A dedicated rescue vehicle system for performing rescue missions within congested tunnels and methods of use thereof

Abstract

This invention relates to a dedicated rescue vehicle system and its method of use for carrying out rescue missions in congested tunnels, comprising a mother vehicle command unit, a subsidiary vehicle operation unit, an unmanned reconnaissance unit, and a communication network unit. The mother vehicle command unit receives the alarm and plans the optimal route; the unmanned reconnaissance unit conducts three-dimensional reconnaissance after the rescue vehicle arrives at the scene and transmits data back to construct a three-dimensional digital twin model of the tunnel; simultaneously, based on the model, AI task planning is performed, and the subsidiary vehicle operation unit is commanded to perform tasks such as clearing obstacles and creating life-saving passages; at the same time, the communication network unit ensures uninterrupted internal and external communication throughout the process. The systematic process and highly coordinated units ensure seamless connection from alarm reception to deployment. Based on the three-dimensional digital twin model and utilizing AI for optimal route planning and task allocation, commanders can quickly and intuitively grasp the overall situation, make rapid decisions, and open the best life-saving passages, greatly improving the accuracy and overall efficiency of rescue operations.

Description

Technical Field

[0001] This invention relates to the field of emergency rescue technology, and in particular to a dedicated rescue vehicle system for performing rescue missions in congested tunnels and its method of use. Background Technology

[0002] Tunnels, especially long and underground tunnels, are highly susceptible to traffic congestion during emergencies such as traffic accidents and fires, which can prevent rescue forces from quickly reaching the core scene. Current tunnel rescue methods rely primarily on manual labor, which has significant drawbacks. For example, they cannot quickly obtain comprehensive information on the overall congestion structure, the precise location of obstacles, the distribution of hazards, and the real-time status of trapped personnel. AI decision-making is mainly based on outdated or static maps, failing to respond to the rapidly changing dynamic environment of the rescue site. This results in a lack of pre-operation information, making decision-making a high-risk gamble. Furthermore, it cannot dynamically assign the most urgent tasks to the most suitable units based on the real-time situation, leading to low resource utilization, response delays, and impacting rescue efficiency. Summary of the Invention

[0003] (a) Technical problems to be solved In view of the problems in existing tunnel emergency rescue systems, this invention transforms the rescue site into a real-time, computable environment through a dynamically updated 3D digital twin model. This allows AI decision-making to be based on real-time data reflecting the latest situation, solving the decision-making bias caused by inaccurate or incomplete information. Simultaneously, through AI collaborative decision-making and simulation algorithms, the consequences of operations are pre-simulated in the digital twin model before action, realizing a shift from experience-based decision-making to simulation-verified decision-making, greatly mitigating the secondary risks of blind action. Their collaboration enables rescue command to proactively plan the globally optimal life-saving passage and dynamically and safely allocate tasks, improving the accuracy, safety, and overall efficiency of rescue operations.

[0004] (II) Technical Solution To achieve the above objectives, the present invention provides the following technical solution: a dedicated rescue vehicle system for performing rescue missions in congested tunnels, characterized in that it comprises: (a) Mother vehicle command unit: As the main command center, it is used for overall command, route planning, situation generation and task allocation, including onboard computer, data fusion module, 3D digital twin modeling module and task planning module; (b) Sub-vehicle operation unit: including unmanned clearing vehicle and rapid rescue pallet, used to perform physical operations such as obstacle clearing, vehicle relocation and opening of life passage in tunnels; (c) Unmanned reconnaissance unit: It includes a group of drones equipped with communication relays, gas sensors and thermal imagers, as well as explosion-proof robotic dogs for close-range reconnaissance, which are used to collect data on the tunnel environment, hazard sources and living beings. (d) Communication network unit: It includes satellite or 5G backbone communication equipment and communication relay nodes carried by drones, which are used to establish and maintain real-time communication links between the mother vehicle command unit, the sub-vehicle operation unit, the unmanned reconnaissance unit and the rear command center.

[0005] The mother vehicle command unit is used to receive and fuse multi-source data. The multi-source data received at least includes data such as laser point clouds, video streams, and prior maps. The mother vehicle command unit will construct and update a three-dimensional digital twin tunnel model in real time; and based on the model, conduct AI task planning and simulation deduction, and link the traffic signal system to enable the ambulance to quickly reach the rescue site.

[0006] The functions of the communication network unit include maintaining contact with the rear command center through satellite or 5G network backbone links; using relay nodes carried by drones to expand communication coverage in the tunnel and ensuring the transmission of commands and data between the mother vehicle, the sub-vehicle, and the reconnaissance unit.

[0007] Preferably, the mother vehicle command unit is further integrated with a V2X communication module, which is used to link with the urban traffic signal system to generate a green wave zone for the mother vehicle to reach the scene. When the task planning module of the mother vehicle plans the driving route to the scene, the road condition information in the V2X module is a key input parameter. The system will preferentially select the path with higher coverage rate and easier to achieve green wave passing analyzed by the V2X module, so as to ensure the optimal speed before departure.

[0008] Secondly, during the process of the mother vehicle rushing to the scene, the real-time position, speed of the mother vehicle, and the signal status of the front intersection obtained through V2X can all be dynamically displayed on the digital twin large screen of the rear command center. The commander can clearly see the traveling trajectory and estimated arrival time of the mother vehicle, and make accurate judgments when deploying subsequent rescues.

[0009] Preferably, the three-dimensional digital modeling module in the mother vehicle command unit includes the following steps: (1) Dynamic voxel grid generation: Divide the registered three-dimensional space into voxel grids, and assign multiple dynamic attribute fields to each voxel. The dynamic attribute fields at least include the passability probability calculated based on the object type and obstacle height, and the risk level calculated by integrating environmental sensor data. (2) Model real-time update: According to the newly received reconnaissance data, update the voxel attributes of the changed area with higher weight, and perform time decay processing on the attribute values of historical data, so that the digital twin model synchronously maps the dynamic changes of the physical world.

[0010] The 3D digital modeling module elevates a static spatial model into a computable virtual environment that can dynamically evolve in sync with the physical world. It can also be understood as a closed loop of perception-modeling-updating formed through the complementary steps of dynamic voxel mesh generation and real-time model updates. This transforms the digital twin model from a static map into a new system capable of dynamically reflecting and even predicting changes in the physical world, thereby supporting subsequent AI path planning and simulation.

[0011] The structure of the unmanned obstacle clearing vehicle in the vehicle operation unit specifically includes: (a) Composite walking module: It consists of an omnidirectional moving chassis and liftable auxiliary tracks integrated on both sides or front and rear of the vehicle body; the omnidirectional moving chassis can flexibly turn and laterally in narrow spaces and quickly and accurately position itself; the liftable auxiliary tracks provide stable support and excellent obstacle crossing ability when facing complex terrains such as ruins and slopes.

[0012] (b) Modular operation tool module: a hydraulically driven quick-connect interface set at the front end or the end of the boom of the unmanned recovery vehicle; by quickly changing different tools, a single recovery vehicle can handle a variety of operation tasks such as pushing, demolition, and towing, which greatly improves the operation coverage and task response efficiency of the equipment.

[0013] (c) Explosion-proof sealed vehicle body: It adopts a sealed design to prevent leaked flammable gases or dust from entering the vehicle body; it can avoid causing secondary explosions.

[0014] (d) Integrated active protection module: including vehicle body temperature sensor, automatic fire extinguishing device and physical crash barrier, used to deal with dangerous environment; can actively deal with extreme situations such as open flame and high temperature, and ensure the vehicle's continuous combat capability in dangerous environment.

[0015] (e) External power output interface: including a high-power waterproof aviation connector located on the side or rear of the unmanned recovery vehicle for use with external rescue equipment. This enables the unmanned recovery vehicle to power other rescue equipment, upgrading it from a single operating unit into a small mobile energy station and enhancing the continuous operational capability of the entire rescue team.

[0016] The composite mobility module ensures the unmanned vehicle can reach work locations that are difficult for traditional equipment to access; modular work tools allow the unmanned vehicle to perform the most suitable work tasks after arrival; explosion-proof and active protection modules provide safety permits and survival guarantees for the above-mentioned mobility and operation processes in hazardous environments; and the external power interface further extends and shares this capability with the entire rescue system. These four components are deeply coupled, together elevating the unmanned vehicle from a single-function robot into a mobile, flexible, comprehensive, safe, and reliable intelligent mobile work platform that can support teammates, effectively solving the core challenge of efficient and safe obstacle removal operations in congested tunnels.

[0017] Preferably, the quick rescue pallet of the sub-vehicle operation unit is equipped with a laser-guided docking mechanism, a lifting mechanism, and a self-driving system, which are used to automatically drive to the bottom of the target vehicle, lift the vehicle wheels off the ground, and then carry it.

[0018] The laser-guided docking mechanism, through enhanced vision and millimeter-level precision docking, makes the entire process fast and reliable, laying the foundation for subsequent operations. The lifting mechanism achieves chassis lifting, four-point synchronous, and stable lifting without damage, allowing rescue personnel to enter the vehicle for assistance, solving the problem of vehicle overturning or damage that often occurs with traditional cranes or forklifts. The self-drive module transforms obstacle clearing operations from traditional pushing, pulling, and dragging actions into efficient moving, transporting, and placing actions, avoiding potential scratches and collisions to the accident vehicle and road surface during the pushing process, making the operation more civilized, controllable, and efficient.

[0019] In other words, the rapid rescue pallet utilizes laser-guided precise sensing, a lifting mechanism for stable execution, and a self-driving module for intelligent movement. The deep collaboration of these three modules transforms a passive tool into an active intelligent agent. It works closely with the main vehicle command unit and digital twin model to form a closed loop of perception-decision-execution, improving the efficiency and safety of tunnel congestion rescue.

[0020] Preferably, the unmanned reconnaissance unit's drone swarm is equipped with anti-collision ball cages and optical flow positioning systems for stable flight in complex and narrow tunnels.

[0021] The drone swarm has the ability to fly in contact, allowing it to fly unimpeded into the gaps between vehicles, close to ceilings or walls to conduct reconnaissance and obtain the most critical and detailed on-site data. This data can help the mother vehicle build a high-precision digital twin model.

[0022] The optical flow positioning module enables drones to hover stably even inside tunnels, providing a stable platform for LiDAR scanning and high-definition photography, ensuring the quality of the collected data. The short-term, high-precision relative velocity and displacement information it provides can complement LiDAR real-time positioning and SLAM (Simultaneous Localization and Mapping) to form a stable and reliable integrated navigation system within tunnels.

[0023] Preferably, the explosion-proof robot dog of the unmanned reconnaissance unit is equipped with a voice interaction module for brief communication with trapped personnel; and is equipped with sensors for detecting fuel leaks under the vehicle and the vehicle's structural condition.

[0024] The voice interaction module enables two-way voice communication with trapped individuals who cannot be directly reached, facilitating information gathering and psychological reassurance. The acquired information, such as the location and status of the trapped individuals, is then input into the digital twin model as the highest priority, dynamically optimizing rescue priorities and route planning.

[0025] Under-vehicle reconnaissance sensors actively detect secondary disaster risks that are difficult to detect with the naked eye, such as fuel leaks and vehicle structural instability in the space under the vehicle. The detected risks are then labeled and quantified in real time into a comprehensive hazard level in a digital twin model. The resulting comprehensive hazard level provides key data support for AI decision-making, enabling the entire system to transform from traditional risk perception to risk prediction.

[0026] Furthermore, the explosion-proof robot dog is not merely a standalone reconnaissance tool; it deeply integrates voice interaction and sophisticated sensing capabilities, intelligently connecting the physical scene with a digital twin model. It simultaneously digitizes and visualizes the status of trapped personnel and the risks of the concealed environment, injecting this data into the decision-making loop of the entire rescue system, greatly enhancing the safety, accuracy, and humanization of rescue operations.

[0027] Preferably, this dedicated rescue vehicle system also relates to a method of using a dedicated rescue vehicle system for performing rescue missions in congested tunnels, the specific steps of which include: (1) Alarm response and initial deployment: After receiving the alarm, the command center quickly activates the system. The main vehicle command unit calls up the tunnel data, plans the optimal route and links the traffic signal system. The main vehicle is dispatched to the predetermined stopping point outside the tunnel entrance. (2) On-site reconnaissance and situation generation: The mother vehicle establishes a forward command post; the communication network unit establishes a backbone link and relay nodes; the unmanned reconnaissance unit releases a swarm of unmanned aerial vehicles and explosion-proof robot dogs to conduct three-dimensional reconnaissance; the mother vehicle command unit integrates reconnaissance data, generates a three-dimensional digital twin model, and performs task planning and allocation based on the model; (3) Unmanned operation and passage opening: The sub-vehicle operation unit enters the site according to the instructions, clears obstacles and opens up life passages by pushing, pulling, grabbing and carrying; at the same time, the unmanned reconnaissance unit cooperates in environmental control and continuous monitoring. (4) Personnel rescue and mission completion: After the life channel is opened, rescuers enter the core area to carry out rescue; the sub-vehicle operation unit provides auxiliary support; after all trapped personnel are rescued, each unit works together to complete the site cleanup, and automatically returns to the mother car after the cleanup is completed. The mother car command unit generates a rescue report during the withdrawal process.

[0028] Preferably, in the on-site reconnaissance and situation generation step, after constructing a three-dimensional digital twin model, an AI collaborative decision-making and simulation algorithm is used to assist in planning the optimal lifeline route and assigning specific moving or clearing targets to the vehicle operation unit. This AI collaborative decision-making and simulation algorithm includes: (a) Decision verification module with simulation and inference first: The core algorithm used is the coupling of Monte Carlo Tree Search (MCTS) and embedded physical simulation engine.

[0029] (b) Intelligent planning module for multi-objective weighted optimization: The algorithm used is the A* algorithm. The cost function is: in, This represents the cost incurred in getting from the rescue entrance to the current node n; This represents an estimate of the remaining cost from the current node n to the target point. (c) Multi-agent collaborative module based on auction mechanism: The core algorithm used is the distributed auction algorithm.

[0030] The A* algorithm used in the intelligent planning module of multi-objective weighted optimization is to find the optimal solution for multiple objectives of safety and efficiency in a three-dimensional twin digital model, and to plan the best life passage based on this problem. The life passage planned in this way is no longer the shortest path in geometry, but the optimal passage obtained by comprehensively balancing the cost of clearing operations, environmental safety risks and the urgency of rescue.

[0031] The auction-based multi-agent collaborative module applies auction algorithms to vehicle clearing units for real-time, dynamic task allocation at disaster sites where communication may be unstable. Auction algorithms, being decentralized decision-making mechanisms, endow the system with strong robustness and adaptability, enabling it to handle complex situations such as the dynamic emergence of tasks at the rescue site and sudden failures of individual units.

[0032] The three major algorithm modules—simulation-based decision verification, multi-objective weighted optimization intelligent planning, and auction-based multi-agent collaboration—together with the 3D digital twin model in the 3D digital twin modeling module, form a complete intelligent closed loop from perception to decision-making, verification, and finally action. Specifically: perception involves data from the unmanned reconnaissance unit driving real-time updates to the digital twin model. Decision-making and verification employ the A* algorithm for global strategic lifeline planning on the dynamic model, and use Monte Carlo Tree Search (MCTS) and physical simulation to tactically rehearse and verify the specific operations required at key nodes of the lifeline on the model. Action involves the auction algorithm rationally allocating verified tasks to the most suitable sub-vehicle units. When the sub-vehicle units execute their tasks, the data on their operational effects is perceived by the reconnaissance unit and fed back to the digital twin model, initiating the next optimization cycle.

[0033] Preferably, during unmanned operation and passage opening, the operator inside the mother car performs precise remote control operation of the sub-car's working unit through a first-person perspective and a three-dimensional situation map.

[0034] The first-person perspective provides the operator with a high-definition, low-latency real-time view of the front of the vehicle, enabling the operator to use the vehicle to perform complex operations such as grabbing, dismantling, and docking in an immersive way, and to intuitively check the equipment status and on-site details, thereby making up for the shortcomings of pure AI models in understanding complex scenarios.

[0035] The 3D situation map visualizes the digital twin model, the real-time location of all units, the AI-planned path, and the task objectives, providing a global situational awareness and a collaborative task view. Operators can always grasp their position and progress in the overall task, understand the AI's decision-making intent, and promptly intervene and replan at the global level in case of emergencies.

[0036] The combined use of a first-person perspective and a 3D situation map allows operators to quickly locate problems under the global guidance of the 3D situation map, and then switch to the first-person perspective for precise operations. Simultaneously, augmented reality guidance information from the digital twin model can be overlaid in the first-person perspective, and the results of the operations are fed back to the 3D model in real time for updates. This combination of first-person perspective and 3D situation map integrates human cognitive judgment with precise machine execution, ensuring efficient progress of the overall task while enabling highly efficient and precise operations at key and complex points, thereby improving the overall system's operational accuracy and reliability.

[0037] (III) Beneficial Effects (1) By deeply integrating the dynamically updated 3D digital twin model with AI collaborative decision-making and simulation algorithms, the rescue site is transformed into a computable environment using a dynamic voxel mesh model. The real-time updated accessibility probability and comprehensive risk level in the model provide an accurate and vivid situational map for AI decision-making. This enables the AI ​​algorithm to form an intelligent closed loop integrating perception, decision-making, verification, and action, rather than isolated calculations based on static maps. The simulation module pre-simulates the consequences of operations in the digital twin model before action, transforming the original experience-based decision-making into simulation-verified decision-making, thereby effectively avoiding secondary risks caused by blind actions. In addition, the collaboration between the 3D digital twin model and AI algorithms allows rescue command to predictively plan the globally optimal lifeline and dynamically and safely allocate tasks, transforming the rescue mode from relying on human experience to data-driven, human-machine collaborative intelligent operations, greatly improving rescue efficiency, accuracy, and safety. Attached Figure Description

[0038] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a sequence diagram of the overall workflow of a specialized rescue vehicle system used to perform rescue missions in congested tunnels; Figure 2 This is a flowchart illustrating the 3D digital twin modeling process in a specialized rescue vehicle system used for performing rescue missions in congested tunnels. Figure 3 This is a structural diagram of an unmanned clearing vehicle in a specialized rescue vehicle system used for performing rescue missions in congested tunnels; Figure 4 This is a schematic diagram of the working principle of a rapid rescue pallet in a specialized rescue vehicle system used for rescue missions in congested tunnels. Figure 5 This is a schematic diagram of the closed-loop AI collaboration and simulation inference algorithm in a dedicated rescue vehicle system for performing rescue missions in congested tunnels; Figure 6 This is a schematic diagram of the human-machine collaborative control interface of a specialized rescue vehicle system used for carrying out rescue missions in congested tunnels.

[0039] Attached diagram descriptions: 11-Omnidirectional mobile chassis, 12-Liftable auxiliary tracks, 13-Hydraulic drive quick-change interface, 14-Explosion-proof sealed body, 15-Integrated active protection, 16-External power output interface. Detailed Implementation

[0040] The following will refer to the appendix in the examples of this invention. Figure 1 - Appendix Figure 6 The technical solutions in the embodiments of the present invention are clearly and completely described. 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.

[0041] A dedicated rescue vehicle system for performing rescue missions in congested tunnels includes: (a) Mother vehicle command unit: As the main command center, it is used for overall command, route planning, situation generation and task allocation. It is equipped with an onboard computer with an operation data fusion module, a three-dimensional digital twin modeling module and a task planning module. (b) Sub-vehicle operation unit: including unmanned clearing vehicle and rapid rescue pallet, used to perform physical operations such as obstacle clearing, vehicle relocation and opening of life passage in tunnels; (c) Unmanned reconnaissance unit: including a swarm of unmanned aerial vehicles equipped with communication relays, gas sensors and thermal imagers, as well as explosion-proof robot dogs for close-range reconnaissance, used to collect data on the tunnel environment, hazards and living organisms; (d) Communication network unit: including satellite or 5G communication equipment and communication relay nodes carried by UAVs, used to establish and maintain real-time communication links between the mother vehicle command unit, the sub-vehicle operation unit, the unmanned reconnaissance unit and the rear command center.

[0042] To further explain, the mother vehicle command unit is the command center of the entire dedicated rescue vehicle system. It can be a specially modified ambulance serving as the mobile command center for the entire rescue operation. The vehicle is equipped with a high-performance onboard computer, which operates three main modules, specifically including: (1) Data fusion module: used to receive and register real-time laser point cloud, video stream and prior tunnel map data from unmanned reconnaissance unit. Among them, objects in the video stream are identified and semantically labeled by computer vision model. (2) Three-dimensional digital twin modeling module: Based on the data fused in the data fusion module, a tunnel space model that is exactly the same as the rescue tunnel environment is constructed in real time in the vehicle computer. This tunnel is not only three-dimensional, but more importantly, every voxel in the tunnel model, that is, every point, has dynamic attributes. These dynamic attributes are used to reflect the changes in the real and reliable real world.

[0043] (3) Task planning module: Based on the dynamic digital model built by the three-dimensional digital twin modeling module, artificial intelligence algorithms are used to plan rescue routes and assign tasks. After the planned routes and assigned tasks are generated, they are first simulated and deduced in the dynamic digital model. Only after the simulation verifies that the plan is safe and feasible can the next step be carried out.

[0044] To further explain, the unmanned reconnaissance unit is mainly used for reconnaissance, including the environment inside the tunnel and the situation of trapped personnel, and for capturing relevant data in real time. The tools used for reconnaissance mainly include drone swarms and bomb-proof robotic dogs. Specifically: (a) The main reconnaissance range of the drone swarm is the air. The drones quickly fly into the tunnel and acquire a wide-ranging global view from the air through laser scanning and cameras. At the same time, they also carry communication relay equipment to connect with the main vehicle command unit to ensure smooth communication inside and outside the tunnel.

[0045] (b) The main range of the explosion-proof robot dog is the ground. The robot dog can crawl under the vehicle and enter a narrow space to conduct a comprehensive inspection of the vehicle. The inspection includes, but is not limited to, damage and fuel leaks. It also has a voice interaction function to communicate with trapped personnel and provide emotional support while obtaining patient information.

[0046] Furthermore, the sub-vehicle operation unit is mainly used for obstacle clearing and rescue, that is, opening up rescue channels and providing timely assistance. Its main working structure includes an unmanned obstacle clearing vehicle and a rapid rescue pallet. Specifically: The unmanned obstacle clearing vehicle is a multi-functional, driverless, explosion-proof obstacle clearing robot used to replace rescue personnel in dangerous environments to perform rescue tasks. This unmanned obstacle clearing vehicle can quickly switch between tools to complete various obstacle clearing actions such as pushing, pulling, jacking, and clamping, and carry out comprehensive obstacle clearing work in tunnels.

[0047] The quick rescue pallet is similar to a rescue stretcher. The pallet can be driven under the stuck vehicle to smoothly lift and move the vehicle away, thereby quickly opening up a rescue channel.

[0048] Furthermore, the main objective of the communication network unit is to enable the communication required inside and outside the tunnel. Specifically, it consists of vehicle-mounted satellite or 5G equipment and relay equipment carried by drones, thereby forming a seamless communication network covering both inside and outside the tunnel. This network is used to maintain constant communication between the mother vehicle command unit, the unmanned reconnaissance unit, and the sub-vehicle operation unit, and to transmit instructions and data in real time.

[0049] The master vehicle command unit is further integrated with a V2X communication module, which is used to interact with the urban traffic signal system to generate a green waveband for the master vehicle to reach the scene. The integrated V2X communication module not only adds communication functions to the master vehicle, but also deeply integrates the master vehicle into the urban intelligent transportation network, transforming from the traditional ambulance's request for right-of-way to a better road condition where the road actively cooperates with right-of-way. Specifically: The V2X communication module installed on the master vehicle adopts mature DSRC or C-V2X communication technologies. Specifically, chipsets such as Autotalks' CRATON2 or similar commercial modules can be used to ensure compatibility with existing road infrastructure. In addition, this module supports standardized vehicle-road communication protocols such as SAE J2735, which are used to encapsulate and send special vehicle status messages with high priority. It should be noted that the above-mentioned communication technologies, chipsets and communication protocols are all conventional existing technologies, and the relevant technical details are well-known to those skilled in the art, so no further detailed description will be made.

[0050] The steps for the master vehicle to generate and apply for a green waveband using the installed V2X communication module are specifically as follows: (I) Triggering and identity authentication: When the master vehicle receives a rescue mission instruction, the on-vehicle system will automatically activate the V2X communication module and continuously broadcast to the roadside units (RSUs) along the way, indicating its special identity identification code for emergency rescue vehicles. It should be noted that the identity identification code used needs to be certified and authorized by the traffic management department to ensure its legality and authority.

[0051] (II) Route prediction and request sending: The navigation system of the master vehicle will plan the optimal driving route to the tunnel entrance. At this time, the V2X communication module will send the route information to the signal control system at the front intersection in advance, indicating that priority passage is required and requesting the cooperation of the traffic system. The specific messages sent include rescue vehicle information, arrival time, passing time, etc.

[0052] (III) Cooperative response of the signal machine: After receiving the request, the intelligent signal machine at the intersection will incorporate it into the real-time scheduling algorithm and make fine-tuning of the traffic conditions. The adjustment methods can be: Appropriately extend the green light time in the driving direction of the master vehicle.

[0053] End the green light of the oncoming lane in advance, or shorten the passing time in other directions.

[0054] Coordinate among a series of intersections to calculate and generate a continuous green waveband for the master vehicle, enabling the master vehicle to continuously pass through multiple intersections at a nearly uniform speed, minimizing stop-and-wait.

[0055] The three-dimensional digital twin modeling module in the master vehicle command unit includes the following steps: (1) Dynamic voxel mesh generation: The registered three-dimensional space is divided into voxel meshes, and each voxel is assigned multiple dynamic attribute fields. The dynamic attribute fields include at least the passability probability calculated based on the object type and obstacle height, and the comprehensive hazard level calculated based on the comprehensive environmental sensor data. (2) Real-time model update: Based on the newly received reconnaissance data, the voxel attributes of the changed area are updated with higher weights, and the attribute values ​​of historical data are processed by time decay, so that the digital twin model can synchronously map the dynamic changes of the physical world.

[0056] Furthermore, the 3D digital twin modeling module constructs a computable and simulable virtual environment that includes physical rules and real-time situational awareness, rather than a simple 3D mirror of the real world. The various steps of the module are interconnected: Data fusion in the data fusion module integrates disparate data from different sources into a unified spatial model with semantic information. This process gradually progresses from aligning raw data to generating a high-level semantic model. In other words, preliminary fusion is required in the data fusion module of the main vehicle command unit. Preliminary fusion involves set registration and data association, primarily the alignment and bundling of underlying data. Specific steps include: a) Spatiotemporal alignment: First, the data fusion module uses an iterative nearest point algorithm to coarsely and finely register the received laser point cloud, video stream, and other data with the prior tunnel model, synchronizing these data in the time and space coordinate systems to ensure that all data are registered in a unified three-dimensional coordinate system.

[0057] b) Data Binding: After completing spatiotemporal alignment, the YOLOv5 computer vision model is used to identify 2D object bounding boxes from the video stream. These 2D object bounding boxes are then associated with corresponding point cloud clusters in the 3D laser point cloud using camera calibration parameters. After the association is completed, the system can understand the object represented by each cluster of 3D points, and thus complete the association between geometric data and preliminary semantics.

[0058] c) Output results: After completing data binding and association, the system finally outputs a preliminary fused 3D point cloud dataset with semantic labels.

[0059] Furthermore, the iterative nearest point algorithm used in the data fusion module is a classic and well-known point cloud registration algorithm in the fields of 3D computer vision and robotics. It iteratively searches for the closest point pairs between two point sets and minimizes the distance between these pairs by calculating rotation and translation matrices, thus completing the registration. This is a conventional existing technology and will not be described in detail. The prior tunnel model is a digital model of the tunnel's foundation acquired and stored before the rescue operation. It includes at least the 3D geometric structure converted from design drawings or BIM models and key fixed facilities within the tunnel, including the location and attribute information of fire hydrants and ventilation openings. Prior tunnel models are a very mature pre-process in fields such as smart cities, digital twins, and BIM, and will not be described in detail here.

[0060] After data fusion is completed, the process moves to the 3D digital twin modeling module, where information extraction and model structuring are performed step-by-step. Using the previously fused data, a model suitable for computation and reasoning is constructed. Specific steps include: (1) Dynamic voxel mesh generation: First, the system divides the entire tunnel space into countless tiny cubes of uniform size. These cubes are voxels. This process can also be called voxel mesh generation, which is to convert continuous and irregular point cloud data into regular, discrete data structures that are easy for computers to process.

[0061] After completing the voxel mesh division, the system will traverse each voxel and calculate and assign a series of dynamic attribute fields to each voxel based on the point cloud semantic labels and sensor readings contained within it. The assigned dynamic attribute fields include at least the passability probability P and the overall hazard level R.

[0062] The passability probability P is calculated based on the object type and obstacle height. It quantifies the likelihood that rescue equipment can safely and successfully pass through the location of the voxel, and is a value between 0 (completely impassable) and 1 (completely passable). The specific algorithm includes: 1) The system assigns an initial access value to the point cloud within the voxel based on its semantic labels. The specific allocation logic is that the more fixed and difficult to move the identified object, the higher its initial pass value. The lower the value, the better. The allocation of initial passage values ​​requires a pre-defined mapping relationship within the system. Based on the allocation logic, this mapping relationship is pre-defined as follows: Areas identified as unobstructed, such as air or road surface, are given a higher initial passability value. Areas identified as movable obstacles are assigned a moderate initial passage value; Areas identified as fixed structures such as walls are assigned the lowest initial passage value.

[0063] After this, the initial passage value will be further modified based on the physical geometry of the obstacle. Specifically, the height H can be used for modification to define a height correction function based on the device's passage capability. This means comparing the relative height H of the obstacle with the passage value of the rescue equipment. For obstacles below the equipment's safe passage threshold, the system will increase their passage value. The closer the value is to 1, the less obstructed the path. For obstacles far exceeding the device's obstacle-crossing capability, the system will lower the passability value. The closer to 0, the more complete the blockage. Therefore, the height correction function... It is a decreasing function from 1 to 0.

[0064] After the height correction is completed, dynamic evolution is required. That is, when the vehicle operation unit clears an obstacle, the reconnaissance data returned by the unmanned reconnaissance unit will show that the area has become passable. Based on this information, the system updates the voxel passability value of the area to a higher value, thereby reflecting the changes in the environment in real time.

[0065] The final calculation of the passability probability P is to use the initial passability value. With height correction value A comprehensive calculation is performed to obtain the final result. As a preferred implementation scheme, a multiplication model can be used for calculation, with the following formula: The multiplicative model can effectively express the joint constraint effect of semantic attributes and physical height on traversability. It should be noted that those skilled in the art will understand that other function combinations can also be used to achieve the combined calculation of the two, but all must adhere to the core idea of ​​the joint constraint described above.

[0066] Furthermore, assigning dynamic attributes to voxels means that these voxels are not completely static, but rather dynamically updated in real time as new reconnaissance data is input. The overall risk level R is calculated based on comprehensive environmental sensor data. It is used to comprehensively evaluate the potential threat level posed by a specific spatial voxel to rescue operations. Specifically, it is calculated using a multi-factor weighted fusion model, and its calculation formula is as follows: in, This represents the weight coefficient of the i-th risk factor, used to reflect the relative importance of different risk factors in the rescue scenario; This represents the normalized hazard value of the i-th risk factor, whose value is limited to between 0 and 1.

[0067] The overall hazard R-value calculated using a multi-factor weighted fusion model is a quantified score; a higher score indicates a higher degree of hazard. The multi-factors include at least: temperature, toxic gas concentration, presence of open flame, and structural stability. The system assigns a weighting coefficient to each factor. The weighted sum of all factors forms the basis of the overall risk R. Specifically: The contribution of the temperature factor is calculated based on the current temperature's position relative to the safe and dangerous thresholds. The higher the temperature, and the closer it is to or exceeds the dangerous threshold, the greater its contribution, and its weighting coefficient. The larger.

[0068] The contribution value of toxic gas factors is calculated based on the ratio of gas concentration to its hazardous concentration threshold. The higher the concentration, the greater the contribution value, and its weighting coefficient... The larger.

[0069] The open flame factor directly determines whether an open flame exists in a voxel region using computer vision and thermal imaging data, and assigns it a higher contribution value, i.e., a larger weighting coefficient. .

[0070] The contribution value of the structural stability factor is calculated based on a visual analysis of the object's deformation and degree of damage. The more severe the deformation and the more obvious the damage, the higher the contribution value, and the higher its weighting coefficient. The larger.

[0071] Finally, the overall risk level R calculated using the multi-factor weighted fusion model is directly stored in the voxel dynamic attributes for use by the AI ​​decision-making module during analysis. In other words, the intelligent planning module with multi-objective weighted optimization will prioritize paths with lower overall risk levels R when planning routes; and will assign work units with appropriate protective capabilities to tasks in high-risk areas when allocating tasks.

[0072] Furthermore, the situation inside the tunnel is constantly changing. Therefore, after the voxel mesh is generated, the model will be updated in real time to achieve synchronous evolution. This ensures the timeliness and accuracy of the rescue. The model update includes incremental updates and intelligent forgetting. Specifically: Incremental updates involve the system automatically locating and updating the voxel attributes of changed areas when new reconnaissance data arrives. The new reconnaissance data includes at least the new point cloud dataset transmitted back from the second round of UAV scans.

[0073] Intelligent forgetting uses a time decay processing algorithm to apply a gradually diminishing weight to historical data, ensuring that the model always leans towards the latest environmental situation. In other words, time decay processing makes the 3D digital twin model more reliant on the latest reconnaissance data, while smoothly diminishing the influence of old data, thus achieving the best balance between rapid response and maintaining stability.

[0074] The algorithm used for time decay processing is the exponentially weighted moving average method, and the calculation formula is as follows: Among them, the calculated This represents the voxel attribute value calculated at time t, which is the latest value. This value has been smoothed by the algorithm and is about to be stored in the digital twin model. This represents the historical voxel attribute values ​​stored in the model at time t-1, which is the previous time step. This represents the voxel attribute observation value directly calculated based on the newly received reconnaissance data at time t. It is called the decay factor, also known as the forgetting coefficient, and is an adjustable parameter between 0 and 1. The larger the value, the stronger the model's memory of historical data, and the slower its response to new changes; conversely, the smaller the value, the slower the model's response to new changes. The smaller the value, the more sensitive the model is to the latest data, but the more susceptible it is to transient noise.

[0075] This formula is equivalent to: in, This represents the voxel attribute observations directly calculated from newly received reconnaissance data. That is The fusion weight.

[0076] To address the dual requirements of rapid response and system stability in tunnel rescue, and after extensive simulation testing, the attenuation factor in this invention... A preferred value range is 0.3-0.7. For an attribute requiring a rapid response, such as the overall risk level R, a lower value can be used. Value, that is =0.3; For a property that needs to remain stable, such as structural stability, a higher value can be used. Value, that is =0.6.

[0077] The exponentially weighted average method is precisely and deeply coupled with other technical modules, forming a whole with a closed-loop capability of perception-decision-verification-action. Using the exponentially weighted average method for real-time model updates maximizes the synergy of the entire system, specifically including: a) Coupling with Dynamic Voxel Mesh: A dynamic voxel mesh is a structured, addressable data unit that can be used in conjunction with the exponentially weighted average method for efficient iterative computation. Each dynamic attribute field of each voxel can be used as an object of the exponentially weighted average method. Furthermore, the regularized and discretized characteristics of the dynamic voxel mesh enable the system to perform exponentially weighted average calculations on massive voxel attributes in parallel and efficiently.

[0078] Without coupling with a dynamic voxel grid, the exponential weighted average method requires processing irregular and massive amounts of raw point clouds, which is computationally complex and resource-intensive. However, the coupling of the exponential weighted average method with a dynamic voxel grid makes the dynamic update of the entire tunnel space clear and computable.

[0079] b) Coupling with the Data Fusion Module: The data fusion module fuses multi-source data to provide high-quality, semantically meaningful input data for the exponentially weighted average method, thereby improving the accuracy of model updates. Specifically, the data fusion module outputs registered and semantically labeled 3D point cloud datasets, which directly trigger the exponentially weighted average method for updates. When the system receives fused data from the unmanned reconnaissance unit, it promptly locates the voxel regions where changes have occurred and directly calculates the voxel attribute observations based on the new semantic labels and the newly received reconnaissance data calculated from sensor readings. Then, the exponentially weighted average method is called to calculate the voxel attribute observations. Compared with historical values ​​stored in the model Perform weighted fusion calculations to obtain the latest voxel attribute values. .

[0080] Furthermore, the exponentially weighted average method relies on accurate data fusion. In other words, only precisely registered and semantically labeled data obtained through data fusion can ensure the accuracy of the spatial location and object attributes updated by the exponentially weighted average model. Simultaneously, the smoothing capability of the exponentially weighted average method compensates for potential transient misidentifications or data fluctuations during data fusion, enabling the system to maintain stability, continue operating, and complete its core tasks even in the face of internal faults and external interference.

[0081] The unmanned vehicle structure of the sub-vehicle operation unit specifically includes: (a) Composite walking module: consisting of an omnidirectional moving chassis and liftable auxiliary tracks integrated on both sides or front and rear of the vehicle body; (b) Modular operation tool module: a hydraulically driven quick-connect interface set at the front end or the end of the boom of the unmanned vehicle; (c) Explosion-proof sealed vehicle body: It adopts a sealed design to prevent leaked flammable gases or dust from entering the vehicle body; (d) Integrated active protection module: including vehicle body temperature sensor, automatic fire extinguishing device and physical crash barrier, used to deal with dangerous environments; (e) External power output interface: including a high-power waterproof aviation plug located on the side or rear of the unmanned recovery vehicle for use with external rescue equipment.

[0082] Unmanned obstacle clearing vehicles are intelligent operational robots specifically designed to perform tasks in congested tunnels where space is limited, the environment is hazardous, and the working conditions are complex. Their specific structure includes: The composite mobility module consists of an omnidirectional mobile chassis and liftable auxiliary tracks. The omnidirectional mobile chassis uses Mecanum wheels or omnidirectional wheels as the main movement mechanism, enabling flexible movement in the longitudinal, lateral, diagonal, and on-the-spot rotation directions. This means that the unmanned clearing vehicle can laterally enter the work position and operate in confined tunnel spaces, improving the mobility and efficiency of the sub-vehicle work unit and buying valuable time to quickly open up life-saving passages.

[0083] The liftable auxiliary tracks utilize an electrically or hydraulically lifting track mechanism, integrated on the sides or front and rear of the vehicle body. When the unmanned recovery vehicle is traveling on a flat road, the liftable auxiliary tracks rise, allowing the omnidirectional wheels to provide efficient and flexible movement. When encountering large obstacles, piles of rubble, or inclined slopes caused by accidents, the liftable auxiliary tracks lower to the ground, giving the unmanned recovery vehicle a larger ground contact area and stronger obstacle-crossing ability. This enables it to stably traverse obstacles that traditional wheeled chassis cannot pass through, ensuring the recovery vehicle's passability and operational stability in complex road conditions.

[0084] The hydraulically driven quick-connect interface in the modular work tool module is located at the front end of the unmanned recovery vehicle or the end of the boom. Specifically, it can be a hydraulic quick-connect coupling conforming to ISO 14567 or similar industry standards. This interface allows for one-button quick connection and disconnection with various work tools via the vehicle's hydraulic system. Multiple work tools can be integrated into a toolset, which includes at least the following replaceable tools: A pusher used to move scattered obstacles or clear debris; Hydraulic shears or spreaders used to break apart deformed vehicles and create rescue space; Wire rope reels and hooks used for towing heavy accident vehicles.

[0085] The quick-connect interface allows unmanned obstacle clearing vehicles to change their tools in a timely manner according to the needs of the site, enabling them to perform corresponding obstacle clearing tasks and achieve multiple uses for one vehicle. This greatly improves the operational coverage of a single device and the cost-effectiveness and responsiveness of the entire system.

[0086] The explosion-proof sealed vehicle body features a sealed design, with key electronic components, including but not limited to clickers and remote controls, housed within explosion-proof enclosures that meet ATEX explosion-proof certification standards. Furthermore, all cable interfaces are sealed. This explosion-proof sealed body design prevents potentially leaked flammable gases or dust from entering the vehicle body, avoiding secondary explosions caused by electrical sparks from electrical equipment, thus ensuring the safety of the equipment and the entire rescue site.

[0087] The integrated active protection module can be viewed as a standalone system, specifically including: a) Vehicle body temperature sensor and automatic fire extinguishing device: When an abnormal surface temperature of the vehicle body is detected or an open flame is encountered, the device will automatically trigger the spraying of ultra-fine dry powder or heptafluoropropane fire extinguishing agent around the bottom of the vehicle and key components to achieve self-extinguishing of the initial fire.

[0088] b) Physical anti-collision guardrails: Alloy anti-collision bars are installed on key parts of the vehicle body, including LiDAR, cameras and other parts, to prevent collisions with obstacles during operation that could damage the core sensors.

[0089] The integrated active protection module design enables the unmanned vehicle to survive in short periods and operate continuously in extreme environments such as high temperatures and open flames.

[0090] The external power output interface is primarily located on the side or rear of the unmanned recovery vehicle, employing a high-power, waterproof aviation connector capable of outputting 48V DC or 220V AC power. This external power output interface is designed to provide on-site power or charging for other rescue equipment, including but not limited to hydraulic rescue tools, lighting systems, life detectors, or another robot with depleted battery. This means that within the system, multiple intelligent agents can achieve energy synergy, upgrading the unmanned recovery vehicle from an independent operating unit into a small mobile energy station, enhancing the continuous operational capability of the entire rescue team.

[0091] The rapid rescue pallet of the sub-vehicle operation unit is equipped with a laser-guided docking mechanism, a lifting mechanism, and a self-driving module. It automatically travels to the bottom of the target vehicle, lifts the vehicle's wheels off the ground, and then transports it. The rapid rescue pallet is a highly autonomous vehicle handling robot capable of quickly and without secondary damage removing critical vehicles blocking emergency access routes in congested tunnels. Specifically: The laser-guided docking mechanism consists of a laser emitter array and a visual recognition system. Its specific workflow is as follows: 1) Coarse Positioning: After receiving instructions from the mother vehicle command unit, the rapid rescue pallet travels to the vicinity of the target vehicle via its self-driving system. During the journey, the rapid rescue pallet emits fan-shaped laser beams towards the underside of the vehicle. These laser beams are clearly displayed on the operator's screen or in the 3D situation map of the mother vehicle behind, to assist the operator in making an initial confirmation.

[0092] 2) Precise Alignment: After initial coarse positioning, the vision sensors and cameras at the front of the quick-rescue pallet begin to identify the relative positions of the vehicle's tires. Based on the tire's geometric characteristics and combined with laser ranging data, the system calculates the lateral and longitudinal deviations between the quick-rescue pallet and the tires in real time.

[0093] 3) Autonomous correction: After completing coarse positioning and fine alignment, the self-drive system will make slight lateral movements and angle adjustments based on the deviation data, aligning the V-shaped or grooved guide mechanism on the quick rescue pallet with and embedding it into both sides of the tire.

[0094] The lifting mechanism primarily employs a four-point synchronous lifting mechanism, meaning that an independent lifting unit is installed at each of the four wheel positions on the quick-rescue pallet. The specific workflow is as follows: 1) Positioning and Contact: After the laser-guided docking mechanism precisely moves the rapid rescue pallet to the designated position under the vehicle, the four lifting units corresponding to the four wheel positions move upwards simultaneously, reliably contacting the top support pallet with the vehicle chassis or tires. The lifting units can be electrically or hydraulically driven scissor jacks or augers.

[0095] 2) Synchronous and smooth lifting: The four lifting units are connected to the central controller, which controls their synchronous operation to lift all four wheels of the vehicle off the ground simultaneously and smoothly. This avoids frame twisting and prevents secondary structural damage to the accident vehicle.

[0096] 3) Locking and holding: After being lifted into position, the system uses a mechanical self-locking or hydraulic pressure holding device to keep the vehicle firmly in the raised state, ensuring absolute safety during subsequent transport.

[0097] The self-driving module includes a compact hub motor, a high-energy-density battery pack, and a navigation control system based on Simultaneous Localization and Mapping (SLAM) technology. Each wheel houses a compact hub motor, enabling small-radius turns and even stationary maneuvers, enhancing the flexibility of the rapid rescue pallet in congested environments. The specific workflow is as follows: 1) Path following: After receiving the optimal evacuation path planned by the 3D digital twin modeling module of the mother vehicle, the rapid rescue pallet accurately follows the path in tunnels without GPS signals by fusing data from lidar, inertial measurement unit (IMU) and wheel speed odometer.

[0098] 3) Transport and transfer: After lifting and locking the target vehicle, the self-driving system, according to instructions, transports the vehicle safely and smoothly along the planned path to the emergency parking lane or a pre-set gathering and dispersal area in the tunnel, thereby quickly clearing the main road.

[0099] Simultaneous localization and mapping (SLAM) is a well-known foundational technology in the fields of robotics and autonomous driving. Its core function is to enable mobile devices to simultaneously estimate their own position and build an environmental map in an unknown environment. It is a mature existing technology and therefore will not be described in detail.

[0100] The unmanned reconnaissance unit's drone swarm is equipped with anti-collision ball cages and optical flow positioning modules, enabling it to conduct ultra-low-altitude, close-range reconnaissance operations in tunnels where visibility may be extremely low, space is narrow, and filled with uncertain obstacles. Specifically: A collision protection cage is designed to protect the core components of a drone from damage and allow it to continue flying in the event of an unavoidable collision. The specific structure includes a geometrically spherical or near-spherical frame that completely encloses the drone body. This spherical frame can be made of carbon fiber composite materials or high-strength lightweight alloys, minimizing weight while ensuring sufficient rigidity and strength.

[0101] The optical flow positioning module consists of a downward-looking camera and an inertial measurement unit. Its specific workflow is as follows: 1) Image acquisition: When the drone flies inside the tunnel, the downward-facing camera continuously captures images of the ground directly below at a high frequency, including road surface texture, cracks, and marking lines.

[0102] 2) Change detection: The optical flow algorithm calculates the moving speed and direction of feature points in the image by comparing consecutive frames.

[0103] 3) Data fusion: By combining these pixel-level motion information with the body acceleration and angular velocity data provided by the inertial measurement unit (IMU), and using data fusion algorithms such as Kalman filtering, the precise three-dimensional velocity vector of the UAV relative to the ground is calculated.

[0104] 4) Feedback control: The system utilizes the fused speed information and uses a PID controller to quickly adjust the speed of each motor to generate thrust in the opposite direction, thereby suppressing the speed of the UAV relative to the ground to zero and achieving precise hovering.

[0105] The explosion-proof robotic dog in the unmanned reconnaissance unit is equipped with a voice interaction module for brief communication with trapped personnel; and sensors to detect fuel leaks under the vehicle and the vehicle's structural condition. The explosion-proof robotic dog is designed to fill the gap in observation and reconnaissance capabilities beyond the macroscopic field of view of the drone swarm, specifically including: The voice interaction module includes at least a high-sensitivity noise-canceling microphone array, a waterproof speaker, and a dedicated audio processing unit. In areas inaccessible to rescuers, it establishes contact with those trapped, obtains crucial information, and provides psychological support. The specific workflow is as follows: 1) Sound source localization and noise reduction: After approaching the trapped person, the microphone array filters the background noise in the tunnel and accurately locates the direction of the distress call, guiding the robot dog to move towards the sound source; the background noise includes but is not limited to wind noise, leakage sound, etc.

[0106] 2) Two-way voice communication: After the information of the trapped person is fed back to the main vehicle command unit, the operator in the main vehicle command unit can conduct real-time, two-way voice communication with the trapped person through the explosion-proof robot dog.

[0107] 3) Offline voice command recognition: When there is no communication network connection in the tunnel, the explosion-proof robot dog's built-in offline voice library will start working to recognize preset keywords and respond with pre-recorded voices to soothe the emotions of trapped personnel in a timely manner.

[0108] The location and status information of personnel transmitted back by the explosion-proof robotic dog is directly input into the 3D digital twin model as the highest priority semantic information. The AI ​​decision-making module will dynamically adjust the weight of the rescue route based on this information to ensure that resources are prioritized for those most in need of rescue.

[0109] Under-vehicle reconnaissance sensors are used to actively detect the space under a vehicle and identify potential secondary disaster risks such as fuel leaks and structural instability. Specifically, this includes fuel leak detection and vehicle structural condition assessment.

[0110] Fuel leak detection is achieved by using a photoionization detector or a high-performance semiconductor volatile organic compound sensor. While the explosion-proof robot dog crawls under the vehicle, the sensors continuously sample the air. When the concentration of gasoline or diesel vapor exceeds the safety threshold, it immediately feeds back to the system, marking the vehicle as a high-risk leak source in the three-dimensional digital twin model and significantly increasing the overall hazard level R of surrounding voxels. Simultaneously, it issues the highest-level alarm to the parent vehicle.

[0111] The system uses depth cameras and 3D laser scanners to assess vehicle condition. The explosion-proof robot dog circles the vehicle, performing a 3D scan to reconstruct its precise outline and posture. Based on the scan data transmitted back by the robot dog, the system analyzes point cloud data to automatically identify whether the vehicle body is severely deformed, whether the vehicle is in an unstable state of balance, or whether the jacks are malfunctioning.

[0112] It is important to note that this structural information will be integrated into the digital twin model in real time.

[0113] More specifically, the explosion-proof robot dog itself has passed ATEX or IECEx explosion-proof certifications, ensuring that it will not generate electrical sparks that could cause an explosion when operating in environments filled with flammable gas mixtures. Adopting a quadrupedal bionic design, it possesses the ability to crawl, cross obstacles, and climb stairs, enabling it to reach complex terrains inaccessible to wheeled or tracked equipment. The explosion-proof robot dog is deeply embedded in the entire intelligent rescue system. Through voice interaction, it integrates the status of trapped personnel into a digital twin model. Hidden risks are visualized and digitized through under-vehicle sensors and input into the 3D digital twin model. Simultaneously, through linkage with AI simulation and deduction, reconnaissance data is transformed into predictable action consequences, achieving a leap from risk perception to risk anticipation.

[0114] A method for using a dedicated rescue vehicle system for performing rescue missions in congested tunnels. The specific steps are as follows: (1) Alarm response and initial deployment: After receiving the alarm, the command center quickly activates the system. The main vehicle command unit calls up the tunnel data, plans the optimal route and links the traffic signal system. The main vehicle is dispatched to the predetermined stopping point outside the tunnel entrance. (2) On-site reconnaissance and situation generation: The mother vehicle establishes a forward command post; the communication network unit establishes a backbone link and relay nodes; the unmanned reconnaissance unit releases a swarm of unmanned aerial vehicles and explosion-proof robot dogs to conduct three-dimensional reconnaissance; the mother vehicle command unit integrates reconnaissance data, generates a three-dimensional digital twin model, and performs task planning and allocation based on the model; (3) Unmanned operation and passage opening: The sub-vehicle operation unit enters the site according to the instructions, clears obstacles and opens up life passages by pushing, pulling, grabbing and carrying; at the same time, the unmanned reconnaissance unit cooperates in environmental control and continuous monitoring. (4) Personnel rescue and mission completion: After the life channel is opened, the rescue personnel enter the core area to carry out rescue; the sub-vehicle operation unit provides auxiliary support; after all the trapped personnel are rescued, each unit works together to complete the site cleanup, and automatically returns to the mother car after the cleanup is completed. The mother car command unit generates a rescue report in the process of being in the car.

[0115] Furthermore, the alarm response and initial deployment steps utilize digital and network technologies to minimize non-operational time. This primarily includes system startup and data retrieval, as well as intelligent route planning and traffic coordination. Specifically: System startup and data retrieval refer to the system being started with a single click after the command center receives the alarm. The main vehicle command unit immediately retrieves the prior tunnel model of the target tunnel, including its structure and facility locations, and loads it as the initial digital base map.

[0116] Intelligent route planning and traffic coordination means that the system plans the optimal route for the mother vehicle to reach the scene based on real-time traffic flow data. At this time, the mother vehicle's V2X communication module is activated, continuously broadcasting its rescue vehicle identity and planned route to the city's traffic signal control system. The signal system then generates a green wave for the mother vehicle, ensuring its smooth passage and reaching the safe, predetermined stopping point outside the tunnel entrance as quickly as possible.

[0117] The on-site reconnaissance and situation generation process rapidly transforms an unknown and chaotic disaster scene into a structured, computable digital twin model. The specific process includes: 1) Establishment of the forward command post and communication coverage: After the mother vehicle stops, it quickly deploys as a forward command post. The communication network unit immediately begins operation, maintaining contact with the rear via satellite or 5G network backbone links, and releasing a swarm of drones carrying communication relay nodes to fly deep into the tunnel to establish a seamless communication network covering the entire accident area.

[0118] 2) Three-dimensional reconnaissance and data fusion: The drone swarm utilizes its anti-collision cages and optical flow positioning system to fly stably within the tunnel, conducting macroscopic scanning and collecting laser point cloud, high-definition video, and thermal imaging data. Simultaneously, explosion-proof robotic dogs are deployed, nimbly crawling under vehicles to detect fuel leaks using VOC sensors (volatile organic compound sensors), assessing vehicle structural stability with 3D cameras, and establishing contact with trapped personnel using a voice interaction module.

[0119] 3) Core Modeling: The data fusion module and the 3D digital twin modeling module of the mother vehicle command unit begin operation. They fuse real-time data with prior models to generate and continuously update a dynamic voxel mesh model. Each voxel in the dynamic voxel mesh contains dynamic attributes such as accessibility probability and overall hazard level.

[0120] 4) AI Decision-Making and Task Allocation: The AI ​​decision-making module performs physics-based simulations of multiple candidate rescue plans on a digital twin model, including at least simulating the consequences of vehicle movement and smoke diffusion paths, to predict risks and select the optimal plan. After determining the global path of the lifeline, an improved A* algorithm is used for fine-grained path planning, and an auction algorithm is used to dynamically allocate specific tasks to the most suitable vehicle operation units and unmanned reconnaissance units.

[0121] It is important to note that the cost function of the A* algorithm here incorporates the passability probability P and the overall risk level R from the dynamic attribute field.

[0122] Unmanned operation and pathway clearing refer to collaborative physical operations performed by unmanned equipment guided by digital models. This mainly includes unmanned obstacle clearing operations and collaborative monitoring and closed-loop feedback. Specifically: Unmanned obstacle removal operations involve unmanned obstacle removal vehicles using their composite walking modules to reach the target location based on instructions. They then connect to appropriate tools via hydraulic quick-connect interfaces to perform tasks such as pushing and towing. Simultaneously, a rapid rescue pallet automatically docks with the target vehicle under laser guidance, and a four-point synchronous lifting mechanism safely transports and moves the vehicle away.

[0123] Collaborative monitoring and closed-loop feedback mean that the unmanned reconnaissance unit continuously monitors the entire operation. Newly generated data is fed back to the digital twin model in real time, triggering real-time updates to the model. The AI ​​then performs simulations and planning again based on the new model, forming a closed loop of perception-decision-action-feedback, dynamically adjusting the operation plan to ensure that the operation remains safe and efficient at all times.

[0124] The personnel rescue and mission closure steps mainly include personnel rescue, auxiliary support and closure, and report generation. Specifically: Once the lifeline is opened, rescuers can enter the core area to carry out rapid and safe manual rescue in the environment where the sub-vehicle work unit has been initially cleared of danger.

[0125] The sub-vehicle operating unit can provide external power output to power the rescue equipment. After all personnel have been rescued, the units work together to complete the site cleanup and finally automatically return to the mother vehicle.

[0126] During the evacuation, the mother vehicle command unit automatically aggregates all data and generates a structured rescue report for post-event review and accountability. This comprehensive data includes, but is not limited to, timelines, resource consumption, model evolution records, and other relevant information.

[0127] In step (2) of on-site reconnaissance and situation generation, after constructing a three-dimensional digital twin model, the optimal lifeline route is planned with the assistance of AI algorithms, and specific moving or clearing targets are assigned to the vehicle operation units. Specifically, this AI algorithm-assisted planning consists of three interlocking core algorithm modules that closely interact with the three-dimensional digital twin model, including: (a) Decision verification module with simulation and inference first: The core algorithm used is the coupling of Monte Carlo Tree Search (MCTS) and embedded physical simulation engine.

[0128] The simulation-driven decision verification module runs directly on the dynamic 3D digital twin model. Before any actual physical operation, AI uses Monte Carlo Tree Search (MCTS) to simulate and generate a large number of candidate key action nodes within the model, including but not limited to how the clearing vehicle approaches, docks with, and pushes the vehicle. For each key action node simulated by the Monte Carlo Tree Search (MCTS), the system calls the built-in simplified physics engine to perform consequence simulation in a voxel model rich in physical properties, evaluating its feasibility, efficiency, and risks, including whether a secondary collision will occur and whether the vehicle structure is stable.

[0129] (b) Intelligent Planning Module with Multi-Objective Weighted Optimization: The algorithm used is an improved A* algorithm. The planning of the intelligent planning module with multi-objective weighted optimization relies entirely on the dynamic attribute fields attached to the voxels in the digital twin model. The cost function of the A* algorithm is: in, This represents the actual cumulative cost from the starting point to node n, which is the cost already incurred in getting from the rescue entrance to the current node n; This represents the heuristic estimated cost from node n to the target point, which is to estimate how much cost is still to get from the current node n to the target point, thereby guiding the search direction.

[0130] Furthermore, it can be Concretize into Here, P and R refer to the probability of passage (P) and the overall risk level (R), respectively, both obtained by querying the corresponding attribute values ​​of the voxel containing node n in the 3D digital twin model in real time. This represents the safety weighting coefficient, an adjustable parameter greater than 0 that determines the trade-off between safety and operational efficiency. A higher value indicates a greater weight given to safety in the planning process, meaning the algorithm will be more inclined to bypass high-risk areas, even if the path becomes longer; The smaller the value, the more emphasis is placed on work efficiency, and the algorithm may choose a more direct but slightly riskier path.

[0131] Furthermore, it can be Concretize into , here This represents the geometric distance, specifically the Euclidean distance from the current node n to the target point g. This distance can be used as a basic guiding term in path search to ensure the algorithm's fundamental efficiency. The urgency weighting coefficient for rescue is an adjustable parameter greater than 0, used to adjust the influence of rescue priority on the overall planning. The higher the value, the more the algorithm tends to prioritize rescuing those in more critical situations. Indicates the urgency and cost of rescuing target point g. The higher the level, the greater its urgency and cost. The larger the value, the more difficult it is to predict the cost heuristically. Overall increase.

[0132] It is important to note here that when The smaller the value, the higher the urgency of rescuing the target point g; conversely, the larger the value, the higher the urgency of rescuing the target point g. The larger the value, the lower the urgency of rescuing the target point g.

[0133] (c) Multi-agent collaborative module based on auction mechanism: The algorithm used is a distributed auction algorithm, and its specific working principle is as follows: The task planning module, i.e. the AI ​​decision-making module, in the mother vehicle command unit acts as the auctioneer, while the various sub-vehicle operation units, i.e. the unmanned clearing vehicle, the rapid rescue pallet, and the unmanned reconnaissance unit, act as bidders, and the specific rescue tasks are the auction items.

[0134] Each agent calculates its bidding cost based on factors such as real-time location, functional capabilities, current state, and path cost, and submits its bid to the AI ​​decision-making module based on the estimated cost. The AI ​​decision-making module selects the appropriate unit with the lowest cost to execute the task based on these bids. The selected unit confirms acceptance of the task and executes it, while other units continue to bid for other tasks.

[0135] Real-time location refers to the distance from the task point; functional capability refers to whether the tool or function to perform the task is available; current status includes, but is not limited to, remaining battery power and workload; path cost is the comprehensive cost of reaching the task point calculated by the A* algorithm.

[0136] The distributed auction algorithm employs decentralized decision-making, ensuring that even if some units experience communication disruptions, others can continue operating and respond in real-time to new tasks and changes in unit state. The entire auction process is visualized and monitored in real-time within a three-dimensional digital twin model.

[0137] The AI ​​simulation algorithm is precisely coupled with the previously described real-time model updates using the exponentially weighted moving average method. Specifically, AI simulations must be conducted in a stable and reliable environment. The exponentially weighted moving average method, by providing smooth and continuous model evolution, forms the cornerstone of simulation reliability. In other words, the initial state of each calculation by simulation engines for rigid body dynamics, fluid dynamics, and other physics simulations is a dynamic digital twin model updated using the exponentially weighted moving average method. If the model is not smoothed using the exponentially weighted moving average method and the raw reconnaissance data is used directly, the AI ​​simulation algorithm will produce unpredictable or even absurd results due to drastic model changes.

[0138] Furthermore, when planning a clearing plan, the AI ​​decision-making and simulation module will use this data as input for physical simulation to predict the risk of vehicle collapse or overturning during the operation, thereby avoiding catastrophic consequences caused by blind actions.

[0139] In step (3) unmanned operation and passage opening, the operator inside the mother car performs precise remote control operation of the sub-vehicle working unit through a first-person perspective and a 3D situation map. The first-person perspective is to provide the remote operator with an intuitive visual and control experience; the 3D situation map is dedicated to providing the operator with a macroscopic, multi-dimensional situational awareness of the entire work site. Specifically: The realization of the first-person perspective includes high-definition, wide dynamic range cameras mounted on the front end of the sub-vehicle's working unit, as well as a possible image-stabilized optical zoom module, which specifically provides the video source for the first-person perspective; it also includes the deep fusion of the video stream with the sub-vehicle's own LiDAR and IMU data, presenting the operator with a low-latency, highly stable real-time image on the mother vehicle's control console screen.

[0140] From this first-person perspective, the operator can clearly see the precise relative position of the robotic arm and the target object, enabling them to perform millimeter-level precision operations such as gripping, cutting, and pushing using a master-slave joystick. Simultaneously, they can visually inspect the details of vehicle damage and observe for signs of trapped personnel—complex information that pure AI models struggle to fully interpret.

[0141] The implementation of the 3D situation map includes the real-time visualization of a dynamic 3D digital twin model on a command center screen. This means that the precise locations of all vehicles, drones, bomb disposal robots, and trapped personnel are displayed on the model in real time. Furthermore, the global lifelines planned by the A* algorithm, the task objectives assigned by the auction algorithm, and the predetermined operation sequences recommended by the Monte Carlo Tree Search (MCTS) simulation are all overlaid on the 3D digital twin model as highlighted arrows, path lines, or virtual labels. Operators can switch between displaying hazard heatmaps, accessibility distribution maps, etc., with a single click, intuitively grasping the overall risk.

[0142] When operators focus on precise first-person operations, the 3D situation map allows them to understand their position and progress in the overall task at any time. They can also intuitively see the work path planned by AI. If adjustments are needed based on unforeseen circumstances, they can directly drag and drop on the 3D map to generate new path points or restricted areas, and the system will immediately replan the path.

[0143] It is important to note that the first-person perspective and the 3D situation map are not independent of each other, but rather form a deeply collaborative interactive loop. The specific workflow is as follows: 1) Task reception: The sub-vehicle receives instructions from the parent vehicle AI and autonomously drives to the vicinity of the target.

[0144] 2) Viewpoint switching and preparation: After the operator sees the vehicle in position on the 3D situation map, he can switch to the first-person view of the vehicle with one click to prepare for fine operation.

[0145] 3) Augmented Reality Overlay: The system overlays virtual guidance information generated by a digital twin model onto the first-person video feed. This virtual guidance information includes: Display a virtual capture box at the location where capture is needed.

[0146] A flashing warning icon will be displayed at the point of hazardous gas leak that needs to be avoided.

[0147] Display the simulated motion trajectory of the robotic arm.

[0148] 4) Human-machine collaborative operation: The operator refers to the virtual guidance and controls the robotic arm to perform actions. At the same time, the 3D situation map will display the movement of the robotic arm in the virtual model in real time, allowing the operator to monitor from a global perspective.

[0149] 5) Feedback and Updates: After the operation is completed, the sensors on the sub-vehicle feed the new environmental data back to the system. The digital twin model is updated in real time, and the corresponding voxel attributes change accordingly. The AI ​​then uses the new model to perform the next round of autonomous planning or recommend the next task objective to the operator.

[0150] Example 1: Tunnel traffic congestion rescue caused by multi-vehicle pile-up The main scenario is as follows: A multi-vehicle rear-end collision occurred in the middle of a long tunnel, with approximately 10 cars stuck together. One of the cars is deformed, and people are trapped inside. Communication inside the tunnel is interrupted, and the specific situation is unclear.

[0151] The specific rescue steps are as follows: (1) Alarm Response and Initial Deployment: After receiving the alarm, the command center activates the rescue system with one click. The main vehicle command unit uses the prior BIM model of the tunnel as the base map, plans the optimal travel route, and links with the traffic signal system through the V2X module to obtain a green wave of green lights. Ultimately, the main vehicle can reach the safe area outside the tunnel entrance at the fastest speed, saving more than 30% of the travel time for in-depth rescue.

[0152] (2) On-site reconnaissance and situation generation: Upon arrival at the scene, the mother vehicle establishes a forward command post. The communication network unit immediately establishes communication coverage within the tunnel via UAV relay. Subsequently, unmanned reconnaissance units are deployed, including: Using anti-collision cage technology, the drone swarm can fly flexibly between vehicles to conduct laser scanning and thermal imaging reconnaissance, quickly locating the heat source of trapped personnel.

[0153] The explosion-proof robot dog crawled under the car and detected no fuel leak at the accident scene, but a 3D scan revealed that the vehicle's structure was unstable and the trapped person was trapped. Simultaneously, the robot dog communicated briefly with the trapped person via its voice module, confirming that they were conscious but stuck.

[0154] (3) Constructing a three-dimensional digital twin model: After this, the mother vehicle command unit will integrate all the data and construct a three-dimensional digital twin model in real time. The model clearly shows the congestion structure, unstable vehicles, i.e., high risk R, and the candidate path of the optimal life channel, i.e., high accessibility P.

[0155] (4) AI Decision Making and Simulation: After modeling is completed, the AI ​​task planning module is activated, which uses the A* algorithm to plan a lifeline route that bypasses the most congested area and has the lowest overall cost on the digital model. Subsequently, the simulation module simulates the plan to move the key obstacle vehicle. The simulation found that direct towing may cause the vehicle to become unstable and collapse, so the plan was rejected and the use of a quick rescue pallet for transport was recommended instead.

[0156] (5) Unmanned Operation and Access Clearance: After the command is generated, unmanned operation and access clearance will commence. That is, the sub-vehicle operation unit will enter the site according to the AI ​​command, and the unmanned clearing vehicle will use its omnidirectional mobile chassis to flexibly shuttle and push the scattered debris and movable vehicles to both sides. For unstable critical accident vehicles, the rapid rescue pallet will autonomously drive to its bottom under the laser guidance, and the vehicle will be smoothly lifted off the ground by the four-point synchronous lifting mechanism, and then safely transported to the emergency parking lane.

[0157] Throughout the process, the drone swarm will continuously monitor and transmit the data of the newly opened space back in real time, the digital twin model will be dynamically updated, and AI will dynamically optimize the subsequent operation sequence.

[0158] (6) Personnel Rescue and Mission Closure: After the road is cleared, the personnel rescue and mission closure phase begins. That is, after the lifeline is cleared to the core area, rescue personnel enter and use the external power provided by the vehicle to operate hydraulic demolition tools to successfully rescue the trapped personnel. After all personnel are rescued, each vehicle unit works together to return the transported vehicles to their positions and clean up the site.

[0159] Meanwhile, during the evacuation process, the mother vehicle automatically generates a rescue report that includes the entire timeline, model evolution, and resource consumption.

[0160] The report shows the final rescue outcome was: All trapped personnel were successfully rescued without secondary injuries.

[0161] The total rescue time is reduced by about 50% compared to the traditional method.

[0162] There were no vehicle collapses or secondary accidents caused by blind operation throughout the entire process.

[0163] The "non-damaged transport" method protects the property of the accident vehicle to the greatest extent possible.

[0164] Example 2: Rescue Operations for a Combined Disaster Involving a Tunnel Fire and Multiple Hazard Sources The main scenario is as follows: A truck spontaneously combusts inside the tunnel, causing a fire and resulting in traffic congestion. Thick smoke is visible at the scene, and the truck's cargo is unknown, posing a potential risk of hazardous chemical leaks. A large number of passengers are trapped inside the vehicle.

[0165] The specific rescue steps are as follows: (1) Alarm response and initial deployment: After receiving an alarm with a fire alarm, the command center immediately activates the system. The mother vehicle carries the sub-vehicle unit with enhanced protection and fire extinguishing capabilities and rushes to the scene. The V2X system also opens a fast lane for it.

[0166] (2) On-site reconnaissance and situation generation: After the wooden vehicle arrives at the scene, it stops at the tunnel entrance. The unmanned reconnaissance unit focuses on reconnaissance of the fire and hazard sources, including: By using a swarm of drones and thermal imagers to penetrate dense smoke, the fire's center temperature and spread range were accurately located, and abnormal concentrations of volatile organic compounds (VOCs) in the air were detected.

[0167] The explosion-proof robot dog conducts close reconnaissance, confirms the truck's markings and the leak point, and marks and transmits this high-risk information back.

[0168] (3) Construction of 3D Digital Twin Model: After receiving the transmitted data, the mother vehicle command unit fuses the data in the data fusion module and enters the 3D digital twin modeling module to quickly construct a 3D digital twin model and dynamically update it. The voxels in the model not only mark the passability P, but also prominently mark the fire source and the area of ​​gas leakage diffusion with an extremely high hazard R value.

[0169] (4) AI Decision Making and Simulation: The simulation module simulates the smoke diffusion path and predicts that the traditional downwind attack route will be blocked by dense smoke, thus planning a rescue route that utilizes tunnel ventilation and enters from the upwind side. At the same time, the simulation shows that directly using water cannons to extinguish the truck compartment may cause unpredictable risks. Therefore, the AI ​​suggests prioritizing the control of the fire spread and cooling the rescue passage.

[0170] (5) Unmanned Operation and Access Opening: Based on the simulated paths and recommendations, the sub-vehicle operation unit will execute high-risk tasks, including: Enhanced protection unmanned clearing vehicles are connected to spraying tools and travel along AI-planned paths, creating mist-like water curtains at key points to block fire and high temperatures, thus creating conditions for subsequent operations.

[0171] Rapid rescue pallets are placed outside the fire scene to efficiently transport fireless vehicles that are blocking the way.

[0172] Meanwhile, inside the main vehicle, the operator precisely controls the direction of the water cannon of the recovery vehicle from a first-person perspective, while also using a 3D situation map to gain a comprehensive understanding of the fire and the progress of the rescue.

[0173] (6) Personnel rescue and mission completion: After the life channel is opened and safety is ensured, rescue personnel quickly enter and evacuate and guide the trapped passengers. After all personnel have been safely evacuated, a professional hazardous chemical disposal team enters and, under the continuous cover of the vehicle unit, handles the leak source and completely extinguishes the fire.

[0174] Meanwhile, a detailed report will be generated while the mother vehicle is inside the vehicle, focusing on recording the discovery and handling of hazards.

[0175] The final rescue result was: All trapped passengers were successfully evacuated, and there were no casualties.

[0176] The fire and the source of danger were effectively contained, preventing a catastrophic explosion or toxic gas spread.

[0177] The rescue operation was highly scientific, using AI simulations to avoid erroneous tactical choices and protect the safety of rescue personnel.

[0178] It provided timely and crucial decision support, and provided accurate on-site situation data for subsequent professional response teams.

Claims

1. A dedicated rescue vehicle system for performing rescue missions within a congested tunnel, characterized by, include: Mother vehicle command unit: As the main command center, it is used for overall command, route planning, situation generation and task allocation, including onboard computer, data fusion module, 3D digital twin modeling module and task planning module; Sub-vehicle operation unit: including unmanned clearing vehicle and rapid rescue pallet, used to perform physical operations such as obstacle clearing, vehicle relocation and opening of emergency access routes in tunnels; Unmanned reconnaissance unit: including a swarm of drones equipped with communication relays, gas sensors and thermal imagers, as well as explosion-proof robot dogs for close-range reconnaissance, used to collect data on the tunnel environment, hazards and living organisms; Communication network unit: including satellite and 5G backbone communication equipment and communication relay nodes carried by UAVs, used to establish and maintain real-time communication links between the mother vehicle command unit, the sub-vehicle operation unit, the unmanned reconnaissance unit and the rear command center.

2. A dedicated rescue vehicle system for performing rescue operations in congested tunnels according to claim 1, characterized in that, The mother vehicle command unit further integrates a V2X communication module for linkage with the urban traffic signal system, generating a green waveband when the mother vehicle arrives at the scene.

3. A dedicated rescue vehicle system for performing rescue operations in congested tunnels as claimed in claim 1, wherein, The three-dimensional digital modeling module in the mother vehicle command unit includes the following steps: Data fusion: Receive and register real-time laser point cloud, video stream and prior tunnel map data from unmanned reconnaissance unit, wherein objects in the video stream are identified and semantically labeled using a computer vision model; Dynamic voxel mesh generation: The registered 3D space is divided into voxel meshes, and each voxel is assigned multiple dynamic attribute fields. The dynamic attribute fields include at least the passability probability calculated based on object type and obstacle height, and the hazard level calculated based on comprehensive environmental sensor data. Real-time model updates: Based on newly received reconnaissance data, the voxel attributes of the changed areas are updated with higher weights, and the attribute values ​​of historical data are subjected to time decay processing, so that the digital twin model can synchronously map the dynamic changes of the physical world.

4. A dedicated rescue vehicle system for performing rescue operations in congested tunnels as claimed in claim 1, wherein, The unmanned obstacle clearing vehicle structure of the sub-vehicle operation unit specifically includes: A composite walking module consisting of an omnidirectional mobile chassis (11) and liftable auxiliary tracks (12); Modular work tool module based on hydraulically driven quick-change interface (13); An explosion-proof sealed vehicle body (14) and an integrated active protection (15) module are designed to adapt to hazardous environments; External power output interface (16) for use in conjunction with external rescue equipment.

5. A dedicated rescue vehicle system for performing rescue operations in congested tunnels as claimed in claim 1, wherein, The rapid rescue pallet of the sub-vehicle operation unit is equipped with a laser-guided docking mechanism, a lifting mechanism, and a self-driving system, which is used to automatically drive to the bottom of the target vehicle, lift the vehicle wheels off the ground, and then carry it.

6. A dedicated rescue vehicle system for performing rescue operations in congested tunnels as claimed in claim 1, wherein, The unmanned reconnaissance unit's drone swarm is equipped with anti-collision ball cages and an optical flow positioning system for stable flight within complex and narrow tunnels.

7. A dedicated rescue vehicle system for performing rescue operations in congested tunnels as claimed in claim 1, wherein, The explosion-proof robot dog of the unmanned reconnaissance unit is equipped with a voice interaction module for brief communication with trapped personnel; and is equipped with sensors for detecting fuel leaks under the vehicle and the vehicle's structural condition.

8. A method of using a dedicated rescue vehicle system according to any one of claims 1-7 for performing rescue missions in congested tunnels, comprising the following steps: Alarm Response and Initial Deployment: Upon receiving the alarm, the command center quickly activates the system. The main vehicle command unit calls up tunnel data, plans the optimal route, and coordinates with the traffic signal system. The main vehicle is dispatched to the predetermined stopping point outside the tunnel entrance. On-site reconnaissance and situation generation: The mother vehicle establishes a forward command post; the communication network unit establishes backbone links and relay nodes; The unmanned reconnaissance unit releases a swarm of drones and explosive-proof robot dogs to conduct three-dimensional reconnaissance; the mother vehicle command unit integrates the reconnaissance data, generates a three-dimensional digital twin model, and performs task planning and allocation based on the model; Unmanned operation and passage opening: The sub-vehicle operation unit enters the site according to instructions and clears obstacles and opens up life passages by pushing, towing, grabbing and carrying; at the same time, the unmanned reconnaissance unit coordinates environmental control and continuous monitoring. Personnel rescue and mission completion: After the life channel is opened, rescuers enter the core area to carry out rescue; the sub-vehicle operation unit provides auxiliary support; after all trapped personnel are rescued, each unit works together to complete the site cleanup, and automatically returns to the mother car after the cleanup is completed. The mother car command unit generates a rescue report during the withdrawal process.

9. The method of using a specialized rescue vehicle system for performing rescue missions in congested tunnels as defined in claim 8, characterized in that, In the on-site reconnaissance and situation generation step, after constructing a three-dimensional digital twin model, the optimal lifeline route is planned with the assistance of AI collaborative decision-making and simulation inference algorithms, and specific moving or clearing targets are assigned to the vehicle operation unit, including the following steps: The AI ​​collaborative decision-making and simulation algorithm includes: The simulation-driven decision verification module employs a core algorithm that couples Monte Carlo Tree Search (MCTS) with an embedded physics simulation engine. The intelligent planning module for multi-objective weighted optimization uses the A* algorithm. The cost function is: in, This represents the cost incurred in getting from the rescue entrance to the current node n; This represents an estimate of the remaining cost from the current node n to the target point. The multi-agent collaborative module based on the auction mechanism uses a distributed auction algorithm as its core algorithm.

10. A method of using a dedicated rescue vehicle system for performing rescue missions in congested tunnels according to claim 8, characterized in that, During the unmanned operation and passage opening process, the operator inside the mother car performs precise remote control operation of the sub-car's working unit through a first-person perspective and a three-dimensional situation map.

Images