A coordinate-free multi-modal edge-seeking navigation system

Abstract

The application discloses a kind of no coordinate multimodal edge navigation system, belong to special fire-fighting robot navigation technical field.The application discards traditional laser radar SLAM complex navigation mode, is not dependent on GNSS, UWB, 5G base station and high-precision map, realizes no coordinate type autonomous navigation.System includes: lateral double millimeter wave radar redundant ranging unit, realizes high-speed high-precision lateral edge control;Forward millimeter wave radar unit, realizes dynamic obstacle autonomous obstacle avoidance;Visual recognition unit, cooperate with the sparse visual label combined with highway mileage sign, correct mileage cumulative error;Dead reckoning unit, provide continuous mileage and emergency positioning;Fusion control unit, integrate EKF and MPC, realize lateral control, longitudinal positioning, obstacle avoidance decision, environment self-adaptation and multistage fault tolerance.

Description

Technical Field

[0001] This invention relates to the field of autonomous navigation technology for special firefighting robots, specifically a coordinate-free multimodal edge-finding navigation system. Background Technology

[0002] Highway tunnels are characterized by enclosed environments, high traffic volumes, dense pipelines, rapid fire spread, and high rescue difficulty, placing extremely high demands on the rapid arrival capabilities, driving stability, and environmental adaptability of firefighting robots. The firefighting robot must travel at high speeds (e.g., 30 km / h) along the right side of the tunnel, maintaining a safe distance (e.g., 50 cm) from the right tunnel wall. It must be able to operate reliably in all weather conditions, including darkness, dense smoke, glare, and water accumulation, and accurately reach the designated fire source (e.g., at K23+150). It must also possess the ability to perceive and avoid obstacles ahead.

[0003] Currently, the mainstream navigation solutions for tunnel firefighting robots generally suffer from the following core defects: 1. LiDAR SLAM navigation system: It relies on real-time mapping, global positioning and precise coordinate reference. The algorithm is complex, the computing power requirement is high and the hardware cost is expensive. In the environment of dense smoke, dust and water mist generated by tunnel fire, the laser point cloud decays rapidly and the map drift is serious, and the positioning and obstacle avoidance functions are directly invalidated. It is necessary to collect high-precision tunnel maps in advance and rely on external coordinate assistance such as GNSS and UWB base stations. The deployment cycle is long and the emergency response capability is poor.

[0004] 2. Pure vision navigation solution: Relies on cameras to identify lane lines or side wall texture features. The recognition fails severely in dark tunnels, under the glare of oncoming headlights, or under the obstruction of thick smoke. The all-weather working capability and reliability are insufficient.

[0005] 3. Pure millimeter-wave radar navigation solution: Although it can adapt to harsh environments and measure the distance to the side wall, it has the following defects: It can only achieve lateral edge control and cannot measure the longitudinal distance of the robot moving along the tunnel; it has slow drift in lateral control during long-term operation; it cannot know the robot's absolute position and cannot locate the fire source; it has no obstacle avoidance capability and cannot deal with dynamic obstacles such as vehicles, pedestrians, and falling objects in front.

[0006] Wheel speed encoder mileage solution: Although it can continuously measure driving distance, tire slippage, wear, tire pressure changes and other factors cause cumulative errors to increase over time, making it unable to provide accurate absolute position and unable to eliminate them.

[0007] Dense visual tagging solution: Densely deploying QR codes or reflective tags along the tunnel sidewalls can provide absolute location information, but the deployment cost is high, the workload of later maintenance is large, the navigation accuracy drops sharply after the tags are damaged, and it cannot solve the obstacle avoidance problem.

[0008] Single sensor solution: Relying on only a single type of sensor, lacking redundancy and fault tolerance mechanisms, the system will be completely paralyzed once the sensor fails, which cannot meet the high reliability requirements of fire-fighting equipment.

[0009] Existing solutions generally only support low-speed travel, which cannot meet the control response requirements of firefighting robots traveling at high speeds (e.g., 30 km / h) along edges. In summary, existing navigation technologies cannot simultaneously meet the comprehensive requirements of tunnel firefighting robots for high-speed travel, lateral edge following, longitudinal positioning, autonomous obstacle avoidance, all-weather reliability, low cost, and easy maintenance. Therefore, there is an urgent need for a multimodal edge-finding navigation system that abandons high-cost LiDAR and complex SLAM algorithms, can resist dense smoke interference, and simultaneously achieves lateral control, longitudinal positioning, and autonomous obstacle avoidance. Summary of the Invention

[0010] The technical problem this invention aims to solve is to address the aforementioned deficiencies in existing technologies by providing a coordinate-free multimodal edge-finding navigation system. Through a heterogeneous sensor fusion architecture, it employs a fusion navigation technology combining lateral dual millimeter-wave radar redundant ranging, sparse visual tag calibration, and forward radar obstacle avoidance. This achieves a three-in-one function of high-speed edge-following driving, precise longitudinal positioning, and autonomous obstacle avoidance, resulting in all-weather, highly reliable, low-cost, and high-precision tunnel fire emergency navigation. This invention does not rely on any external positioning facilities such as GNSS, UWB, 5G base stations, or lidar mapping, achieving coordinate-free autonomous navigation.

[0011] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: A coordinate-free multimodal edge-finding and navigation system, characterized in that it includes: The robot itself; A lateral dual-radar redundant ranging unit is fixedly installed on the right side of the robot body, including a first millimeter-wave radar and a second millimeter-wave radar, used to measure the lateral distance between the robot and the right sidewall of the tunnel in real time. The forward obstacle avoidance radar unit is fixedly installed at the front of the robot body and is used to detect obstacle information in front in real time. A visual recognition unit is fixedly installed on the robot body and is used to identify visual tags deployed along the right sidewall of the tunnel and read the absolute mileage information pre-stored in the visual tags. The visual tags are set in conjunction with existing highway mileage markers in the tunnel, and each tag is encoded with an absolute mileage value corresponding to the mileage marker. The dead reckoning unit, including a wheel speed encoder and an inertial measurement unit, is used to continuously measure the robot's travel distance and provide emergency positioning data when the radar fails. The fusion control unit is electrically connected to the lateral dual-radar redundant ranging unit, the forward obstacle avoidance radar unit, the visual recognition unit, and the dead reckoning unit, respectively. The fusion control unit incorporates an extended Kalman filter (EKF) and a model predictive control (MPC) algorithm. The fusion control unit is configured as follows: 1) In normal working mode, a lateral control command is generated based on the lateral distance data collected by the lateral dual radar redundant ranging unit to drive the robot to travel along the right side wall of the tunnel while maintaining a preset safe distance. 2) When the visual recognition unit recognizes the visual tag, the mileage accumulation error of the dead reckoning unit is corrected according to the absolute mileage information; 3) Assess collision risk based on obstacle information collected by the forward obstacle avoidance radar unit and generate obstacle avoidance decision instructions; 4) When the lateral dual-radar redundant ranging unit is determined to be ineffective, switch to the dead reckoning unit and limit the robot's speed. 5) When the robot's lateral distance is determined to be less than the preset safety threshold, emergency braking is triggered.

[0012] Preferably, the first millimeter-wave radar and the second millimeter-wave radar use staggered timing sampling, so that if one radar fails, the other radar can seamlessly take over, achieving uninterrupted distance measurement.

[0013] Preferably, the visual tags are sparsely deployed along the right sidewall of the tunnel, with a deployment spacing of 100m to 300m.

[0014] Preferably, the fusion control unit also dynamically adjusts the preset distance and safety threshold according to the curvature of the tunnel curve, increasing the preset distance at the curve to provide a greater safety margin.

[0015] Preferably, the fusion control unit also dynamically adjusts the preset distance and safety threshold according to the robot's current driving speed; the higher the speed, the larger the safety threshold.

[0016] Preferably, the fusion control unit obtains real-time smoke concentration through the robot's built-in smoke sensor. The fusion control unit obtains real-time smoke concentration information and dynamically adjusts the fusion weight of the lateral millimeter-wave radar data and the visual recognition unit data according to the smoke concentration. The higher the smoke concentration, the greater the weight of the radar data and the smaller the weight of the visual data.

[0017] Preferably, it also includes a mechanical tentacles protection device, installed on the right side of the robot, with a two-level triggering mechanism: the first level triggers deceleration and fine adjustment, and the second level triggers emergency braking.

[0018] Preferably, when the fusion control unit detects an interruption in the right tunnel sidewall, it enters a sidewall interruption mode, stops using sidewall distance data for lateral control, and instead maintains the course based on the heading angle and dead reckoning unit data before entering the interruption, until the sidewall reappears.

[0019] Preferably, the fusion control unit executes a graded avoidance strategy based on the obstacle information from the forward obstacle avoidance radar unit: 1) For stationary small obstacles, temporarily shift to the left to bypass them, and then smoothly return to the original edge-hugging trajectory after passing through them; 2) For stationary large obstacles, slow down, stop, and call the alarm; 3) For slower vehicles traveling in the same direction, slow down and maintain a safe distance; 4) When encountering oncoming vehicles, stay close to the right side.

[0020] A coordinate-free multimodal edge-finding navigation method, applied to the system described in any one of claims 1-9, includes the following steps: S1. The lateral distance between the robot and the right tunnel sidewall is measured in real time using a dual millimeter-wave lateral radar. S2. Real-time detection of obstacles ahead using forward-facing millimeter-wave radar; S3. Continuously measure the robot's travel distance using a wheel speed encoder; S4. Use a visual recognition unit to identify sparse visual labels combined with tunnel mileage markers and read absolute mileage information; S5. Perform multimodal fusion navigation through the fusion control unit: 1) Lateral control: The steering command is generated based on lateral radar data to maintain a preset edge distance; 2) Longitudinal calibration: When a visual tag is detected, the encoder's accumulated error is corrected using absolute mileage; 3) Obstacle avoidance decision-making: Assess risks based on forward radar data and implement graded obstacle avoidance strategies; 4) Environmental Adaptation: Adjusts fusion weights and operating modes based on smoke concentration; 5) Sidewall Discontinuity Emergency: When a sidewall discontinuity is detected, maintain course based on historical heading and dead reckoning; 6) Fault tolerance: Automatically downgrades driving mode when a sensor fails.

[0021] Preferably, the tunnel is at least one of the following: highway tunnel, urban tunnel, railway tunnel, underground utility tunnel, or mine tunnel.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Three-in-one complete navigation capability: It simultaneously realizes three core functions: lateral edge control, longitudinal mileage positioning, and autonomous obstacle avoidance, which meets the full-scenario navigation needs of tunnel fire fighting robots and solves the problem of fragmented functions in existing technologies.

[0023] 2. Eliminate external dependence and achieve "coordinate-free" navigation: Completely abandon LiDAR and complex SLAM algorithms, without the need for external coordinate assistance such as GNSS, base stations, and high-precision maps, relying only on the physical features of the tunnel sidewalls and sparse visual tags, it can be used immediately after installation, and the emergency response capability is greatly improved.

[0024] 3. Precise longitudinal positioning: By combining existing road mileage markers in the tunnel with sparse visual tags, the cumulative error of the wheel speed encoder is periodically corrected, enabling the robot to obtain a longitudinal position with meter-level accuracy without external coordinates (calibrated every 200 meters, with an error of <0.3 meters), ensuring accurate arrival at the fire source.

[0025] 4. Strong survivability in extreme environments: With millimeter-wave radar as the core sensor, it is unaffected by darkness, dense smoke, glare, rain, or snow, and can still operate stably in tunnel fire environments. The higher the smoke concentration, the more automatically the radar data fusion weights are increased to ensure the main navigation system does not fail.

[0026] 5. Dynamic Environment Adaptation: Forward obstacle avoidance radar enables the robot to perceive and respond to dynamic obstacles, allowing it to drive safely in complex tunnel environments and meet the needs of fire rescue operations.

[0027] 6. High System Robustness and Safety: The system features three levels of heterogeneous backup: lateral dual radar redundancy, visual calibration, and dead reckoning. Failure of any single sensor does not affect the system's basic functions. Combined with multiple mechanisms such as distance monitoring, graded degradation, two-level protection via mechanical tentacles, and emergency braking, the system comprehensively avoids robot collisions with tunnel structures, meeting the high safety requirements of firefighting equipment.

[0028] 7. Low deployment and maintenance costs: Visual tags are integrated with existing highway mileage markers at intervals of 100m to 300m, eliminating the need for additional measurements and dense installation, significantly reducing deployment and subsequent maintenance costs. Hardware costs are reduced by more than 60% compared to laser SLAM solutions, and computing power requirements are reduced by 70%, enabling stable operation on embedded low-end industrial control computers.

[0029] 8. High-speed and high-precision edge-following driving: At a driving speed of, for example, 30km / h, the lateral control accuracy can reach ±3-5cm, with stable edge-following and fast response, fully meeting the high-speed driving requirements of tunnel fire-fighting robots. Attached Figure Description

[0030] Figure 1 is a block diagram of the overall structure of the present invention. Figure 2 is a block diagram of the fusion control system of the present invention; Figure 3 is a flowchart of the overall workflow of the navigation method of the present invention; Figure 4 is a schematic diagram of the dual-radar interleaved timing sampling principle of the present invention; Figure 5 is a schematic diagram of the combination of visual labels and highway mileage signs according to the present invention; Figure 6 is a schematic diagram of the sidewall interruption mode control logic of the present invention; Figure 7 is a schematic diagram of the autonomous obstacle avoidance decision-making process of the present invention; Figure 8 is a schematic diagram of the installation position of the robot sensor of the present invention. Detailed Implementation

[0031] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that these descriptions are for the purpose of aiding understanding the present invention, but do not constitute a limitation thereof. Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0032] Reference Figure 1-8 A coordinate-free multimodal edge-finding navigation system, comprising: The robot itself; A lateral dual-radar redundant ranging unit is fixedly installed on the right side of the robot body, including a first millimeter-wave radar and a second millimeter-wave radar, used to measure the lateral distance between the robot and the right sidewall of the tunnel in real time. The forward obstacle avoidance radar unit is fixedly installed at the front of the robot body and is used to detect obstacle information in front in real time. A visual recognition unit is fixedly installed on the robot body and is used to identify visual tags deployed along the right sidewall of the tunnel and read the absolute mileage information pre-stored in the visual tags. The visual tags are set in conjunction with existing highway mileage markers in the tunnel, and each tag is encoded with an absolute mileage value corresponding to the mileage marker. The dead reckoning unit, including a wheel speed encoder and an inertial measurement unit, is used to continuously measure the robot's travel distance and provide emergency positioning data when the radar fails. The fusion control unit is electrically connected to the lateral dual-radar redundant ranging unit, the forward obstacle avoidance radar unit, the visual recognition unit, and the dead reckoning unit, respectively. The fusion control unit incorporates an extended Kalman filter (EKF) and a model predictive control (MPC) algorithm. The fusion control unit is configured as follows: 1) In normal working mode, a lateral control command is generated based on the lateral distance data collected by the lateral dual radar redundant ranging unit to drive the robot to travel along the right side wall of the tunnel while maintaining a preset safe distance. 2) When the visual recognition unit recognizes the visual tag, the mileage accumulation error of the dead reckoning unit is corrected according to the absolute mileage information; 3) Assess collision risk based on obstacle information collected by the forward obstacle avoidance radar unit and generate obstacle avoidance decision instructions; 4) When the lateral dual-radar redundant ranging unit is determined to be ineffective, switch to the dead reckoning unit and limit the robot's speed. 5) When the robot's lateral distance is determined to be less than the preset safety threshold, emergency braking is triggered.

[0033] In detail, two 77GHz 4D imaging millimeter-wave radars (ranging accuracy ±2cm, theoretical basis: radar manufacturer's rated parameters and simulation verification) are used, installed on the right side of the robot, with a ranging range of 0.2-15m, ranging accuracy ±2cm, and a data update rate of 50Hz. The two radars use staggered timing sampling (e.g., Figure 3 As shown, radar 1 samples at even times and radar 2 samples at odd times. After timestamp alignment, an equivalent refresh rate of 100Hz is achieved. Seamless switching is possible when a single radar fails, ensuring uninterrupted ranging.

[0034] Specifically, the first millimeter-wave radar and the second millimeter-wave radar use staggered timing sampling, so that if one radar fails, the other radar can seamlessly take over, achieving uninterrupted distance measurement.

[0035] In detail, a 77GHz mid-range millimeter-wave radar is used, which is installed in the center of the front of the robot. It has a detection range of 1-100m, a horizontal field of view of ±60°, and can track 64 targets at the same time for real-time detection of obstacles such as vehicles, pedestrians, and fallen objects in front.

[0036] Preferably, the visual tags are sparsely deployed along the right sidewall of the tunnel, with a deployment spacing of 100m to 300m.

[0037] In detail, a global shutter industrial camera, 850nm infrared illumination, and narrowband filters are used to identify the sparsely deployed AprilTag visual tags on the right tunnel sidewall. For example... Figure 4 As shown, the visual tags are integrated with existing highway mileage markers within the tunnel. Each tag is encoded with an absolute mileage value corresponding to the mileage marker (e.g., K23+100). The tag size is 200×200mm, the deployment spacing is 100m–300m, and the installation height is 1.2m. When the visual recognition unit detects a tag, it reads the absolute mileage value and calculates the robot's precise pose relative to the tag using the PnP algorithm. The deployment spacing of 100m–300m minimizes tag deployment and maintenance costs while ensuring mileage calibration accuracy.

[0038] Furthermore, the fusion control unit also dynamically adjusts the preset distance and safety threshold according to the curvature of the tunnel curve, increasing the preset distance at the curve to provide a greater safety margin.

[0039] In detail, it consists of a high-resolution wheel speed encoder (four wheels, 17-bit) and a tactical-grade inertial measurement unit (IMU). The encoder continuously measures the robot's travel distance, while the IMU measures angular velocity and acceleration. Dead reckoning is achieved through extended Kalman filtering.

[0040] Furthermore, the fusion control unit dynamically adjusts the preset distance and safety threshold according to the robot's current driving speed; the higher the speed, the larger the safety threshold.

[0041] Furthermore, the fusion control unit obtains the real-time smoke concentration through the robot's built-in smoke sensor, and dynamically adjusts the fusion weight of the lateral millimeter-wave radar data and the visual recognition unit data according to the smoke concentration. The higher the smoke concentration, the greater the weight of the radar data and the smaller the weight of the visual data.

[0042] In detail, an embedded real-time industrial control computer is used, integrating an extended Kalman filter (EKF) and a model predictive controller (MPC).

[0043] EKF fusion: Inputs include lateral radar distance, forward radar obstacle data, visual tag odometer, encoder odometer, and IMU angular velocity / acceleration; outputs robot state estimates (lateral deviation, heading angle, longitudinal odometer, distance to sidewall, velocity, etc.).

[0044] MPC control: With a target edge distance of 0.5 meters, based on the bicycle kinematics model, considering steering angle constraints (±30°), steering angle rate constraints (±10° / step) and lateral acceleration constraints (≤1.0m / s²), the optimal steering command is generated through rolling optimization, with a control cycle of 50Hz.

[0045] Longitudinal calibration logic: When the visual recognition unit detects a tag, it reads the absolute mileage value L_abs and records the current encoder cumulative mileage L_odo, calculating the error ΔL = L_abs - L_odo. The EKF uses this error to update the status and correct subsequent mileage estimates. Actual measurements show that calibration every 200 meters keeps the longitudinal positioning error within 0.3 meters.

[0046] Obstacle avoidance decision logic (e.g.) Figure 6As shown): The forward-facing radar continuously scans and generates an obstacle list; it calculates the collision time (TTC) for each obstacle = distance / relative speed; it assesses the threat level based on the TTC and obstacle size; Threat level 0 (no threat): TTC > 5s, normal driving; Threat level 1 (low threat): 3s < TTC ≤ 5s, decelerate to 20km / h, prepare to avoid the obstacle; Threat level 2 (medium threat): 1s < TTC ≤ 3s, deviate 0.3-0.5m to the left to bypass the obstacle, and smoothly return to the original trajectory after passing; Threat level 3 (high threat): TTC ≤ 1s, emergency braking, triggering an alarm.

[0047] Furthermore, it also includes a mechanical tentacle protection device, installed on the right side of the robot, with a two-level triggering mechanism: the first level triggers deceleration and fine adjustment, and the second level triggers emergency braking.

[0048] In detail, the robot has a set of elastic mechanical tentacles, 45cm in length, installed at both the front and rear of its right side. First-level trigger: When the tentacles slightly contact the sidewall, the system decelerates and makes minor adjustments in the opposite direction; second-level trigger: When the tentacles bend beyond their limit, the system immediately applies emergency braking.

[0049] Furthermore, when the fusion control unit detects an interruption in the right tunnel sidewall, it enters a sidewall interruption mode, stops using sidewall distance data for lateral control, and instead maintains the course based on the heading angle and dead reckoning unit data before entering the interruption, until the sidewall reappears.

[0050] Furthermore, the fusion control unit executes a graded avoidance strategy based on the obstacle information from the forward obstacle avoidance radar unit: 1) For stationary small obstacles, temporarily shift to the left to bypass them, and then smoothly return to the original edge-hugging trajectory after passing through them; 2) For stationary large obstacles, slow down, stop, and call the alarm; 3) For slower vehicles traveling in the same direction, slow down and maintain a safe distance; 4) When encountering oncoming vehicles, stay close to the right side.

[0051] Furthermore, a coordinate-free multimodal edge-finding navigation method, applied to the system described in any one of claims 1-9, includes the following steps: S1. The lateral distance between the robot and the right tunnel sidewall is measured in real time using a dual millimeter-wave lateral radar. S2. Real-time detection of obstacles ahead using forward-facing millimeter-wave radar; S3. Continuously measure the robot's travel distance using a wheel speed encoder; S4. Use a visual recognition unit to identify sparse visual labels combined with tunnel mileage markers and read absolute mileage information; S5. Perform multimodal fusion navigation through the fusion control unit: 1) Lateral control: The steering command is generated based on lateral radar data to maintain a preset edge distance; 2) Longitudinal calibration: When a visual tag is detected, the encoder's accumulated error is corrected using absolute mileage; 3) Obstacle avoidance decision-making: Assess risks based on forward radar data and implement graded obstacle avoidance strategies; 4) Environmental Adaptation: Adjusts fusion weights and operating modes based on smoke concentration; 5) Sidewall Discontinuity Emergency: When a sidewall discontinuity is detected, maintain course based on historical heading and dead reckoning; 6) Fault tolerance: Automatically downgrades driving mode when a sensor fails.

[0052] Furthermore, the tunnel is at least one of the following: highway tunnel, urban tunnel, railway tunnel, underground utility tunnel, or mine tunnel.

[0053] The system workflow of this invention is as follows: Figure 2 As shown: 1) Receive fire alarm task instructions; 2) Power-on self-test, sensor initialization; 3) Enter normal edge-finding mode, use lateral radar for ranging, and use MPC to control the vehicle to stay close to the edge, for example, maintain a speed of 30km / h; 4) The forward-facing radar continuously scans and assesses obstacle risks in real time; 5) Automatically calibrate longitudinal mileage when recognizing visual tags; 6) Automatic redundancy switching in case of radar failure; if both radars fail, it will enter dead reckoning at a limited speed (≤10km / h). 7) When an obstacle is detected, implement an avoidance strategy according to the threat level; 8) Monitor the distance to the sidewall in real time; trigger emergency braking if the distance falls below the safety threshold. 9) Automatically increases safe distance on curved sections and automatically maintains course when sidewalls are interrupted.

[0054] At a speed of 30km / h, this system has a lateral control accuracy RMS of 2.8cm on straight sections, a maximum deviation of 7.2cm on curved sections, and a longitudinal positioning error of <0.3m / 200m. It can successfully identify and avoid obstacles ahead and can work stably in dense smoke and dark environments, fully meeting the requirements for use as a fire-fighting robot in highway tunnels.

[0055] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. For those skilled in the art, various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and these variations still fall within the protection scope of the present invention.

Claims

1. A coordinate-free multimodal edge-finding navigation system, characterized in that, include: The robot itself; A lateral dual-radar redundant ranging unit is fixedly installed on the right side of the robot body, including a first millimeter-wave radar and a second millimeter-wave radar, used to measure the lateral distance between the robot and the right sidewall of the tunnel in real time. The forward obstacle avoidance radar unit is fixedly installed at the front of the robot body and is used to detect obstacle information in front in real time. A visual recognition unit is fixedly installed on the robot body and is used to identify visual tags laid along the right sidewall of the tunnel and read the absolute mileage information pre-stored in the visual tags. The visual labels are set up in conjunction with the existing highway mileage markers inside the tunnel, and each label is coded with an absolute mileage value corresponding to the mileage marker; The dead reckoning unit, including a wheel speed encoder and an inertial measurement unit, is used to continuously measure the robot's travel distance and provide emergency positioning data when the radar fails. The fusion control unit is electrically connected to the lateral dual-radar redundant ranging unit, the forward obstacle avoidance radar unit, the visual recognition unit, and the dead reckoning unit, respectively. The fusion control unit incorporates an extended Kalman filter (EKF) and a model predictive control (MPC) algorithm. The fusion control unit is configured as follows: 1) In normal working mode, a lateral control command is generated based on the lateral distance data collected by the lateral dual radar redundant ranging unit to drive the robot to travel along the right side wall of the tunnel while maintaining a preset safe distance. 2) When the visual recognition unit recognizes the visual tag, the mileage accumulation error of the dead reckoning unit is corrected according to the absolute mileage information; 3) Assess collision risk based on obstacle information collected by the forward obstacle avoidance radar unit and generate obstacle avoidance decision instructions; 4) When the lateral dual-radar redundant ranging unit is determined to be ineffective, switch to the dead reckoning unit and limit the robot's speed. 5) When the robot's lateral distance is determined to be less than the preset safety threshold, emergency braking is triggered.

2. The coordinate-free multimodal edge-finding navigation system according to claim 1, characterized in that, The first and second millimeter-wave radars use staggered timing sampling, so that if one radar fails, the other radar can seamlessly take over, achieving uninterrupted distance measurement.

3. The coordinate-free multimodal edge-finding navigation system according to claim 1, characterized in that, The visual labels are sparsely deployed along the right sidewall of the tunnel, with a deployment spacing of 100m to 300m.

4. The coordinate-free multimodal edge-finding navigation system according to claim 1, characterized in that, The fusion control unit also dynamically adjusts the preset distance and safety threshold according to the curvature of the tunnel curve, increasing the preset distance at the curve to provide a greater safety margin.

5. The coordinate-free multimodal edge-finding navigation system according to claim 1, characterized in that, The fusion control unit also dynamically adjusts the preset distance and safety threshold according to the robot's current driving speed; the higher the speed, the larger the safety threshold.

6. The coordinate-free multimodal edge-finding navigation system according to claim 1, characterized in that, The fusion control unit obtains real-time smoke concentration through the robot's built-in smoke sensor. The fusion control unit obtains real-time smoke concentration information and dynamically adjusts the fusion weight of lateral millimeter-wave radar data and visual recognition unit data according to the smoke concentration. The higher the smoke concentration, the greater the weight of radar data and the smaller the weight of visual data.

7. The coordinate-free multimodal edge-finding navigation system according to claim 1, characterized in that, It also includes a mechanical tentacles protection device, installed on the right side of the robot, with a two-level triggering mechanism: the first level triggers deceleration and fine adjustment, and the second level triggers emergency braking.

8. The coordinate-free multimodal edge-finding navigation system according to claim 1, characterized in that, When the fusion control unit detects an interruption in the right tunnel sidewall, it enters a sidewall interruption mode, stops using sidewall distance data for lateral control, and instead maintains the course based on the heading angle and dead reckoning unit data before entering the interruption, until the sidewall reappears.

9. The coordinate-free multimodal edge-finding navigation system according to claim 1, characterized in that, The fusion control unit executes a graded avoidance strategy based on the obstacle information from the forward obstacle avoidance radar unit: • For stationary small obstacles, temporarily shift to the left to bypass them, and then smoothly return to the original edge-hugging trajectory after passing through them; 1) For stationary large obstacles, slow down, stop, and call the alarm; 2) For slower vehicles traveling in the same direction, slow down and maintain a safe distance; 3) When encountering oncoming vehicles, stay close to the right side.

10. A coordinate-free multimodal edge-finding navigation method, applied to the system described in any one of claims 1-9, characterized in that, Includes the following steps: S1. The lateral distance between the robot and the right tunnel sidewall is measured in real time using a dual millimeter-wave lateral radar. S2. Real-time detection of obstacles ahead using forward-facing millimeter-wave radar; S3. Continuously measure the robot's travel distance using a wheel speed encoder; S4. Use a visual recognition unit to identify sparse visual labels combined with tunnel mileage markers and read absolute mileage information; S5. Perform multimodal fusion navigation through the fusion control unit: 1) Lateral control: The steering command is generated based on lateral radar data to maintain a preset edge distance; 2) Longitudinal calibration: When a visual tag is detected, the encoder's accumulated error is corrected using absolute mileage; 3) Obstacle avoidance decision-making: Assess risks based on forward radar data and implement graded obstacle avoidance strategies; 4) Environmental Adaptation: Adjusts fusion weights and operating modes based on smoke concentration; 5) Sidewall Discontinuity Emergency: When a sidewall discontinuity is detected, maintain course based on historical heading and dead reckoning; 6) Fault tolerance: Automatically downgrades driving mode when a sensor fails.

11. The coordinate-free multimodal edge-finding navigation system according to claim 1, characterized in that, The tunnel is at least one of the following: highway tunnel, urban tunnel, railway tunnel, underground utility tunnel, or mine tunnel.

Images