatmap maintenance module, and pre-deployment scheduling module.

[0023] Each mobile charging robot's hardware configuration includes: a four-wheel independent drive wheeled walking mechanism on the mobile chassis, with each drive wheel equipped with a 750W rated power DC brushless motor, a maximum travel speed of 1.5m / s, and the ability to turn on the spot; its dimensions are 1200mm (L) × 700mm (W) × 900mm (H); the energy storage unit uses a 60kWh lithium iron phosphate battery module with a rated voltage of 384V; the charging output interface supports the GB / T20234.3-2015 DC charging standard, with a maximum output power of 60kW; the robotic arm system uses a six-degree-of-freedom articulated robotic arm, with a charging gun holder and a six-dimensional force / torque sensor at the end, with a range of ±200N and a resolution of 0.1. N, the robotic arm has a working radius of 850mm and a repeatability of ±0.5mm; the perception and positioning system includes a 16-line 360° scanning LiDAR, a forward-facing infrared thermal imaging camera, a visible light camera, and a UHF RFID reading module installed on the bottom of the vehicle; the autonomous charging mechanism is an automatic docking charging interface set at the rear of the robot, which is automatically coupled to the charging pile of the dedicated charging position via an electric push rod; the on-board computing unit adopts a high-performance embedded computing platform, running a multi-sensor fusion positioning module, a layered map maintenance module, a segmented path planning and obstacle avoidance module, a charging port identification and docking module, and a status reporting module. Each module communicates with each other based on the robot's operating system environment.

[0024] The user's mobile client is an application running on mainstream mobile operating systems, providing a charging request submission interface. Users enter the target parking space number (formatted as "floor-area code-parking space number", e.g., "B1-A-035") and the desired charging amount (unit: kWh, ranging from 5-80 kWh, in 5 kWh increments). The client then sends the charging request to the dispatch center's interface via the mobile communication network or parking lot Wi-Fi.

[0025] Combined with appendix Figure 1Before entering operational mode, the mobile charging robot needs to construct a two-layer map of the parking lot. Upon its first entry into the parking lot, the operator uses a wireless remote control to traverse all accessible areas of the parking lot at a speed not exceeding 0.5 m / s. The robot's onboard LiDAR collects 3D point cloud data of the environment at a frequency of 10 Hz. After the point cloud data is collected, the onboard computing unit uses the Cartographer2DSLAM algorithm to perform offline batch processing on the point cloud data, generating a global 2D grid map with a resolution of 5 cm. Each grid cell in this map stores the occupancy probability value for that location (0 indicates completely free, 100 indicates completely occupied, and -1 indicates an unknown area), with a grid size of 5 cm × 5 cm.

[0026] After the grid map is constructed, dispatch center operators manually annotate the following features in the map editing interface: center lines of each lane, all intersections of passageways, the start and end points of ramps between floors, and the stopping floor locations of elevators. Once annotated, the system automatically generates a parking lot topology navigation map. ,in For a set of nodes, Let it be a set of edges. Each edge Connect adjacent nodes and And record the following attributes: road segment length Road section width constraints And the basic traffic hours for the road section: ; In the formula: The robot's nominal speed for obstacle-free travel is set at 1.2 m / s.

[0027] For multi-story parking garages, each floor stores its own raster map, and the connections between floors are represented by cross-floor topological edges. Cross-floor topological edges are divided into two categories: ramp edges connect the start and end points of ramps on different floors, with additional ramp gradient and ramp length parameters, and the basic passage time is corrected according to the slope; elevator edges connect the stopping positions of elevators on different floors, with additional single transfer time parameters.

[0028] The robot's localization during operation employs a three-layer fusion scheme: the global localization layer uses adaptive Monte Carlo localization (AMCL) by matching LiDAR point clouds with pre-built maps, and outputs a global pose estimate. ; The local dead reckoning layer uses wheeled odometers to output incremental pose based on a differential motion model; the absolute correction layer uses ground-deployed passive UHF RFID tags to provide absolute pose observation. The three positioning signals are fused using an extended Kalman filter, and the state vector is defined as: ; In the formula: Global coordinates; For heading angle; Linear velocity; ω is the angular velocity.

[0029] The prediction step is driven by dead reckoning results from wheel odometry, while the observation update step uses AMCL global pose and RFID tag pose as observations, fusing and outputting a global pose estimate at a frequency of 10Hz. Under the conditions that the RFID tag deployment spacing is no greater than 15m and the standard deviation of AMCL matching error does not exceed 0.1m, the root mean square error of global positioning is better than 0.15m.

[0030] Combined with appendix Figure 3 The construction and updating of the dynamic cost map is a fundamental step in achieving bidirectional coupling between scheduling and navigation. Each robot publishes a status packet to the scheduling center via the MQTT protocol at a frequency of 1Hz. The status packet includes the robot number, timestamp, global pose, measured passage cost of each road segment, estimated remaining time for the current task, and its own charge state.

[0031] The method for calculating the measured passage cost of a road segment is as follows: Dynamic obstacles detected within a 30m fan-shaped perception area in front of the robot are projected onto a grid map using global coordinates. The number of grid cells occupied along the length of the road segment is counted, and the measured passage cost is calculated using the following formula: ; In the formula: Section The number of grid cells occupied in the direction of travel; Section Total number of grid cells along the length direction; Section length.

[0032] When the grid occupancy rate exceeds 90%, the road segment is considered completely blocked. Taking infinity indicates that the road section is impassable.

[0033] After receiving the measured traffic costs of each road segment, the dispatch center dynamically assigns cost weights to each edge of the topology graph. An exponential decay model with a forgetting factor is used for maintenance: if the difference between the current time and the most recent reporting time does not exceed the decay time threshold. (If 30 seconds is taken), the weights are updated directly with the latest measured values; if the threshold is exceeded, the basic passage cost is decayed according to the following formula: ; In the formula: For road section Basic toll costs; The most recent reporting time; Let be the attenuation coefficient, and take . .

[0034] The model means that when the measured data is first received, the measured values ​​are used as the standard. After a certain period of no update, the weights gradually return to the base values, reflecting the prior assumption that obstacles may have been removed. The scheduling center summarizes the current dynamic cost weights of all topological edges to form a global dynamic cost map, which serves as the real-time traffic condition basis for subsequent scheduling decisions and path planning.

[0035] Combined with appendix Figures 1-3 The core scheduling mechanism of this invention is the asynchronous task timing chain prediction scheduling driven by the charging curve. The lithium-ion battery charging process is controlled by the battery management system and is divided into two stages: constant current (CC) and constant voltage (CV). In the constant current stage, the battery is charged with a constant current, and the terminal voltage gradually rises to the cutoff voltage. In the constant voltage stage, the cutoff voltage is kept constant, and the charging current gradually decreases until the charging is completed when the cutoff current is reached.

[0036] The specific steps for predicting the charging curve release time are as follows: Step one: The robot communicates with the target vehicle's battery management system via the charging gun's CAN bus communication interface to obtain the battery's rated capacity. Current state of charge Battery temperature Maximum allowable charging power At the same time, it obtains the user's requested charging amount from the dispatch center. .

[0037] Step 2, calculate the actual amount of electricity needed to charge: ; Step 3: Decompose the constant current and constant voltage stages and calculate their durations separately: Duration of constant current phase The average voltage is calculated based on the relationship between the required charging capacity and the chargeable capacity during the constant current stage. The average voltage is the average of the starting voltage and the cutting-off voltage during the constant current stage. Duration of constant pressure phase Based on the exponential decay model, the current satisfies the following time-varying condition: ,in The time constant for the constant voltage stage is obtained from a table based on battery type and temperature, and is derived by inverse solving using the cutoff current. .

[0038] Step 4, predict the release time using the following formula: ; In the formula: The fixed operation time for the robotic arm to remove the gun, hang the gun, and close the protective cover is set to 15 seconds.

[0039] During the charging process, the robot continuously collects actual charging parameters. When the current deviates from the predicted value by more than 10% for three consecutive cycles, or the battery temperature rises by more than 15°C, the robot recalculates using the current parameters as the initial value and corrects the predicted value of the release time online.

[0040] The rules for pre-allocating asynchronous task time chains are as follows: The pre-allocation threshold adopts a dual constraint of the remaining charging time ratio and a fixed duration. The baseline ratio threshold is 20% of the remaining charging time to the total charging time, with a maximum limit of 300 seconds. If the calculated value exceeds 300 seconds, it is set to 300 seconds to avoid locking scheduling resources too early. When the scheduling center receives a user's charging request, the candidate robot set includes all idle robots and charging robots whose predicted release time is less than the pre-allocation threshold.

[0041] For each robot in the candidate set, calculate the expected start time: the expected start time for an idle robot is the predicted travel time overlaid on the dynamic cost map at the current time; the expected start time for a charging robot is the larger of the predicted release time and the travel time overlaid on the current time.

[0042] The dispatch center selects the robot with the shortest expected start time to assign tasks. If the selected robot is charging, task pre-locking is performed: the new task is appended to the end of the robot's task queue. After the robot completes its current charging task and performs the gun removal and cover closing operation, it automatically retrieves the task from the queue and starts navigation without returning to the standby area. The task queue follows a first-in, first-out order, with a maximum queue depth of 3 tasks.

[0043] Combined with appendix Figure 1 The joint optimization of the space-energy dual-cost functional is the core allocation criterion of this invention. In the candidate robot scoring comparison, a dual-cost functional is constructed: ; In the formula: Spatial cost, i.e., the predicted travel time for a candidate robot to reach the target parking space on a dynamic cost map; Energy cost, which is the predicted value of the robot's own energy consumption throughout the entire process of performing the task; and The spatial weights and energy weights are adjusted in real time according to the system state.

[0044] The energy cost consists of three stages of energy consumption: ; In the formula: The energy consumption for traveling from the starting position to the target parking space is equal to the average power consumption multiplied by the travel time. The energy consumption for charging output is equal to the amount of charging requested by the user divided by the overall charging efficiency. This refers to the energy consumption required to return to the nearest standby node after completing a task.

[0045] Real-time adjustment rule for weighting factors: When the number of concurrent charging requests in the system exceeds the dense threshold, increase the weighting factor. Weights are prioritized to compress response time; when the average energy storage of robots in the system is below a safety threshold, the weights are increased. Prioritize reducing system energy consumption.

[0046] Candidate robots must simultaneously satisfy two energy constraints: The current energy storage minus the predicted energy consumption is greater than the preset safe energy storage lower limit; The remaining energy storage can support the robot to return from the task completion location to the nearest dedicated charging point.

[0047] Robots that do not meet the constraints are removed from the candidate set; when the candidate set is empty, a request is made to enter the waiting queue and the waiting time is reported to the user.

[0048] After task allocation, the time window reservation phase begins. The assigned robot plans a global path, calculates the estimated entry and exit times for each topological edge on the path, forms an occupied time window sequence, and reports it to the scheduling center. The scheduling center maintains an occupied time window list for each topological edge, sorted in ascending order of time. In subsequent path planning, if the overlap between a candidate path and an existing time window exceeds the minimum safety interval (5 seconds), it is considered a timing conflict, the corresponding edge is temporarily disabled during that period, and the path planning module searches for an alternative path.

[0049] Combined with appendix Figure 3 The core innovative mechanism of this invention is the two-level progressive rescheduling during transit. Existing mobile robot scheduling systems generally adopt a univariate judgment logic of "triggering replanning when the measured cost exceeds a threshold." However, in parking lot scenarios, more than 80% of dynamic obstacles are transient obstacles such as pedestrians crossing and vehicles temporarily picking up or dropping off passengers. A single trigger logic will frequently lead to invalid replanning, time window chain corrections, and waste of system computing resources. To address the above shortcomings, this invention constructs a two-level progressive judgment process of "initial judgment of magnitude → continuous judgment → differentiated handling." It distinguishes obstacle attributes through time and space dual-dimensional judgment, and only triggers full-path replanning for confirmed continuous obstacles.

[0050] In this embodiment, the rescheduling trigger threshold Historical data statistical period number The short-term observation window length is three consecutive reporting cycles.

[0051] During mission execution, the robot reports the measured travel cost for each road segment it reaches at the midpoint of each topological edge. The scheduling center calculates the ratio of the measured cost to the planned expected cost. The process involves two levels of judgment: the first level is an initial judgment based on the magnitude of the cost deviation; if... If it is determined to be a normal traffic fluctuation, no rescheduling action will be triggered; if Then proceed to the second level of judgment.

[0052] The second level is obstacle attribute review, which is judged in parallel from both temporal and spatial dimensions: Time dimension: statistical continuity The percentage of times the measured cost exceeded the threshold during each reporting period ; Spatial dimension: Extracting the projection occupancy of obstacle point cloud in the road segment width direction. .

[0053] Based on the results from both dimensions, a three-tiered response strategy is output: The first scenario, and The obstacle was determined to be transient. The dispatch center did not trigger a global replanning or modify the reserved time window, but only issued a local deceleration and obstacle avoidance command. After the robot decelerated and passed the obstacle, it automatically resumed its nominal speed, and the extra time was absorbed by the subsequent driving phase.

[0054] The second scenario, and The obstacle was identified as a persistent obstacle. The dispatch center triggered a full path replanning, researching for a detour path starting from the robot's current position, simultaneously correcting the reserved time window, and performing a system-wide time window conflict check. The time windows of the affected robots were coordinated according to task priority until there were no conflicts.

[0055] In the third scenario, the remaining conditions are judged as suspected persistent obstacles. The dispatch center initiates a short-term observation window, marks the road segment as a high-risk segment, and applies a temporary cost weighting of 2 times to guide subsequent routes to prioritize detours; if the cost recovers to below the threshold for two consecutive cycles within the window, the marking is lifted; if the obstacle has not dissipated by the end of the window, it is transferred to persistent obstacle handling.

[0056] The quantifiable technical effects of this two-level progressive judgment mechanism include: a reduction of approximately 65% ​​in CPU load at the scheduling center, an average reduction of approximately 80% in the number of time window chain corrections per day, and an average reduction of approximately 120m per day in the unnecessary detour mileage of robots.

[0057] Combined with appendix Figure 1 The segmented adaptive precision navigation with charging port orientation constraints breaks down the navigation process into a global path planning stage and a precise end-point docking stage.

[0058] In the global path planning phase, the robot uses the center point of the target parking space as its endpoint and employs the A* algorithm on the topology map to search for the path with the minimum travel cost. The path cost includes the dynamic costs of each edge and the turning penalty. The robot travels along the global path to the observation stop point 3 meters in front of the target parking space, at which point the global phase ends.

[0059] After entering the precise docking phase, the robot activates its infrared thermal imaging camera to scan and image the rear of the target vehicle. The onboard computing unit runs a lightweight target detection network to locate the charging port protective cover area in the infrared image, and takes the result with the highest confidence level for 5 consecutive frames as the final location. Combining the camera's intrinsic and extrinsic parameters, the EPnP algorithm is used to calculate the 3D position and surface normal vector of the charging port protective cover in the robot's coordinate system, and the orientation of the charging port is determined based on the normal vector components: right side of the rear, left side of the rear, or middle.

[0060] The robot selects its approach direction based on the orientation of the charging port and the obstruction of adjacent parking spaces: if the charging port is on the same side as the robot's current position and there are no obstructions from adjacent vehicles, the robot moves forward in a straight line to the robotic arm's operating range; if the orientation is on the opposite side or on the same side and there are obstructions, the robot plans a local detour path in the form of a Bézier curve, adjusting from the unobstructed side to the matching docking posture. If there are obstructions on both sides and the rear center is unreachable, the robot reports a "target parking space unreachable" status, and the dispatch center sends a notification to the user.

[0061] After reaching the docking position, the robotic arm uses a combination of visual servo and flexible force control to sequentially perform the operations of opening the charging port protective cover, retrieving the gun, and inserting the gun; after charging is completed, it performs the actions of removing the gun, hanging the gun, and closing the protective cover in reverse order.

[0062] Combined with appendix Figure 1 The establishment and maintenance process of the preventive pre-deployment mechanism based on the spatiotemporal heat map of charging demand is as follows: The dispatch center divides a 24-hour day into 24 time windows with 1-hour units, and divides the parking lot into several parking areas with 10 adjacent parking spaces as units. It maintains the historical charging request frequency matrix of each area in each time window and updates the data using a 30-day sliding window.

[0063] Calculate the demand probability of each region in each time window to generate a spatiotemporal heatmap of charging demand: ; In the formula: For the region In the time window The historical number of charging requests; For the region In the time window The probability of demand.

[0064] The pre-deployment operation is triggered once 15 minutes before the start of each time window. The pre-deployment steps are as follows: Step 1: Determine the number of robots to be deployed, taking the smaller value between the number of idle robots and the preset upper limit. More than 30% of idle robots were included in the pre-deployment candidate; Step 2: Sort the robots by probability of demand from high to low, and use a roulette wheel algorithm to assign target areas to each robot. Areas with a probability of less than 5% will not participate in this pre-deployment. Step 3: Send navigation instructions to the corresponding robot. The target is a preset standby node in the area. After the robot arrives, it stops and waits without occupying the user's parking space.

[0065] When a real charging request arrives, pre-deployed robots within the target area are prioritized for dispatch, reducing the response time to the duration of short-distance movement within the area. If there are no requests in the pre-deployed area within a 15-minute deviation judgment window, but requests occur in the non-pre-deployed area, robots are recalled from the original pre-deployed area to respond across areas. At the same time, the demand probability of the corresponding area in the heatmap is corrected online using the exponential moving average method.

[0066] Example 1: Complete charging task operation process.

[0067] This embodiment uses the 9:00-10:00 AM operating period on a weekday morning as an example to illustrate the complete process of the system from pre-deployment to task execution. At 8:45 AM (15 minutes before entering the 9:00-10:00 AM time window), the spatiotemporal heat map of charging demand shows that the demand hotspots during this period are concentrated in area A3 (entrance elevator area, historical demand probability 0.37) and area A12 (supermarket entrance area, historical demand probability 0.24). The currently idle robots are R2 (State of Charge (SOC) = 0.72) and R3 (State of Charge (SOC) = 0.65). The dispatch center performs preventative pre-deployment: R2 is assigned to the standby node in area A3, and R3 is assigned to the standby node in area A12.

[0068] At 9:12 AM, a user initiated a charging request: the target parking space is B1-A3-035, and the requested charging amount is 40kWh. At this time, robot R1 is performing a charging task in area A8. The predicted task release time is 27 minutes from the current time, which exceeds the pre-allocation threshold, so it is not included in the candidate robot set; the candidate set is {R2, R3}.

[0069] Calculate the expected start time for each candidate task: R2 is located in area A3, with a predicted travel time of 38 seconds under the dynamic cost map, and an expected start time of 9:12:38; R3 is located in area A12 and needs to move across areas, with a predicted travel time of 195 seconds, and an expected start time of 9:15:15. R2 is initially selected as the optimal candidate.

[0070] Execution space-energy dual cost verification: R2's current energy storage is 43.2 kWh, and the predicted energy consumption for the entire task is 43.53 kWh. After completing the task, the remaining energy storage is below the safe lower limit of 6 kWh, failing to meet the energy constraint condition. Therefore, R2 is removed from the candidate set. Similarly, R3 also fails to meet the energy constraint, and its candidate set is empty.

[0071] In response to this scenario, the dispatch center implements a three-level response strategy: First, it pushes optional options to the user, including "partial charging (approximately 8kWh, wait 5 minutes)" and "full charging (wait approximately 35 minutes, to be executed after the robot is recharged)," which the user can choose independently; Second, it dynamically increases the recharging priority of dedicated charging stations to speed up the recharging speed of idle robots; Third, it monitors the charging progress of R1 in real time, and automatically adds it to the waiting queue for task allocation after its task is released and energy constraints are met.

[0072] Example 2: In-transit two-level progressive rescheduling operation scenario.

[0073] This embodiment uses the time period of 14:00-15:00 on a Saturday as an example to illustrate the actual operation process of the two-level progressive rescheduling mechanism. Currently, robot R1 is idle and waiting in area A7 with SOC=0.81; at 14:05, the user initiates a charging request for parking space A8 in area A8. R1 is the only candidate robot, the energy constraint verification passes, and the task is officially assigned to R1.

[0074] R1 plans a global path, passing through topological edges 23 and 24 to reach the target parking space, and reports the corresponding path's occupancy time window to the dispatch center. When R1 reaches the middle section of topological edge 23, it encounters a temporarily occupied SUV and construction barriers, causing a slight increase in the measured passage cost of the section. However, the ratio R of the measured cost to the expected cost is lower than the rescheduling trigger threshold of 2.0, which is judged as normal traffic fluctuation. R1 slows down and detours through, without triggering rescheduling.

[0075] When R1 reaches the entrance of topology edge 24, it finds a truck completely blocking the road. The actual road passage cost rises to infinity, and the cost ratio far exceeds the trigger threshold. The first level of cost deviation is initially judged as passing and enters the second level of obstacle attribute re-judgment.

[0076] The dispatch center retrieved historical data from the last five reporting periods: In terms of time, the measured cost exceeded the threshold in three of the five periods, representing a sustained exceedance rate of 60%, which is within the middle range; in terms of space, the obstacle width projection occupancy rate was 81.25%, indicating a large-area obstacle. The results from these two dimensions contradict each other, leading to a combined assessment that it is a suspected persistent obstacle. A short-term observation window of three reporting periods was initiated, marking topological edge number 24 as a high-risk section and applying a temporary cost weighting of 2x.

[0077] Within the three observation periods, the road segment remained completely blocked, and the obstacle persisted. After the window expired, it was officially classified as a persistent obstacle, triggering a full path replanning: starting from R1's current position, it would detour via topological branches 25 and 26 to reach the target parking space. This detour increased the total path length by 65 meters and the estimated travel time by approximately 67 seconds. Simultaneously, the dispatch center corrected R1's reserved time window sequence and performed a system-wide time window conflict check, confirming that there were no other robots experiencing time window conflicts.

[0078] After R1 reaches the target parking space following its detour path, it enters the precise docking phase: the infrared camera identifies the charging port as facing the right rear, with no obstructions from the adjacent parking space on the right. The robot moves forward in a straight line to the robotic arm's operating range, where the robotic arm sequentially opens the charging port protective cover and inserts the charging gun, officially initiating charging. During charging, R1 continuously reports charging curve parameters, and the dispatch center updates the predicted release time of the task in real time.

[0079] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An autonomous navigation and scheduling method for a mobile charging robot in a parking lot, characterized in that: Includes the following steps: S1. Map Building: Construct a two-layer raster topology map structure. The bottom raster map is used for local obstacle avoidance, and the upper topology navigation map is used for global path planning and scheduling decisions. S2. Dynamic Cost Maintenance: Each robot periodically reports the measured passage cost of the road segment and its own status. The dispatch center uses an exponential decay model to maintain the dynamic cost weight of each edge of the topology graph and generates a global dynamic cost map. S3. Predictive task allocation: After receiving a charging request, the candidate set is traversed. The set includes idle robots and charging robots that meet the pre-allocation threshold. The threshold is determined based on the proportion of remaining charging time. The expected start time of each candidate is calculated based on the predicted release time to initially screen candidate robots. S4. Dual-cost joint optimization: Using the spatial energy dual-cost functional as the scoring criterion, the optimal robot task allocation is selected from the candidate set; the functional integrates the travel time and total energy consumption dimensions, and the weights are dynamically adjusted according to the system's operating status, with energy safety constraints applied during the allocation process; S5. Time Window Sequential Scheduling: After task allocation, the robot declares that the path occupies a time window. The scheduling center treats the time window as a temporal obstacle and performs conflict avoidance in subsequent path planning. S6, Segmented Precise Navigation: Navigation is divided into two stages: global path planning and precise terminal docking. The docking posture is adjusted by combining the orientation of the charging port and the obstruction of neighboring vehicles to complete the charging docking operation. S7, Two-level progressive rescheduling: During on-the-go travel, a two-level judgment mechanism is used to trigger rescheduling. After the ratio of the measured passage cost to the expected passage cost exceeds the threshold, the obstacle persistence attribute is further determined, and a differentiated handling strategy is output. When triggering replanning, the time window is adjusted simultaneously.

2. The autonomous navigation and scheduling method for a mobile charging robot in a parking lot according to claim 1, characterized in that: In step S1, the topology navigation map uses lane intersections, ramp entrances, and elevator locations as nodes, and lanes connecting adjacent nodes as edges. Each edge records the length, width constraints, and basic travel time of the corresponding road segment. For multi-story parking lots, each floor independently stores a raster map and a topological submap, and the floors are connected by cross-floor topological edges. When navigating across floors, map switching and pose re-initialization are automatically completed.

3. The autonomous navigation and scheduling method for a mobile charging robot in a parking lot according to claim 1, characterized in that: In step S2, the method for calculating the measured passage cost of the road segment is as follows: project the dynamic obstacles detected by the laser radar within the preset perception distance in front of the robot onto the grid map, calculate the grid occupancy rate in the length direction of the road segment, and multiply the occupancy rate by the basic passage time of the road segment to obtain the measured passage cost. The exponential decay model satisfies the following: when a measured passage cost is received within a preset update period, the dynamic cost weight of the corresponding topological edge is directly updated with the measured value; when no measured data is updated after the preset update period, the dynamic cost weight is restored to the basic passage cost according to the following formula: ； In the formula: This is the cost of basic traffic flow on the road section; The cost of the most recently reported actual measurement; The most recent reporting time; Preset update cycle; This is the attenuation coefficient.

4. The autonomous navigation and scheduling method for a mobile charging robot in a parking lot according to claim 1, characterized in that: In step S3, the predicted release time is calculated by the physical model of the charging process: based on the current state of charge of the target vehicle, the requested charging amount and the nominal characteristics of the battery, the charging process is decomposed into a constant current stage and a constant voltage stage, and the duration is calculated separately. The predicted release time is obtained by superimposing the fixed duration of the robotic arm operation, and is corrected in real time as the actual charging parameters deviate. The pre-allocation threshold is the ratio threshold of the remaining charging time to the total charging time. When the proportion of the robot's remaining charging time is less than this ratio threshold, it is included in the candidate set. The calculation rule for the expected start time is as follows: the expected start time of an idle robot is the predicted travel time overlaid on the dynamic cost map at the current time; the expected start time of a charging robot is the larger of the predicted release time and the travel time overlaid on the current time. The task pre-locking involves adding new tasks to the task queue of the corresponding robot. After the robot completes the current task, it automatically executes the next task navigation according to the queue order without having to return to the waiting area.

5. The autonomous navigation and scheduling method for a mobile charging robot in a parking lot according to claim 1, characterized in that: In step S4, the spatial energy dual-cost functional expression is: ； In the formula: The predicted travel time for candidate robots to reach the target parking space on the dynamic cost map; The predicted energy consumption of the robot throughout the entire task includes the energy consumption of driving to the target parking space, the energy consumption of charging output, and the energy consumption of driving back to the standby node. and As a weighting factor; The weighting factor , The adjustment rule is as follows: when the number of concurrent charging requests in the system exceeds the preset dense threshold, increase... Weights are assigned to prioritize and compress response time. When the average energy storage of robots in the system is lower than the preset safety threshold, increase Weights are assigned to prioritize reducing system energy consumption; The energy safety constraint is: the current energy storage of the candidate robot minus the predicted energy consumption is greater than the preset safe energy storage lower limit, and the remaining energy storage can support the robot to return from the task completion position to the dedicated charging position.

6. The autonomous navigation and scheduling method for a mobile charging robot in a parking lot according to claim 1, characterized in that: In step S5, the time window occupied by the path declared by the robot includes the expected entry time and expected exit time of each topological edge on the planned path; the scheduling center stores each time window and the corresponding topological edge in association. When planning the path in the future, if the overlap between the candidate path and the stored time window in the corresponding time period exceeds the safe interval, it is determined to be a timing conflict, and the path planning module automatically searches for an alternative path to complete the avoidance. When path replanning is triggered, the time window of the corresponding robot is updated synchronously, and the conflict between the updated time window and the time windows of other robots is checked. The time windows of the affected robots are coordinated in order of task priority until there is no conflict in the entire system.

7. The autonomous navigation and scheduling method for a mobile charging robot in a parking lot according to claim 1, characterized in that: In step S6, during the global path planning phase, the robot travels to the preset observation stop point in front of the target parking space; during the precise end-point docking phase, the following steps are executed: The infrared camera is activated to scan the rear of the target vehicle. The charging port protective cover area is located through the target detection model. The three-dimensional position and surface normal vector of the charging port are calculated by combining the installation pose calculation of the camera to determine the orientation category of the charging port. Based on the orientation of the charging port and the obstruction of adjacent parking spaces, a local obstacle avoidance path is generated by selecting an unobstructed direction, and the robot is adjusted to a pose that matches the orientation of the charging docking surface and the charging port. The robotic arm uses a combination of vision servo and flexible force control to sequentially open the charging port protective cover and insert the charging gun.

8. The autonomous navigation and scheduling method for a mobile charging robot in a parking lot according to claim 1, characterized in that: In step S7, the two-level judgment mechanism is specifically as follows: The first level is the initial judgment of the cost deviation level: calculate the ratio of the measured traffic cost of the road segment to the planned expected traffic cost, compare it with the preset trigger threshold, and if it does not exceed the threshold, it is judged as normal fluctuation; if it exceeds the threshold, it enters the second level of judgment. The second level is obstacle attribute review: It statistically analyzes the percentage of road segments exceeding the threshold over N consecutive reporting periods (time dimension) and extracts the projected occupancy rate of obstacles along the road segment width (spatial dimension). Combining these two dimensions, a three-level handling strategy is output. In the first scenario, when the continuous percentage of the obstacle is less than 30% and the width occupancy rate is less than 30%, it is judged as a momentary obstacle, and a local deceleration and obstacle avoidance command is issued without adjusting the global path and the reserved time window. In the second scenario, if the continuous percentage is higher than 70% and the width occupancy rate is higher than 60%, it is judged as a continuous obstacle, triggering full path replanning and simultaneously correcting the reserved time window; In the third scenario, all other scenarios are judged as suspected persistent obstacles, a short-term observation window is activated, and the road segment is marked as high-risk. The short-term observation window lasts for multiple consecutive reporting cycles. Within the window, a temporary cost weighting is applied to high-risk road sections to guide subsequent detours. If the obstacle disappears within the window, the marker is removed; if it does not disappear by the end of the window, the entire route will be replanned.

9. The autonomous navigation and scheduling method for a mobile charging robot in a parking lot according to claim 8, characterized in that: It also includes the following operational scheduling steps: First, preventative pre-deployment: Divide the day into several equal-length windows, count the historical charging request frequency of each parking area in each time window, and generate a spatiotemporal heat map of charging demand; with a preset lead time before the start of the current time window, pre-deploy idle robots to the standby nodes in the corresponding areas according to the demand probability weight. When a real charging request arrives, the pre-deployed robots in the corresponding area are prioritized for scheduling; when the demand deviates from the forecast, the heat map parameters are corrected online. Second, autonomous power replenishment scheduling: continuously monitor the energy storage status of each robot, and remove the robot from the candidate set when its energy storage is lower than the preset low power threshold; after the robot completes the current task, it automatically navigates to the dedicated charging position to replenish power, and is reinstated into the candidate set after replenishing power to the preset recovery threshold.

10. An autonomous navigation and scheduling system for a mobile charging robot in a parking lot, characterized in that: This includes a dispatch center and multiple mobile charging robots, which interact bidirectionally via wireless communication links; The dispatch center includes: The dual-layer map maintenance module is used to build and maintain a raster topology dual-layer map structure, and supports switching between multi-layer parking lot maps; The dynamic cost maintenance module is used to receive the measured passage costs of road segments reported by the robot and generate a global dynamic cost map using an exponential decay model. The predictive task allocation module is used to traverse the candidate robot set, filter candidates by predicting the release time based on the charging curve, and pre-lock the task to be performed by the robot during charging. The space energy dual-cost optimization module is used to calculate candidate scores through dual-cost functionals, select the optimal robot to assign tasks, and apply energy safety constraints. The time window management module is used to record the time window occupied by the path declared by the robot, and to coordinate the time window during execution of timing conflict avoidance and rescheduling. The segmented navigation control module is used to control the robot to perform a segmented navigation process of global path planning and precise end-effector docking; The two-level progressive rescheduling module is used to perform two-level obstacle attribute determination, output differentiated handling strategies, and synchronously correct the time window; The operation and scheduling module is used to maintain the spatiotemporal heat map of charging demand and execute robot pre-deployment, as well as robot power management and autonomous power replenishment scheduling; The mobile charging robot is equipped with a mobile chassis, an energy storage unit, a robotic arm system, and a perception and positioning system. The on-board unit runs a multi-sensor fusion positioning, path planning, and charging docking module, which is used to execute the corresponding steps of the method described in any one of claims 1 to 9.

Images