A multi-robot collaborative construction system for lunar soil in-situ construction

Abstract

The application discloses a kind of multi-robot collaborative construction systems for lunar soil in-situ construction, including control center and multifunctional construction robot cluster;Among them, control center includes communication positioning system, central task scheduling module, in-situ construction control platform and remote interaction and monitoring terminal, multifunctional construction robot cluster includes lunar soil brick manufacturing unit, lunar soil brick transport robot and construction robot.The application guarantees stable communication and high-precision positioning in complex environment through communication positioning system, in-situ construction control platform is responsible for path planning, parameter adjustment and state monitoring, and remote terminal realizes visual management;Robot cluster can be dynamically combined, and algorithm is used to complete lunar soil processing, transportation, laying whole-process operation;In addition, the application breaks through the limitation of lunar environment adaptation of traditional scheme through algorithm and hardware deep cooperation, significantly improves construction efficiency and reliability, and provides core technical support for automated construction of lunar base and other facilities.

Description

Technical Field

[0001] This invention relates to an in-situ lunar soil construction system, specifically a multi-robot collaborative construction system for in-situ lunar soil construction, belonging to the field of space exploration and construction technology. Background Technology

[0002] As humanity's exploration of outer space deepens, the Moon has become a crucial target for extraterrestrial activities. Establishing a base on the lunar surface is indispensable for scientific research, resource development, and long-term manned colonization. Utilizing lunar regolith for in-situ construction is a key step in realizing the construction of a lunar base.

[0003] In-situ lunar soil construction refers to the direct use of lunar surface soil resources to construct necessary facilities and structures locally on the moon using specific technologies and processes. Compared to transporting building materials from Earth, this method significantly reduces costs and enables sustainable development of lunar base construction.

[0004] In the existing technology, there have been some related technological explorations regarding in-situ construction of lunar soil: 1) The method and device for simulating lunar soil laser sintering molding disclosed in Announcement No. CN111958750B adopts a layer-by-layer deposition and laser beam sintering melting molding method for in-situ construction of lunar soil. The layer-by-layer sintering method is inefficient and troublesome to handle, and cannot be effectively used for large-scale construction; 2) The lunar soil concrete preparation vehicle based on material extrusion molding technology and its application disclosed in Announcement No. CN113561289B adopts the extrusion molding method. However, the composition of lunar soil includes silicate and aluminosilicate minerals, and pressure molding alone cannot meet the conditions for use on the moon.

[0005] In summary, existing in-situ lunar soil construction technologies are mostly single-method lunar soil processing methods, lacking clustered operation systems and processes. There is an urgent need for an innovative multi-robot collaborative construction system for in-situ lunar soil construction to solve the above problems and promote the development of lunar base construction. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-robot collaborative construction system for in-situ lunar soil construction in order to solve at least one of the above-mentioned technical problems.

[0007] The present invention achieves the above objectives through the following technical solution: a multi-robot collaborative construction system for in-situ construction of lunar soil, the multi-robot collaborative construction system including a control center and a multi-functional cluster robot, the multi-functional cluster robot being scheduled by the control center;

[0008] The control center includes a communication and positioning system, a central task scheduling module, an in-situ constructed control platform, and a remote interactive and monitoring terminal;

[0009] The multi-functional swarm robot includes a lunar soil brick manufacturing unit, a lunar soil brick transport robot, and a construction robot.

[0010] As a further technical solution of the present invention: the lunar soil brick manufacturing unit collects lunar soil and performs crushing and screening, and manufactures lunar soil bricks with mortise and tenon structures by using vibration screening and hot sintering molding processes.

[0011] As a further technical solution of the present invention: the lunar soil brick manufacturing unit also includes a lunar soil crushing device, a vibrating screen, a lunar soil brick mold, a solar focusing heating device, a conveyor belt, a rotary demolding device and lunar soil bricks. The lunar soil crusher is located above the vibrating screen, the lunar soil brick mold is located below the vibrating screen, the solar focusing heating device is connected to the lunar soil brick mold, and the rotary demolding device is located at the end of the conveyor belt.

[0012] The timing of the lunar soil brick manufacturing unit is controlled by a multi-constraint dynamic task allocation algorithm of the central task scheduling module: when the communication and positioning system detects sunlight intensity ≥600W / m 2 At that time, the algorithm assigns crushing and sintering tasks to the lunar regolith brick manufacturing unit by increasing the manufacturing task association weight in the fitness function; when the sunlight intensity is <400W / m 2 At that time, the algorithm triggers the reward function. The penalty logic stops the sintering task and switches to mold cleaning and equipment self-check tasks.

[0013] As a further technical solution of the present invention: the lunar soil brick includes a tenon brick and a mortise and tenon brick. One end of the tenon brick is provided with a protruding tenon structure, and one end of the mortise and tenon brick is provided with a mortise and tenon groove that matches the tenon. The inside of the mortise and tenon groove is provided with a protrusion that matches the anti-slip texture of the tenon.

[0014] The quantity of lunar soil bricks manufactured and the number of transport batches are dynamically adjusted by the multi-constraint dynamic task allocation algorithm of the central task scheduling module according to the construction task progress: the algorithm calculates the fitness function in real time. and The relationship between the two is used to control the quantity of lunar soil bricks manufactured and the batches of transportation, ensuring that the construction robots have no waiting for materials and that the inventory of lunar soil bricks is ≤ 15% of the total load of the transportation robots.

[0015] As a further technical solution of the present invention: the construction robot includes a mobile platform, a multi-degree-of-freedom robotic arm and a gripping device. The mobile platform is connected to the multi-degree-of-freedom robotic arm, the multi-degree-of-freedom robotic arm is connected to the gripping device, and the gripping device grips the lunar soil bricks in a clamping manner.

[0016] The priority of the construction robot's operations and the start and stop times are determined by the multi-constraint dynamic task allocation algorithm of the central task scheduling module. When the algorithm detects that the execution error of the construction robot is greater than 0.2 cm, it automatically increases the fitness function. The value of is increased, and the task interval of the robot is adjusted to increase the frequency of visual positioning calibration, ensuring that the subsequent execution error is ≤0.2cm.

[0017] As a further technical solution of the present invention: the central task scheduling module is equipped with a multi-constraint dynamic task allocation algorithm, which adopts a hybrid optimization framework composed of an improved genetic algorithm and reinforcement learning, and specifically performs the following operations:

[0018] 1) Receive the overall construction task input from the remote interactive and monitoring terminal, and decompose it into a task set T={T1,T2,...,T...} according to the three-level structure module-subtask-operation unit. n Each subtask carries its type, priority, resource requirements, time and space constraints, and precision requirements.

[0019] 2) Collect environmental status data, robot status data, and task progress data through the communication and positioning system to construct an environment-robot-task status matrix;

[0020] 3) Initial task allocation is achieved based on an improved genetic algorithm;

[0021] 4) Dynamic optimization based on reinforcement learning: Each robot in the multifunctional construction robot cluster is defined as an independently executing intelligent agent, and a central task scheduling module is used as a globally coordinated intelligent agent;

[0022] 5) Built-in conflict detection and resolution mechanism:

[0023] By comparing the temporal-spatial parameters of each robot task, overlapping conflicts were identified, and the conflict types were clarified as: temporal conflicts, spatial conflicts, and mixed conflicts.

[0024] Conflict levels are categorized by the priority of the tasks involved: High-level conflicts: conflicts involving P1 task bots and mixed conflicts; Medium-level conflicts: conflicts involving P2 task bots; Low-level conflicts: conflicts involving P3 task bots.

[0025] The conflict is resolved according to the following rules: "P1 task robots take priority over P2 / P3 task robots, robots with remaining energy <30% take priority over high-energy robots, and robots with historical execution accuracy ≥98% take priority over low-accuracy robots". After resolution, the time-space parameters of each robot task are re-checked. If the conflict occurrence rate is >3%, the above resolution rules are re-executed until the conflict occurrence rate is ≤3%, ensuring that the conflict occurrence rate is ≤3%.

[0026] 6) The top-level iterative hub of the loop optimization layer.

[0027] As a further technical solution of the present invention: in 3), the initial task allocation is implemented based on the improved genetic algorithm, specifically including:

[0028] 31) Chromosome encoding: adopts the format of "robot number, task number, start time, end time, constraint satisfaction", where constraint satisfaction represents the degree to which the spatiotemporal constraints, terrain constraints, and energy constraints of the subtask are satisfied;

[0029] 32) Fitness function, used to evaluate the quality of task allocation schemes:

[0030] ;

[0031] Where F is the fitness value; , , The weights are dynamic, and It is dynamically adjusted according to the lunar environment; The total construction period for the current task allocation plan; The theoretical maximum allowable period for this construction task is predetermined by the requirements of the construction task; The average energy efficiency of all robots; Energy utilization rate when the robot is running at full load; The average execution error of all subtasks; This represents the maximum permissible execution error for the subtask.

[0032] 33) Genetic operations: After selection, crossover, mutation, and iteration, the initial task allocation scheme P0 is output;

[0033] As a further technical solution of the present invention: in 4), dynamic optimization is achieved based on reinforcement learning: each robot in the multifunctional construction robot cluster is defined as an independently executing intelligent agent, and a central task scheduling module is used as a globally coordinated intelligent agent, specifically including:

[0034] 41) State space construction: Define the system state s at time t. t =[Sunlight intensity S, Robot remaining energy E, Real-time execution accuracy δ, Task completion rate C, Fault status F];

[0035] Among them, the solar intensity S is quantified in intervals, with S1≥600W / m 2 400W / m 2 ≤S2<600W / m 2 S3 < 400W / m 2The robot's remaining energy E is quantified proportionally, with E1≥60%, 30%≤E2<60%, and E3<30%; the real-time execution accuracy δ is quantified according to the error threshold, with δ1≤0.2cm and δ2>0.2cm; the task completion rate C is quantified as a percentage, with 0≤C≤100%; the fault status F is quantified in binary, with F=0 indicating no fault and F=1 indicating a fault.

[0036] 42) Action Space Definition: The actions of the agent include: task acceptance / rejection; task type switching; task transfer request; job parameter adjustment; time window request;

[0037] 43) Action selection strategy: The ε-greedy strategy is adopted, with a 90% probability of selecting the optimal action with the largest current Q value and a 10% probability of randomly selecting an action;

[0038] 44) Reward function: Used to evaluate the rationality of the agent's actions, as shown in the following formula:

[0039] ;

[0040] in: Let be the reward value at time t; The reward value for completing the task on time; The penalty value for energy waste; The penalty value is for exceeding the accuracy limit; The penalty value for task conflicts;

[0041] 45) Dynamic adjustment triggering conditions: The coordinating agent triggers dynamic adjustment when any of the following conditions are met:

[0042] ① The intensity of sunlight S decreases from S1 to below S2 or increases from S2 to above S1;

[0043] ② The robot's remaining energy E ≤ 30%, real-time execution accuracy δ > 0.2 cm, or fault state F = 1;

[0044] ③ The task completion rate C is more than 10% lower than the planned value;

[0045] ④ Communication delay > 2 seconds lasting for more than 1 minute;

[0046] If triggered, a dynamic adjustment scheme P1 is generated; otherwise, the initial task allocation scheme P0 is retained.

[0047] 46) Lunar environment adaptation mechanism:

[0048] Sunlight-dependent task adaptation: S < 400W / m 2 Stop the sintering task and switch to mold cleaning / equipment self-check task; S≥600W / m 2 Prioritize the allocation of sintering tasks to maximize the utilization of solar energy;

[0049] Low gravity accuracy compensation: triggered when δ > 0.2cm The penalty was imposed, and the task interval of the construction robot was adjusted, while the frequency of visual positioning calibration was increased.

[0050] Energy scarcity adaptation: Triggered when E < 30% and energy utilization rate < 60%. Punishment: Prioritize assigning short-cycle, low-energy-consumption tasks to them;

[0051] Fault robustness adaptation: When F=1, the faulty robot's task is transferred to the backup robot, ensuring that the task interruption time is ≤5min;

[0052] 47) Iterative optimization: Every 5 minutes, based on the data fed back by the robot, update the Q-value using the Q-learning algorithm, as shown in the following formula:

[0053] ;

[0054] in, γ = 0.9 is the learning rate and the discount factor.

[0055] As a further technical solution of the present invention: in 6), the top-level iteration hub of the loop optimization layer specifically includes:

[0056] 61) Multi-source feedback data reception and fusion: Four types of core data are received every 30 minutes: scheduling effect data reported by the dynamic adjustment layer, conflict resolution reports reported by the conflict resolution layer, global status statistics collected by the communication positioning system, and manual intervention instructions issued by the remote interaction and monitoring terminal.

[0057] 62) Multi-dimensional indicator calculation: Five core indicators are calculated based on the received data to form the basis for optimization:

[0058] Periodic indicator: Periodic deviation rate = Threshold ≤ 5%;

[0059] Resource indicator: Average energy efficiency E avg The target is ≥60%;

[0060] Accuracy metric: Accuracy compliance rate = (Number of tasks with δ≤0.2cm / Total number of tasks) × 100%, Target ≥ 98%;

[0061] Conflict indicator: Conflict occurrence rate, threshold ≤3%;

[0062] Stability index: The fluctuation range of each index before and after adjustment, with a threshold of ≤10%;

[0063] 63) Optimize the dynamic determination of target priority:

[0064] Prioritize construction progress as follows: In the initial stage, prioritize resource utilization and task completion rate; in the core stage, prioritize construction accuracy; in the final stage, prioritize construction cycle.

[0065] 64) Quantitative adjustment of core parameters:

[0066] Adjusting key algorithm parameters based on indicator trigger conditions to ensure weights and values ​​comply with regulations: Fitness Function Weights , , Improvement when periodic deviation rate > 5% 0.1-0.2, E avg Improve when <60% 0.1-0.15, improved when accuracy compliance rate is <98%. 0.1-0.2; Reinforcement learning parameters: When the effect fluctuation is >10%, the learning rate α is reduced from 0.1 to 0.05-0.08; when the short-term reward accounts for >80%, the discount factor γ is increased from 0.9 to 0.95; Reward function weight: When the conflict occurrence rate caused by energy waste is >1%, R... energy Increasing from 5 to 10, and raising R when the total conflict rate is >3%. conflict Increased from 15 to 25;

[0067] 65) Triple validation of parameter validity:

[0068] Constraint verification: ensure α∈[0.05,0.2], γ∈[0.9,0.95];

[0069] Simulation pre-verification: Simulate the effect of parameter adjustments based on historical data; if the results are not satisfactory, readjust the parameters.

[0070] Manual review: Adjustments to key parameters require manual confirmation via remote interaction and monitoring terminals;

[0071] 66) Inter-layer collaboration and result output:

[0072] Parameter distribution: Optimized parameters will be sent. , , The initial scheme P0 is generated by sending data to the global planning layer, and the weights of α, γ and reward function are sent to the dynamic adjustment layer to optimize the reinforcement learning strategy.

[0073] Report output: Generate an optimization strategy report, push it to the remote interactive and monitoring terminal for visualization, and at the same time feed back the optimization basis to the conflict resolution layer to help avoid conflict at the source.

[0074] A construction method for a multi-robot collaborative construction system for in-situ lunar soil construction, the method comprising:

[0075] Step 1: The central task scheduling module analyzes the sunlight intensity data fed back by the communication and positioning system using a multi-constraint dynamic task allocation algorithm. If the sunlight intensity is ≥600W / m 2 The algorithm issues task instructions to the lunar soil brick manufacturing unit based on the optimization results of the fitness function. The lunar soil crushing device crushes the lunar soil, and the vibrating screen sends the crushed lunar soil powder into the lunar soil brick mold.

[0076] Step 2: The solar focusing heating device focuses sunlight to heat the lunar soil brick mold at a temperature ≥1200℃ and a holding time ≥30min, causing the lunar soil to sinter and form lunar soil bricks. A multi-constraint dynamic task allocation algorithm monitors the sintering density in real time; when the sintering density is detected to be ≥1.8g / cm³, the algorithm will proceed. 3 At that time, the algorithm triggers the reward function. The reward logic issues a transportation instruction to the conveyor belt; the conveyor belt transports the lunar soil brick mold to the end, the rotating demolding device clamps the mold and rotates it to demold it, and the algorithm simultaneously assigns brick receiving tasks to the lunar soil brick transport robot to ensure that the lunar soil brick loading is completed within 3 minutes after demolding.

[0077] Step 3: The lunar soil brick transport robot transports lunar soil bricks to the construction area according to the task priority and conflict resolution rules planned by the multi-constraint dynamic task allocation algorithm. During transportation, the algorithm detects the terrain slope in real time through the communication and positioning system. If the slope is greater than 12°, the fitness function is recalculated and the path is adjusted to ensure transportation safety. After transportation is completed, the algorithm performs conflict detection and verification on the operation plan of the robot in the construction area. If there is a time-space overlap conflict, it is handled according to the resolution rules and recorded in the conflict resolution report. The construction robot receives the precise placement instructions issued by the algorithm, uses the gripping device to pick up the lunar soil bricks, and places them according to the construction instructions of the in-situ construction control platform. After the algorithm completes the placement of 10 bricks, it extracts the execution error data. If the error is greater than 0.2cm, the reward function is triggered. The penalty logic pauses the task and adjusts the robotic arm posture parameters of the building robot.

[0078] Step 4: Repeat steps 1 to 3 above. Every 30 minutes, the iterative optimization layer of the multi-constraint dynamic task allocation algorithm integrates four types of data: ① scheduling effect data of the dynamic adjustment layer, ② conflict resolution report of the conflict resolution layer, ③ global status statistics, and ④ manual intervention instructions. Based on the integrated data, calculate the cycle deviation rate and resource utilization rate E. avg The system comprises five core indicators: accuracy compliance rate, conflict occurrence rate, and stability index. Optimization objectives are prioritized based on construction progress, and the fitness function is adjusted according to quantitative rules. , , The system employs a triple validation process: reinforcement learning α / γ, reward function weights, and execution parameter validity. Once validated, the parameters are distributed to the global planning layer and the dynamic adjustment layer, generating scheduling effect reports and optimization strategy reports. These reports are then displayed via a remote interactive and monitoring terminal. Based on the data from these two reports, the system dynamically iterates and optimizes the parameters until the construction task is completed.

[0079] The beneficial effects of this invention are:

[0080] 1) This invention adopts an in-situ construction method, directly utilizing lunar soil, avoiding the high cost of transporting materials from Earth, and significantly reducing costs. At the same time, through concentrated solar energy sintering technology, it utilizes abundant solar energy from the moon to accelerate the sintering of lunar soil, improve material forming efficiency, and meet construction needs more quickly.

[0081] 2) Multiple robots work collaboratively in a cluster mode, with each robot handling construction tasks in parallel according to its function. The manufacturing, transportation and construction of lunar soil bricks are closely linked, reducing waiting time for processes, significantly shortening the construction cycle and improving overall construction efficiency.

[0082] 3) This invention clearly plans the clustered construction architecture and process. The control center coordinates the overall situation, and the central task scheduling module accurately allocates tasks according to the task and robot status. From lunar soil collection to lunar soil brick manufacturing and transportation, and then to the construction robots building according to the plan, each link is promoted in an orderly manner, ensuring that the construction process is stable and controllable.

[0083] 4) This invention ensures stable communication and high-precision positioning in complex environments through a communication and positioning system, while an in-situ construction control platform is responsible for path planning, parameter adjustment, and status monitoring. A remote terminal enables visual management. The robot cluster can be dynamically combined and work with algorithms to complete the entire process of lunar soil processing, transportation, and laying.

[0084] 5) This invention overcomes the limitations of traditional solutions in adapting to the lunar environment through deep collaboration between algorithms and hardware, significantly improving construction efficiency and reliability, and providing core technical support for the automated construction of facilities such as lunar bases. Attached Figure Description

[0085] Figure 1 This is a schematic diagram of the system of the present invention; Figure 2 This is a schematic diagram of the multifunctional construction robot cluster of the present invention; Figure 3 This is a schematic diagram of the lunar soil brick manufacturing unit of the present invention; Figure 4 This is a schematic diagram of the robot constructed according to the present invention; Figure 5 This is a schematic diagram of the lunar soil brick of the present invention; Figure 6 This is a schematic diagram of the construction process of the present invention; Figure 7 This is a flowchart of the Multi-Constraint Dynamic Task Assignment Algorithm (MCDTA) of the present invention; Figure 8 This is a schematic diagram of the global planning layer of the multi-constraint dynamic task allocation algorithm of the present invention; Figure 9 This is a schematic diagram of the dynamic adjustment layer of the multi-constraint dynamic task allocation algorithm of the present invention; Figure 10 This is a schematic diagram of the conflict resolution layer of the multi-constraint dynamic task allocation algorithm of the present invention; Figure 11 This is a schematic diagram of the loop optimization layer of the multi-constraint dynamic task allocation algorithm of the present invention;

[0086] In the diagram: 1. Control Center; 2. Multifunctional Construction Robot Cluster; 3. Communication and Positioning System; 4. Central Task Scheduling Module; 5. In-situ Construction Control Platform; 6. Remote Interaction and Monitoring Terminal; 7. Lunar Soil Brick Manufacturing Unit; 701. Lunar Soil Crusher; 702. Vibrating Screen; 703. Lunar Soil Brick Mold; 704. Sunlight Focusing Heating Device; 705. Conveyor Belt; 706. Rotary Demolding Device; 707. Lunar Soil Brick; 708. Tenon Brick; 709. Mortise and Tenon Brick; 8. Lunar Soil Brick Transport Robot; 9. Construction Robot; 901. Mobile Platform; 902. Multi-DOF Robotic Arm; 903. Grasping Device. Detailed Implementation

[0087] 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.

[0088] Example 1, as Figure 1 , Figure 2 and Figure 4 As shown, this embodiment provides a multi-robot collaborative construction system for in-situ construction on lunar soil. The multi-robot collaborative construction system includes a control center 1 and a multi-functional construction robot cluster 2, which is scheduled by the control center 1.

[0089] The control center 1 includes a communication and positioning system 3, a central task scheduling module 4, an in-situ construction control platform 5, and a remote interaction and monitoring terminal 6. The communication and positioning system 3 adopts high-reliability wireless networking technology to ensure stable communication between robots and with the control center in the complex lunar environment. It integrates visual positioning and inertial navigation and utilizes the inherent topographic features of the moon to achieve high-precision robot positioning. The central task scheduling module 4 aims to ensure construction accuracy, minimize construction cycle, and maximize resource utilization. It is equipped with a multi-constraint dynamic task allocation algorithm (MCDTA), which adopts a hybrid optimization framework composed of improved genetic algorithm and reinforcement learning. The in-situ construction control platform 5 uses algorithms to plan construction paths, adjust forming parameters, and monitor construction status based on lunar terrain and environmental data, ensuring construction quality and efficiency in a low-gravity environment. The remote interaction and monitoring terminal 6 visualizes construction information and the operating status of the multi-constraint dynamic task allocation algorithm (MCDTA), supports emergency remote intervention, and has historical data storage and analysis functions.

[0090] The multi-functional construction robot cluster 2 includes a lunar soil brick manufacturing unit 7, a lunar soil brick transport robot 8, and a construction robot 9.

[0091] Among them, the lunar soil brick manufacturing unit 7 collects lunar soil and crushes and screens it, and uses vibration screening and hot sintering molding process to manufacture lunar soil bricks 707 with mortise and tenon structure; the lunar soil brick transport robot 8 adapts to the low gravity and complex terrain of the moon, and transports the lunar soil bricks 707 to the construction site efficiently and stably according to the path planning algorithm and obstacle avoidance technology; the construction robot 9 accurately grabs and lays the lunar soil bricks 707 to complete the construction of the building structure according to the task instructions issued by the central task scheduling module 4 and the printing path planned by the in-situ construction control platform 5, and can automatically adjust its posture and movement according to the terrain changes.

[0092] Example 2, in addition to the technical solutions in Example 1, includes, for example... Figure 3 As shown, the lunar soil brick manufacturing unit 7 also includes a lunar soil crusher 701, a vibrating screen 702, a lunar soil brick mold 703, a sunlight focusing heating device 704, a conveyor belt 705, and a rotary demolding device 706. The lunar soil crusher 701 is located above the vibrating screen 702, the lunar soil brick mold 703 is located below the vibrating screen 702, the sunlight focusing heating device 704 is connected to the lunar soil brick mold 703, and the rotary demolding device 706 is located at the end of the conveyor belt 705.

[0093] Furthermore, the working timing of lunar soil brick manufacturing unit 7 is controlled by the multi-constraint dynamic task allocation algorithm (MCDTA) of the central task scheduling module 4: when the communication positioning system 3 detects sunlight intensity ≥600W / m 2At that time, the algorithm assigns crushing and sintering tasks to lunar soil brick manufacturing unit 7 by increasing the manufacturing task association weight in the fitness function; when the sunlight intensity is <400W / m 2 When the time comes, the algorithm triggers the penalty logic of Renergy in the reward function, stops the sintering task, and switches to mold cleaning and equipment self-check tasks to avoid energy waste.

[0094] like Figure 5 As shown, the lunar soil brick 707 includes a tenon brick 708 and a mortise and tenon brick 709. One end of the tenon brick 708 is provided with a protruding tenon structure. The tenon is trapezoidal, narrow at the top and wide at the bottom, and has anti-slip texture on both sides. One end of the mortise and tenon brick 709 is provided with a mortise and tenon groove that matches the tenon. The mortise and tenon groove has a protrusion that matches the anti-slip texture of the tenon.

[0095] Furthermore, the manufacturing quantity and transportation batches of Lunar Soil Brick 707 are dynamically adjusted by the Multi-Constraint Dynamic Task Allocation Algorithm (MCDTA) of the Central Task Scheduling Module 4 according to the construction task progress: the algorithm calculates the fitness function in real time... (Execution error) and The correlation of (total cycle) controls the number of lunar soil bricks manufactured and the number of transport batches, ensuring that the construction robot 9 has no waiting for materials, and that the lunar soil brick inventory is ≤ 15% of the total load of the transport robot 8 (total load = number of transport robots × maximum brick capacity per unit).

[0096] like Figure 4 As shown, the construction robot 9 includes a mobile platform 901, a multi-degree-of-freedom robotic arm 902, and a gripping device 903. The mobile platform 901 is connected to the multi-degree-of-freedom robotic arm 902, and the multi-degree-of-freedom robotic arm 902 is connected to the gripping device 903. The gripping device 903 grips the lunar soil brick 707 in a clamping manner.

[0097] Furthermore, the task priority and start / stop time of construction robot 9 are determined by the multi-constraint dynamic task allocation algorithm (MCDTA) of the central task scheduling module 4: when the algorithm detects that the execution error of construction robot 9 is >0.2cm, it automatically promotes the fitness function. The value of (accuracy weight) was increased from 0.3 to 0.4, and the task interval of the robot was adjusted from 5 min / time to 3 min / time. The frequency of visual positioning calibration was increased to ensure that the subsequent execution error was ≤0.2cm.

[0098] Example 3, in addition to the technical solution in Example 1, also includes:

[0099] like Figure 7 , Figure 8 , Figure 9 , Figure 10 and Figure 11As shown, the central task scheduling module 4, with the optimization objectives of ensuring construction accuracy, minimizing construction cycle, and maximizing resource utilization, is equipped with a multi-constraint dynamic task allocation algorithm (MCDTA). The MCDTA algorithm adopts a hybrid optimization framework composed of an improved genetic algorithm and reinforcement learning, and specifically performs the following operations:

[0100] S1. Receive the overall construction task input from the remote interactive and monitoring terminal 6, and decompose it into a task set T={T1, T2, ..., T...} according to the three-level structure module-subtask-operation unit. n Each subtask carries its type (manufacturing / transportation / construction), priority (P1 critical structure / P2 auxiliary structure / P3 decorative structure), resource requirements, and spatiotemporal constraints (including sunlight intensity ≥600W / m). 2 The sintering time window and precision requirements (mortise and tenon joint error ≤ 0.2cm);

[0101] S2. Collect environmental status data (sunlight intensity, terrain slope / obstacle distribution, communication delay 0.5-2s), robot status data (remaining energy, working status, load capacity, historical execution accuracy) and task progress data through the communication and positioning system 3 to construct an environment-robot-task status matrix.

[0102] S3. Implement initial task allocation based on an improved genetic algorithm, specifically including:

[0103] S31. Chromosome Encoding: The format is “Robot ID, Task ID, Start Time, End Time, Constraint Satisfaction”, where the constraint satisfaction represents the degree to which the spatiotemporal constraints, terrain constraints, and energy constraints of the subtask are satisfied (value range 0-1, 1 indicates complete satisfaction).

[0104] S32. Evaluate the merits of task allocation schemes using fitness functions:

[0105] ;

[0106] Where F is the fitness value (ranging from 0 to 1, with a larger value indicating a better solution); , , The weights are dynamic, and It is dynamically adjusted according to the lunar environment; Total construction period (in hours) for the current task allocation plan; The theoretical maximum allowable period for this construction task (in hours) is predetermined by the requirements of the construction task. The average energy utilization rate of all robots (value range 0-1, i.e. 0%-100%). This represents the energy utilization rate of the robot when it is running at full load (value 1, i.e., 100%). The average execution error for all subtasks (unit: cm); This represents the maximum allowable execution error for the subtask (valued at 0.2cm, which is the precision threshold for mortise and tenon joints).

[0107] S33. Genetic operations: After selection (roulette wheel + elite retention strategy, retaining the top 20% of chromosomes in fitness), crossover (task-time segment crossover, verifying constraint satisfaction after crossover), and mutation (randomly modifying the robot number of 10% of chromosomes), the initial task allocation scheme P0 is output after 100-200 generations.

[0108] S4. Dynamic optimization based on reinforcement learning: Each robot in the multifunctional construction robot cluster 2 is defined as an independently executing intelligent agent, and the central task scheduling module 4 acts as the globally coordinating intelligent agent; specifically including:

[0109] S41. State Space Construction: Define the system state s at time t. t =[Sunlight intensity S, Robot remaining energy E, Real-time execution accuracy δ, Task completion rate C, Fault status F];

[0110] Among them, the solar intensity S is quantified in intervals, with S1≥600W / m 2 400W / m 2 ≤S2<600W / m 2 S3 < 400W / m 2 The robot's remaining energy E is quantified proportionally, with E1≥60%, 30%≤E2<60%, and E3<30%; the real-time execution accuracy δ is quantified according to the error threshold, with δ1≤0.2cm and δ2>0.2cm; the task completion rate C is quantified as a percentage, with 0≤C≤100%; the fault status F is quantified in binary, with F=0 indicating no fault and F=1 indicating a fault.

[0111] S42. Action Space Definition:

[0112] The actions performed by the intelligent agent include:

[0113] ① Task acceptance / rejection (low-energy robots can reject P3 priority tasks); ② Task type switching (only for Lunar Brick Manufacturing Unit 7, such as switching from sintering tasks to mold cleaning / equipment self-inspection tasks); ③ Task transfer request (triggered when F=1, requesting to transfer the task to the backup robot); ④ Operation parameter adjustment (such as increasing the frequency of visual positioning calibration in the δ2 ​​state of the construction robot); ⑤ Time window request (requesting to delay the execution of the task when the transport robot encounters terrain obstacles).

[0114] S43. Action selection strategy: Adopt the ε-greedy strategy (ε=0.1), with a 90% probability of selecting the optimal action with the largest current Q value and a 10% probability of randomly selecting an action, balancing strategy exploration and utilization to avoid getting trapped in local optima;

[0115] S44. Reward Function: Used to evaluate the rationality of the agent's actions, as shown in the following formula:

[0116] ;

[0117] in, Let be the reward value at time t; The reward value for completing the task on time (fixed value of 10, value of 0 if the task is not completed on time); This is a penalty value for energy waste (value 5, triggered when robot energy utilization rate is <60%, value 0 when utilization rate is ≥60%). The penalty value for exceeding the accuracy limit is 20, triggered when the execution error is >0.2cm, and 0 when the error is ≤0.2cm. The penalty value for task conflicts (value 15, triggered when there is a "time-space" overlapping conflict, and 0 when there is no conflict).

[0118] S45. Dynamic Adjustment Triggering Conditions: The coordinating agent triggers dynamic adjustment when any of the following conditions are met:

[0119] ① The intensity of sunlight S decreases from S1 to below S2 or increases from S2 to above S1;

[0120] ② The robot's remaining energy E ≤ 30%, real-time execution accuracy δ > 0.2 cm, or fault state F = 1;

[0121] ③ The task completion rate C is more than 10% lower than the planned value;

[0122] ④ Communication delay > 2 seconds lasting for more than 1 minute;

[0123] If triggered, a dynamic adjustment scheme P1 is generated; otherwise, the initial task allocation scheme P0 is retained.

[0124] S46. Lunar environment adaptation mechanism:

[0125] ① Sunlight-dependent task adaptation: S < 400W / m 2 Stop the sintering task and switch to mold cleaning / equipment self-check task; S≥600W / m 2 Prioritize the allocation of sintering tasks to maximize the utilization of solar energy;

[0126] ②Low gravity accuracy compensation: Triggered when δ>0.2cm Penalties were imposed, and the task interval for the construction robot was adjusted (from 5 minutes / time to 3 minutes / time), while the frequency of visual positioning calibration was increased.

[0127] ③ Energy scarcity adaptation: Triggered when E < 30% and energy utilization rate < 60%. Punishment: Prioritize assigning short-cycle, low-energy-consumption tasks to them;

[0128] ④ Fault robustness adaptation: When F=1, the faulty robot's task is transferred to the backup robot to ensure that the task interruption time is ≤5min;

[0129] S47. Iterative Optimization: Every 5 minutes, based on the (state-action-reward) data fed back by the robot, update the Q value using the Q-learning algorithm, as shown in the following formula:

[0130] ;

[0131] in, The learning rate is γ=0.9, which is the discount factor. The task allocation is dynamically adjusted to cope with scenarios such as sunlight fluctuations and robot malfunctions.

[0132] S5, Built-in conflict detection and resolution mechanism:

[0133] By comparing the time-space parameters of each robot task, overlapping conflicts were identified, and the conflict types were clarified as follows: ① Time conflict (only the task start / end time windows overlap, and the work areas do not intersect); ② Spatial conflict (only the work areas overlap, and the time windows do not intersect); ③ Mixed conflict (both the time window and the work area overlap, which can easily lead to task interruption).

[0134] Conflict levels are categorized based on task priority: ① High-level conflicts: Conflicts involving P1 task robots, and mixed conflicts; ② Medium-level conflicts: Conflicts involving P2 task robots; ③ Low-level conflicts: Conflicts involving P3 task robots. Conflicts are resolved according to the following rules: "P1 task robots take precedence over P2 / P3 task robots; robots with less than 30% remaining energy take precedence over high-energy robots; robots with historical execution accuracy ≥ 98% take precedence over low-accuracy robots." After resolution, the time-space parameters of each robot's task are re-checked. If the conflict incidence rate is > 3%, the above resolution rules are re-executed until the conflict incidence rate is ≤ 3%, ensuring that the conflict incidence rate is ≤ 3%.

[0135] S6, the top-level iterative hub of the loop optimization layer, specifically includes:

[0136] S61. Multi-source feedback data reception and fusion: Receive four types of core data every 30 minutes: ① Scheduling effect data reported by the dynamic adjustment layer (including total construction cycle deviation, resource utilization rate E) avg① Accuracy compliance rate and conflict occurrence rate); ② Conflict resolution reports reported by the conflict resolution layer; ③ Global status statistics collected by the communication and positioning system (frequency of lunar environment sunlight fluctuations, statistics of terrain obstacle distribution, average failure rate of robot clusters, and energy consumption trends); ④ Manual intervention commands issued by the remote interaction and monitoring terminal 6.

[0137] S62. Multi-dimensional indicator calculation: Based on the received data, five core indicators are calculated to form the basis for optimization.

[0138] ① Periodic indicator: Periodic deviation rate = (Threshold ≤ 5%)

[0139] ② Resource indicators: Average energy utilization rate E avg (Target ≥ 60%)

[0140] ③ Accuracy indicator: Accuracy compliance rate = (number of tasks with δ≤0.2cm / total number of tasks) × 100% (target ≥98%);

[0141] ④ Conflict indicator: Conflict occurrence rate (threshold ≤ 3%);

[0142] ⑤ Stability indicators: Fluctuation range of each indicator before and after adjustment (threshold ≤ 10%).

[0143] S63. Optimize the dynamic determination of target priority:

[0144] Prioritization based on construction progress:

[0145] ① Initial stage (0-30% progress): Prioritize resource utilization → task completion rate;

[0146] ② Core Phase (30%-70% progress): Prioritize ensuring construction accuracy (focus on P1 task);

[0147] ③ Final stage (70%-100% progress): Prioritize ensuring the construction period;

[0148] S64, Core Parameter Quantization Adjustment:

[0149] Adjust key algorithm parameters based on indicator trigger conditions to ensure weights and values ​​comply with regulations.

[0150] ①Fitness function weights , , ( Improvement when the period deviation rate is >5% 0.1-0.2, E avg Improve when <60% 0.1-0.15, improved when accuracy compliance rate is <98%. 0.1-0.2;

[0151] ② Reinforcement learning parameters: When the effect fluctuation is greater than 10%, the learning rate α is reduced from 0.1 to 0.05-0.08; when the short-term reward accounts for more than 80%, the discount factor γ is increased from 0.9 to 0.95.

[0152] ③ Reward function weight: When the conflict occurrence rate caused by energy waste is >1%, R will be weighted. energy Increasing from 5 to 10, and raising R when the total conflict rate is >3%. conflict Increased from 15 to 25;

[0153] S65. Triple validation of parameter validity:

[0154] ①Constraint verification: Ensure α∈[0.05, 0.2], γ∈[0.9, 0.95];

[0155] ② Simulation pre-verification: Based on historical data, simulate the effect of parameter adjustments; if the results are not satisfactory, readjust the parameters.

[0156] ③ Manual review: Adjustment of key parameters (such as...) Values ​​≥0.5 and α≤0.08 require manual confirmation via a remote interactive monitoring terminal.

[0157] S66. Inter-layer collaboration and result output:

[0158] ①Parameter distribution: The optimized parameters will be distributed. , , The initial scheme P0 is generated by sending data to the global planning layer, and the weights of α, γ and reward function are sent to the dynamic adjustment layer to optimize the reinforcement learning strategy.

[0159] ② Report Output: Generate an optimization strategy report (including the basis for parameter adjustment, expected results, and iteration suggestions), and push it to the remote interactive and monitoring terminal for visualization; at the same time, feed back the optimization basis to the conflict resolution layer to help avoid conflicts at the source.

[0160] Example 3, as Figure 6 As shown in the figure, this embodiment discloses a construction method for a multi-robot collaborative construction system for in-situ lunar soil construction. The construction method includes the following steps:

[0161] Step 1: The central task scheduling module 4 analyzes the sunlight intensity data fed back by the communication and positioning system 3 using the Multi-Constraint Dynamic Task Allocation Algorithm (MCDTA). If the sunlight intensity is ≥600W / m 2The algorithm issues task instructions to lunar soil brick manufacturing unit 7 based on the fitness function optimization results. Lunar soil crusher 701 crushes lunar soil, and vibrating screen 702 sends the crushed lunar soil powder (particle size ≤ 0.5 mm) into lunar soil brick mold 703.

[0162] Step 2: The sunlight focusing heating device 704 focuses sunlight to heat the lunar soil brick mold 703 at a heating temperature ≥1200℃ and a holding time ≥30min, so that the lunar soil is sintered to form lunar soil bricks 707; the multi-constraint dynamic task allocation algorithm (MCDTA) monitors the sintering density in real time, and when the sintering density is detected to be ≥1.8g / cm³, the brick is sintered. 3 At that time, the algorithm triggers the reward function. The reward logic sends a transport instruction to the conveyor belt 705; the conveyor belt 705 transports the lunar soil brick mold 703 to the end, the rotating demolding device 706 clamps the mold and rotates it to demold, and the algorithm simultaneously assigns brick receiving tasks to the lunar soil brick transport robot 8 to ensure that the lunar soil brick loading is completed within 3 minutes after demolding.

[0163] Step 3: The lunar soil brick transport robot 8 transports lunar soil bricks 707 to the construction area according to the task priority and conflict resolution rules planned by the Multi-Constraint Dynamic Task Allocation Algorithm (MCDTA). During transportation, the algorithm detects the terrain slope in real time through the communication and positioning system 3. If the slope is greater than 12°, the fitness function is recalculated and the path is adjusted to ensure transportation safety. After transportation is completed, the algorithm performs conflict detection and verification on the work plan of the construction area robot 9. If there is a time-space overlap conflict, it is handled according to the resolution rules and recorded in the conflict resolution report. The construction robot 9 receives the precise placement instructions from the algorithm, uses the gripping device 903 to pick up the lunar soil bricks 707, and places them according to the construction instructions of the in-situ construction control platform 5. After the algorithm completes the placement of 10 bricks, it extracts the execution error data. If the error is greater than 0.2cm, the reward function is triggered. The penalty logic pauses the task and adjusts the robotic arm posture parameters of the construction robot 9.

[0164] Step 4: Repeat steps 1 to 3 above. The cyclic optimization layer of the Multi-Constraint Dynamic Task Assignment Algorithm (MCDTA) integrates four types of data every 30 minutes: ① scheduling effect data of the dynamic adjustment layer, ② conflict resolution report of the conflict resolution layer, ③ global state statistics, and ④ manual intervention instructions. Based on the integrated data, the cycle deviation rate and resource utilization rate E are calculated. avg Five core indicators are considered: accuracy achievement rate, conflict occurrence rate, and stability index. Optimization objectives are prioritized based on construction progress (resource utilization rate in the initial stage, accuracy in the core stage, and cycle time in the final stage), and the fitness function is adjusted according to quantitative rules. , , The system employs reinforcement learning α / γ and reward function weights, and performs a triple verification of parameter validity (constraint verification → simulation pre-verification → manual review of key parameters). After verification, the parameters are distributed to the global planning layer and the dynamic adjustment layer, generating scheduling effect reports and optimization strategy reports. These reports are displayed together through a remote interactive and monitoring terminal. Based on the data from these two types of reports, the parameters are dynamically iterated and optimized until the construction task is completed.

[0165] The system consists of a control center and a cluster of multifunctional construction robots. The control center includes a communication and positioning system, a central task scheduling module, an in-situ construction control platform, and a remote interaction and monitoring terminal. The core central task scheduling module is equipped with a Multi-Constraint Dynamic Task Allocation (MCDTA) algorithm, constructing a global planning-dynamic adjustment-conflict resolution-cyclic optimization collaborative optimization framework. It outputs an initial scheme P0 based on an improved genetic algorithm, and combines reinforcement learning to generate a dynamic scheme P1 to cope with the dynamic lunar environment. A built-in time-space dual-dimensional conflict detection and resolution mechanism ensures a conflict rate of ≤3%. Through iterative adjustments of core parameters via cyclic optimization layers, it achieves the goals of improving accuracy, shortening the cycle, and increasing efficiency. The communication and positioning system ensures stable communication and high-precision positioning in complex environments. The in-situ construction control platform is responsible for path planning, parameter adjustment, and status monitoring. The remote terminal enables visual management. The robot cluster can dynamically combine to complete the entire process of lunar soil processing, transportation, and laying, in conjunction with the algorithm.

[0166] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0167] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A multi-robot collaborative construction system for in-situ construction on lunar soil, characterized in that: It includes a control center (1) and a multi-functional cluster robot (2), which is scheduled by the control center (1); The control center (1) includes a communication positioning system (3), a central task scheduling module (4), an in-situ construction control platform (5), and a remote interaction and monitoring terminal (6). The multifunctional cluster robot (2) includes a lunar soil brick manufacturing unit (7), a lunar soil brick transport robot (8), and a construction robot (9).

2. The multi-robot collaborative construction system according to claim 1, characterized in that: The lunar soil brick manufacturing unit (7) collects lunar soil and performs crushing and screening processes, and manufactures lunar soil bricks (707) with mortise and tenon structures using vibrating screening and hot sintering molding processes.

3. The multi-robot collaborative construction system according to claim 1, characterized in that: The lunar soil brick manufacturing unit (7) further includes a lunar soil crushing device (701), a vibrating screen (702), a lunar soil brick mold (703), a sunlight focusing heating device (704), a conveyor belt (705), a rotary demolding device (706), and lunar soil bricks (707). The lunar soil crusher (701) is located above the vibrating screen (702), the lunar soil brick mold (703) is located below the vibrating screen (702), the sunlight focusing heating device (704) is connected to the lunar soil brick mold (703), and the rotary demolding device (706) is located at the end of the conveyor belt (705). The working sequence of the lunar soil brick manufacturing unit (7) is controlled by the multi-constraint dynamic task allocation algorithm of the central task scheduling module (4): when the communication positioning system (3) detects that the sunlight intensity is ≥600W / m 2 At that time, the algorithm assigns crushing and sintering tasks to the lunar soil brick manufacturing unit (7) by increasing the manufacturing task association weight in the fitness function; when the sunlight intensity is <400W / m 2 At that time, the algorithm triggers the reward function. The penalty logic stops the sintering task and switches to mold cleaning and equipment self-check tasks.

4. The multi-robot collaborative construction system according to claim 1, characterized in that: The lunar soil brick (707) includes a tenon brick (708) and a mortise and tenon brick (709). One end of the tenon brick (708) is provided with a protruding tenon structure, and one end of the mortise and tenon brick (709) is provided with a mortise and tenon groove that matches the tenon. The mortise and tenon groove is provided with a protrusion that matches the anti-slip texture of the tenon. The manufacturing quantity and transportation batches of the lunar soil bricks (707) are dynamically adjusted by the multi-constraint dynamic task allocation algorithm of the central task scheduling module (4) according to the construction task progress: the algorithm calculates the fitness function in real time. and The relationship between the two is used to control the number of lunar soil bricks manufactured and the batches of transportation, to ensure that the construction robot (9) has no waiting for materials and that the lunar soil brick inventory is ≤ 15% of the total load of the transportation robot (8).

5. The multi-robot collaborative construction system according to claim 1, characterized in that: The construction robot (9) includes a mobile platform (901), a multi-degree-of-freedom robotic arm (902), and a gripping device (903). The mobile platform (901) is connected to the multi-degree-of-freedom robotic arm (902), and the multi-degree-of-freedom robotic arm (902) is connected to the gripping device (903). The gripping device (903) grips the lunar soil brick (707) in a clamping manner. The operation priority and task start / stop time of the construction robot (9) are determined by the multi-constraint dynamic task allocation algorithm of the central task scheduling module (4): when the algorithm detects that the execution error of the construction robot (9) is >0.2cm, it automatically increases the fitness function. The value of is increased, and the task interval of the robot is adjusted to increase the frequency of visual positioning calibration, so as to ensure that the subsequent execution error is ≤0.2cm.

6. The multi-robot collaborative construction system according to claim 1, characterized in that: The central task scheduling module (4) is equipped with a multi-constraint dynamic task allocation algorithm. The multi-constraint dynamic task allocation algorithm adopts a hybrid optimization framework composed of an improved genetic algorithm and reinforcement learning, and specifically performs the following operations: 1) Receive the total construction task input from the remote interactive and monitoring terminal (6), and decompose it into a task set T={T1,T2,...,T...} according to the three levels of structural module-subtask-operation unit. n Each subtask carries its type, priority, resource requirements, time and space constraints, and precision requirements. 2) Collect environmental status data, robot status data and task progress data through the communication positioning system (3) to construct an environment-robot-task status matrix; 3) Initial task allocation is achieved based on an improved genetic algorithm; 4) Dynamic optimization based on reinforcement learning: Each robot in the multifunctional construction robot cluster (2) is defined as an independently executing intelligent agent, and the central task scheduling module (4) is used as the globally coordinated intelligent agent; 5) Built-in conflict detection and resolution mechanism: By comparing the temporal-spatial parameters of each robot task, overlapping conflicts were identified, and the conflict types were clarified as: temporal conflicts, spatial conflicts, and mixed conflicts. Conflict levels are categorized by the priority of the tasks involved: High-level conflicts: conflicts involving P1 task bots and mixed conflicts; Medium-level conflicts: conflicts involving P2 task bots; Low-level conflicts: conflicts involving P3 task bots. The conflict is resolved according to the following rules: "P1 task robots take priority over P2 / P3 task robots, robots with remaining energy <30% take priority over high-energy robots, and robots with historical execution accuracy ≥98% take priority over low-accuracy robots". After resolution, the time-space parameters of each robot task are re-checked. If the conflict occurrence rate is >3%, the above resolution rules are re-executed until the conflict occurrence rate is ≤3%, ensuring that the conflict occurrence rate is ≤3%. 6) The top-level iterative hub of the loop optimization layer.

7. The multi-robot collaborative construction system according to claim 6, characterized in that: In step 3), the initial task allocation based on the improved genetic algorithm specifically includes: 31) Chromosome encoding: adopts the format of "robot number, task number, start time, end time, constraint satisfaction", where constraint satisfaction represents the degree to which the spatiotemporal constraints, terrain constraints, and energy constraints of the subtask are satisfied; 32) Fitness function, used to evaluate the quality of task allocation schemes: ； Where F is the fitness value; , , The weights are dynamic, and It is dynamically adjusted according to the lunar environment; The total construction period for the current task allocation plan; The theoretical maximum allowable period for this construction task is predetermined by the requirements of the construction task; The average energy efficiency of all robots; Energy utilization rate when the robot is running at full load; The average execution error of all subtasks; This represents the maximum permissible execution error for the subtask. 33) Genetic operation: After selection, crossover, mutation, and iteration, the initial task allocation scheme P0 is output.

8. The multi-robot collaborative construction system according to claim 7, characterized in that: In step 4), dynamic optimization is achieved based on reinforcement learning: each robot in the multifunctional construction robot cluster (2) is defined as an independently executing intelligent agent, and the central task scheduling module (4) serves as the globally coordinating intelligent agent, specifically including: 41) State space construction: Define the system state s at time t. t =[Sunlight intensity S, Robot remaining energy E, Real-time execution accuracy δ, Task completion rate C, Fault status F]; Among them, the solar intensity S is quantified in intervals, with S1≥600W / m 2 400W / m 2 ≤S2<600W / m 2 S3 < 400W / m 2 The robot's remaining energy E is quantified proportionally, with E1≥60%, 30%≤E2<60%, and E3<30%; the real-time execution accuracy δ is quantified according to the error threshold, with δ1≤0.2cm and δ2>0.2cm; the task completion rate C is quantified as a percentage, with 0≤C≤100%; the fault status F is quantified in binary, with F=0 indicating no fault and F=1 indicating a fault. 42) Action Space Definition: The actions of the agent include: task acceptance / rejection; task type switching; task transfer request; job parameter adjustment; time window request; 43) Action selection strategy: The ε-greedy strategy is adopted, with a 90% probability of selecting the optimal action with the largest current Q value and a 10% probability of randomly selecting an action; 44) Reward function: Used to evaluate the rationality of the agent's actions, as shown in the following formula: ； in: Let be the reward value at time t; The reward value for completing the task on time; The penalty value for energy waste; The penalty value is for exceeding the accuracy limit; The penalty value for task conflicts; 45) Dynamic adjustment triggering conditions: The coordinating agent triggers dynamic adjustment when any of the following conditions are met: ① The intensity of sunlight S decreases from S1 to below S2 or increases from S2 to above S1; ② The robot's remaining energy E ≤ 30%, real-time execution accuracy δ > 0.2 cm, or fault state F = 1; ③ The task completion rate C is more than 10% lower than the planned value; ④ Communication delay > 2 seconds lasting for more than 1 minute; If triggered, a dynamic adjustment scheme P1 is generated; otherwise, the initial task allocation scheme P0 is retained. 46) Lunar environment adaptation mechanism: Sunlight-dependent task adaptation: S < 400W / m 2 Stop the sintering task and switch to mold cleaning / equipment self-check task; S≥600W / m 2 Prioritize the allocation of sintering tasks to maximize the utilization of solar energy; Low gravity accuracy compensation: triggered when δ > 0.2cm The penalty was imposed, and the task interval of the construction robot was adjusted, while the frequency of visual positioning calibration was increased. Energy scarcity adaptation: Triggered when E < 30% and energy utilization rate < 60%. Punishment: Prioritize assigning short-cycle, low-energy-consumption tasks to them; Fault robustness adaptation: When F=1, the faulty robot's task is transferred to the backup robot, ensuring that the task interruption time is ≤5min; 47) Iterative optimization: Every 5 minutes, based on the data fed back by the robot, update the Q-value using the Q-learning algorithm, as shown in the following formula: ； in, γ = 0.9 is the learning rate and the discount factor.

9. The multi-robot collaborative construction system according to claim 8, characterized in that: In step 6), the top-level iteration hub of the loop optimization layer specifically includes: 61) Multi-source feedback data reception and fusion: every 30 minutes, four types of core data are received: scheduling effect data reported by the dynamic adjustment layer, conflict resolution report reported by the conflict resolution layer, global status statistics collected by the communication positioning system, and manual intervention instructions issued by the remote interaction and monitoring terminal (6). 62) Multi-dimensional indicator calculation: Five core indicators are calculated based on the received data to form the basis for optimization: Periodic indicator: Periodic deviation rate = Threshold ≤ 5%; Resource indicators: Average energy efficiency E avg The target is ≥60%; Accuracy metric: Accuracy compliance rate = (Number of tasks with δ≤0.2cm / Total number of tasks) × 100%, Target ≥ 98%; Conflict indicator: Conflict occurrence rate, threshold ≤3%; Stability index: The fluctuation range of each index before and after adjustment, with a threshold of ≤10%; 63) Optimize the dynamic determination of target priority: Prioritize construction progress as follows: In the initial stage, prioritize resource utilization and task completion rate; in the core stage, prioritize construction accuracy; in the final stage, prioritize construction cycle. 64) Quantitative adjustment of core parameters: Adjusting key algorithm parameters based on indicator trigger conditions to ensure weights and values ​​comply with regulations: Fitness Function Weights , , Improvement when periodic deviation rate > 5% 0.1-0.2, E avg Improve when <60% 0.1-0.15, improved when accuracy compliance rate is <98%. 0.1-0.2; Reinforcement learning parameters: When the effect fluctuation is >10%, the learning rate α is reduced from 0.1 to 0.05-0.08; when the short-term reward accounts for >80%, the discount factor γ is increased from 0.9 to 0.95; Reward function weight: When the conflict occurrence rate caused by energy waste is >1%, R... energy Increasing from 5 to 10, and raising R when the total conflict rate is >3%. conflict Increased from 15 to 25; 65) Triple validation of parameter validity: Constraint verification: ensure α∈[0.05,0.2], γ∈[0.9,0.95]; Simulation pre-verification: Simulate the effect of parameter adjustments based on historical data; if the results are not satisfactory, readjust the parameters. Manual review: Adjustments to key parameters require manual confirmation via a remote interactive and monitoring terminal (6); 66) Inter-layer collaboration and result output: Parameter distribution: Optimized parameters will be sent. , , The initial scheme P0 is generated by sending data to the global planning layer, and the weights of α, γ and reward function are sent to the dynamic adjustment layer to optimize the reinforcement learning strategy. Report output: Generate an optimization strategy report and push it to the remote interaction and monitoring terminal (6) for visualization display. At the same time, the optimization basis is fed back to the conflict resolution layer to help avoid conflict at the source.

10. A construction method for a multi-robot collaborative construction system for in-situ lunar soil construction, wherein the construction includes the multi-robot collaborative construction system as described in any one of claims 1 to 9, characterized in that: The construction method includes: Step 1: The central task scheduling module (4) analyzes the sunlight intensity data fed back by the communication positioning system (3) through a multi-constraint dynamic task allocation algorithm. If the sunlight intensity is ≥600W / m 2 The algorithm issues a task instruction to the lunar soil brick manufacturing unit (7) based on the fitness function optimization result. The lunar soil crushing device (701) crushes the lunar soil, and the vibrating screen (702) sends the crushed lunar soil powder into the lunar soil brick mold (703). Step 2: The solar focusing heating device (704) focuses sunlight to heat the lunar soil brick mold (703) at a heating temperature ≥1200℃ and a holding time ≥30min, so that the lunar soil is sintered to form lunar soil bricks (707); the multi-constraint dynamic task allocation algorithm monitors the sintering density in real time, and when the sintering density is detected to be ≥1.8g / cm³, the algorithm will determine the sintering density. 3 At that time, the algorithm triggers the reward function. The reward logic sends a transport instruction to the conveyor belt (705); the conveyor belt (705) transports the lunar soil brick mold (703) to the end, the rotating demolding device (706) clamps the mold and rotates to demold it, and the algorithm synchronously assigns brick receiving tasks to the lunar soil brick transport robot (8) to ensure that the lunar soil brick loading is completed within 3 minutes after demolding. Step 3: The lunar soil brick transport robot (8) transports the lunar soil bricks (707) to the construction area according to the task priority and conflict resolution rules planned by the multi-constraint dynamic task allocation algorithm. During the transportation process, the algorithm detects the terrain slope in real time through the communication positioning system (3). If the slope is greater than 12°, the fitness function is recalculated and the path is adjusted to ensure transportation safety. After the transportation is completed, the algorithm performs conflict detection and verification on the operation plan of the construction area robot (9). If there is a time-space overlap conflict, it is handled according to the resolution rules and recorded in the conflict resolution report. The construction robot (9) receives the precise placement instructions issued by the algorithm, uses the gripping device (903) to pick up the lunar soil bricks (707), and places them according to the construction instructions of the in-situ construction control platform (5). After the algorithm completes the placement of 10 bricks, it extracts the execution error data. If the error is greater than 0.2cm, the reward function is triggered. The penalty logic is to pause the task and adjust the robotic arm posture parameters of the building robot (9); Step 4: Repeat steps 1 to 3 above. Every 30 minutes, the iterative optimization layer of the multi-constraint dynamic task allocation algorithm integrates four types of data: ① scheduling effect data of the dynamic adjustment layer, ② conflict resolution report of the conflict resolution layer, ③ global status statistics, and ④ manual intervention instructions. Based on the integrated data, calculate the cycle deviation rate and resource utilization rate E. avg The system comprises five core indicators: accuracy compliance rate, conflict occurrence rate, and stability index. Optimization objectives are prioritized based on construction progress, and the fitness function is adjusted according to quantitative rules. , , The parameters are validated three times: reinforcement learning α / γ, reward function weight, and execution parameter validity. After the validation is passed, the parameters are sent to the global planning layer and the dynamic adjustment layer respectively, generating scheduling effect report and optimization strategy report. They are displayed together through remote interaction and monitoring terminal (6). The parameters are dynamically iterated and optimized based on the two types of report data until the construction task is completed.

Images