Robotic dynamic odor field modeling and autonomous tracing system based on biomimetic sampling

Abstract

The present application relates to the technical field of mobile gas detection and leakage source positioning, and discloses a robot dynamic odor field modeling and autonomous tracing system based on bionic sampling, wherein the bionic sampling end unit is used for realizing multi-directional active sampling, the environment and state sensing unit is used for acquiring environment wind field and robot motion state information, the dynamic odor field modeling unit is used for constructing a dynamic odor field model in combination with sampling information, and the tracing positioning unit is used for performing candidate source area extraction, guided approach and source point locking based on the dynamic odor field model. The present application can improve the multi-directional gas sampling quality, local odor field modeling authenticity and leakage source positioning reliability in a complex environment, and is suitable for target gas search and autonomous tracing in petrochemical park, industrial device area, storage area and unorganized emission scene.

Description

Technical Field

[0001] This invention relates to the field of mobile gas detection and leak source location technology, specifically a robot dynamic odor field modeling and autonomous source tracing system based on biomimetic sampling. Background Technology

[0002] In scenarios such as industrial safety monitoring, environmental pollution investigation, and hazardous gas leak early warning, existing technologies typically rely on fixed-point monitoring equipment or handheld detection instruments to complete gas monitoring and leak investigation. However, fixed-point monitoring has limited coverage and is difficult to adapt to inspection scenarios where leak sources are randomly located, diffusion paths are complex, and there are blind spots. While manual inspection has a certain degree of mobility, it depends on the operator's experience, has low detection efficiency, and poses significant safety risks in toxic, harmful, flammable, explosive, or highly corrosive environments.

[0003] In recent years, mobile robots equipped with gas sensors have provided a new technological approach for target gas search and leak source tracing. However, existing solutions still have the following shortcomings: First, most systems use fixed sampling ports or rely solely on the overall rotation of the robot to change the sampling direction, lacking the active sweeping, directional probing, and multi-angle comparison capabilities similar to those in biological olfaction. This makes it difficult to quickly obtain effective sampling information under complex wind fields, obstacle disturbances, and confined space conditions. Second, existing systems mostly use fixed sampling frequencies and fixed suction flow rates, lacking adaptive sampling strategies for different distance scales and modeling needs, making it difficult to... The existing methods have several drawbacks. First, they cannot effectively combine the ability to detect low concentrations at long distances with the ability to model local high-resolution near-source phases. Second, while some methods can collect environmental wind speed and direction information, they do not uniformly couple platform attitude, movement speed, sampling end orientation, and actual incoming flow state. As a result, the modeling and source tracing process still relies mainly on single-point concentration detection or simple hotspot tracking, which cannot accurately reconstruct the local odor field structure and plume diffusion trend in complex environments. Third, many existing methods lack a complete closed loop, including candidate source region extraction, priority ranking, near-source confirmation, false peak elimination, and source point locking criteria. This makes them prone to misjudgment under conditions of local turbulence, obstacle backflow, and instantaneous concentration fluctuations.

[0004] On the other hand, real dogs or other olfactory animals, when searching for odor sources, do not simply rely on overall body movement. Instead, they gradually build a cognition of the local odor field through flexible head and nose movements, directional probing, changes in sniffing frequency, and continuous perception of the direction of incoming airflow. This, combined with movement behavior, enables them to approach and lock onto the target source. Current robotic systems have not yet developed a complete engineering solution that organically couples biomimetic multi-directional sampling, relative incoming airflow estimation, probabilistic grid odor field modeling, wind field asymmetry correction, gradient-guided path generation, candidate source region confirmation, and final locking.

[0005] Therefore, there is an urgent need for a robotic gas detection and autonomous traceability system and method that combines high mobility, high environmental adaptability, and high traceability reliability. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a robot dynamic odor field modeling and autonomous source tracing system based on biomimetic sampling. This system solves problems such as single sampling direction, fixed suction flow rate, insufficient environmental compensation, weak local odor field modeling capability, source tracing path deviation, and low reliability of final source point confirmation in existing technologies. It enables multi-directional active perception of target gases in complex environments, dynamic reconstruction of spatial odor fields, guided approach to candidate source areas, near-source confirmation, and autonomous locking of leakage sources.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: The robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling includes: a robot platform, a biomimetic sampling end unit, an odor perception and suction control unit, an environment and state perception unit, a dynamic odor field modeling unit, a tracing and positioning unit, and a power supply and communication unit. The robot platform is used to drive the system to move in uneven ground, narrow passages, obstacle environments or other complex scenarios. The robot platform can be any one of quadruped robots, wheeled robots, tracked robots, humanoid robots, composite mobile robots or other special robots with autonomous movement and task carrying capabilities. The bionic sampling end unit is mounted on the robot platform and includes a sampling end, a sampling port, and an attitude adjustment mechanism that drives the sampling end to adjust its posture. The attitude adjustment mechanism enables the sampling end to adjust its orientation in at least one degree of freedom to form active sampling in different directions. The odor sensing and suction control unit is connected to the sampling end and includes at least one gas sensor array and an adjustable flow pump for actively suctioning and sampling the gas in the target area and adjusting the suction flow rate according to the current sampling stage. The environment and state perception unit is used to acquire information such as environmental wind speed, wind direction, robot position, platform posture, platform movement speed, and sampling end orientation. The dynamic odor field modeling unit is connected to the bionic sampling end unit, the odor sensing and suction control unit, and the environment and state sensing unit, respectively. It is used to obtain the relative inflow information at the sampling port based on the sampling point location, sampling time, gas concentration information, sampling end orientation information, environmental wind field information, and platform motion state information. Based on the relative inflow information, it performs directional weighted updates on the local odor field in the inspection area to generate a dynamic odor field model. The source tracing and localization unit is connected to the dynamic odor field modeling unit and the robot platform. It is used to extract candidate source areas based on the dynamic odor field model, evaluate the priority of candidate source areas, and control the sampling end posture, suction flow rate and robot motion strategy to perform candidate source area guidance approach, near source confirmation search and leakage source locking. The power supply and communication unit is used to supply power to the system and send the dynamic odor field model, inspection trajectory, candidate source area information, suspected leak source location, confirmed leak source location and alarm information to the remote terminal.

[0008] Preferably, the attitude adjustment mechanism includes at least one of a pitch drive mechanism, a yaw drive mechanism, and a roll drive mechanism, so that the sampling end can adjust the sampling orientation in the horizontal plane and the vertical plane; The dynamic odor field modeling unit preferentially obtains sampling data in different directions by adjusting the posture of the sampling end head, so as to improve the directional resolution of dynamic odor field modeling; When the adjustable range of the sampling end is insufficient or the change in the direction of the local odor field exceeds the preset threshold, the robot is then controlled to perform overall turning or path switching.

[0009] Preferably, the environment and state perception unit includes a wind speed and direction sensor, an inertial measurement unit, a position acquisition unit, and a sampling end attitude feedback unit; The dynamic odor field modeling unit performs coordinate transformation and velocity compensation on the environmental wind vector based on the environmental wind field, robot posture, and platform motion velocity to obtain the relative inflow information at the sampling port. This relative inflow information is used as the basis for updating the dynamic odor field model, evaluating sampling effectiveness, and determining the direction of candidate source areas. The relative inflow information at the sampling port is determined based on the environmental wind vector, platform posture matrix, and platform velocity vector. When the sampling end has a local deflection posture, the relative inflow information is also transformed based on the local posture of the sampling end to obtain the relative inflow information in the local coordinate system of the sampling end.

[0010] Preferably, the dynamic odor field modeling unit uses at least one equivalent spatial distribution model from a probabilistic grid, a directional weighted grid, or a spatiotemporal concentration distribution map to represent the dynamic odor field model, and adopts a probabilistic grid update method based on Gaussian decay mapping to transform discrete sampling points into continuous spatial influences on neighboring grid cells. The dynamic odor field modeling unit uses different update weights or influence ranges in the direction relative to the incoming flow, the direction opposite to the incoming flow, and the lateral direction to perform asymmetric correction of the wind field.

[0011] Preferably, the dynamic odor field modeling unit is further used to evaluate the sampling validity of the current sampling data based on the angle between the sampling end head and the relative incoming flow direction, the robot platform's movement speed, the platform's posture change, and the sampling end head's movement state; when the current sampling data is determined to be low-confidence sampling data, its influence weight on the dynamic odor field model update is reduced.

[0012] Preferably, the suction flow rate of the adjustable flow pump is switched according to the current sampling stage to match the data requirements of dynamic odor field modeling at different stages; The sampling phase includes at least the following stages, and at least two of the sampling stages correspond to different suction flow ranges: a wide-area exploration stage, in which a first suction flow range is used to improve the ability to detect low-concentration, long-distance odors and form a coarse-grained odor field profile; a directional probing stage, in which a second suction flow range is used to improve the ability to compare concentrations in different directions; a local verification stage, in which a third suction flow range is used to improve the consistency of local repeated sampling; and a near-source fine modeling stage, in which a fourth suction flow range is used to reduce sampling mixing effects and improve local high-resolution modeling capabilities. The dynamic odor field modeling unit and the source tracing and localization unit switch between the wide-area exploration stage, the directional trial stage, the local verification stage, and the near-source fine modeling stage according to the indicators. The criteria for determining the switching of the sampling phase include at least one or a combination of the following: the increase in the current gas concentration relative to the background value, the change in the gradient direction of the dynamic odor field model, the frequency of recurrence of local high-value areas, the degree of fluctuation of the environmental wind field, and the continuity of the odor signal.

[0013] Preferably, the source tracing and localization unit is used to extract one or more candidate source regions based on at least one of the following in the dynamic odor field model: high concentration accumulation region, local gradient convergence feature, directional consistency feature, and multi-period repeating high value region: wherein the candidate source region is a connected region in the dynamic odor field model that exceeds the background increment threshold and has local peak features. The source tracing and localization unit prioritizes the candidate source regions based on at least one of the following: peak intensity, spatial clustering, gradient consistency, consistency with the upstream direction of the environmental wind field, and historical recurrence frequency. During the process of guiding the approach to the candidate source area, the source tracing and localization unit determines the motion direction of the robot platform based on at least two of the following: the center direction of the candidate source area, the local gradient direction, and the headwind direction.

[0014] Preferably, the source tracing and localization unit controls the sampling end to perform micro-angle high-resolution directional sampling during the near-source fine modeling stage, and controls the robot platform to perform at least one of headwind approach, local circling approach, neighborhood surround sampling or multi-directional comparison sampling, so as to perform near-source confirmation search for candidate source regions; The source tracing and localization unit is also used to distinguish between true and false sources of local high peaks in the dynamic odor field model, and to exclude false peaks by at least one of the following methods: continuous time-series position offset comparison, neighborhood direction attenuation relationship comparison, or multi-cycle repeated modeling results. When at least two of the preset locking conditions are met simultaneously in multiple consecutive sampling periods of the candidate source region, the leakage source is determined to be locked.

[0015] Preferably, the system operation steps include: S1. Acquire environmental wind field information, robot position and posture information, motion speed information, and sampling end head orientation information; S2. Active suction sampling is performed through the odor sensing and suction control unit to obtain the current gas concentration information; S3. Based on the environmental wind field information, platform motion status information, and sampling end orientation information, the relative inflow information at the sampling port is obtained. Combined with the current gas concentration information, sampling point location, and spatial relationship, the local odor field is constructed and updated with directional weighting to generate a dynamic odor field model. S4. Control the sampling end to perform directional exploration sampling according to the current sampling stage, and adjust the suction flow rate of the air pump simultaneously; S5. Extract candidate source regions based on the dynamic odor field model and evaluate their priority. S6. For candidate source regions with higher priority, control the sampling head and robot platform to perform candidate source region guided approach and near-source confirmation search; S7. When the candidate source region meets the source point locking condition and the influence of false peak values ​​is eliminated, output the location of the leakage source and send the detection results and alarm information to the remote terminal.

[0016] Compared with existing technologies, this invention provides a robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling, which has the following beneficial effects: 1. This robot dynamic odor field modeling and autonomous source tracing system based on biomimetic sampling achieves multi-directional active sampling through a biomimetic sampling end with posture adjustment capability, thereby improving the ability to acquire gas information from different directions.

[0017] 2. This robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling integrates environmental wind field, platform motion state and sampling end orientation information to construct a dynamic odor field model based on the direction weighting of relative incoming flow, thereby improving the realism of local odor field modeling in complex environments.

[0018] 3. This biomimetic sampling-based robotic dynamic odor field modeling and autonomous source tracing system, through phased suction flow control and sampling action switching, enables the wide-area exploration, directional probing, local verification, and near-source fine modeling processes to match the odor field modeling requirements.

[0019] 4. This biomimetic sampling-based robotic dynamic odor field modeling and autonomous source tracing system improves the stability and reliability of leak source location through candidate source region extraction, priority evaluation, near-source confirmation search, and false peak elimination. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall structure of the robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the biomimetic sampling end unit and its attitude adjustment mechanism in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the data interaction and control relationships between various functional units of the system in this embodiment of the invention; Figure 4 This is a schematic diagram of the sampling control process in the wide-area exploration stage, the directional exploration stage, the local verification stage, and the near-source fine modeling stage in an embodiment of the present invention. Figure 5 This is a schematic diagram illustrating the directional weighted grid update of the dynamic odor field model in a local region in an embodiment of the present invention.

[0021] In the diagram: 101, Environment and State Sensing Unit; 102, Sampling Pipeline; 103, Odor Sensing and Suction Control Unit; 104, Bionic Sampling End Unit; 201, Sampling Pipeline; 202, Sampling End; 203, Sampling Port; 204, Pitch Drive Mechanism; 205, Yaw Drive Mechanism. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] As described in the background section, there are shortcomings in the existing technology. In order to solve the above-mentioned technical problems, this application proposes a robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling.

[0024] Please refer to Figures 1-2A robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling includes a robot platform, a biomimetic sampling end unit, an odor perception and suction control unit, an environment and state perception unit, a dynamic odor field modeling unit, a tracing and positioning unit, and a power supply and communication unit. The robot platform includes, but is not limited to, any one of the following: quadruped robot, wheeled robot, tracked robot, humanoid robot, composite mobile robot, or other special robots with autonomous movement and task carrying capabilities. It is used to move stably in uneven ground, narrow passages, areas with dense obstacles, and other complex working conditions, and to serve as the supporting foundation for other modules.

[0025] The biomimetic sampling end unit is installed at the front or top of the robot platform and includes a sampling end, a sampling port, and an attitude adjustment mechanism for driving the sampling end to adjust its attitude. The attitude adjustment mechanism preferably includes a yaw drive mechanism, a pitch drive mechanism, or a combination thereof, so that the sampling end can adjust its orientation in at least one degree of freedom, thereby forming multi-directional active sampling. By simulating the sweeping, probing, and directional sniffing actions of a biological head and nose, the sampling end can generate continuous directional observation data as the robot travels along the path, improving the system's ability to collect gas information from different propagation directions.

[0026] The odor sensing and suction control unit is connected to the sampling end and includes at least one multi-component gas sensor array, a sampling gas path, and an adjustable flow pump. The pump switches suction flow rate and suction stability according to different sampling stages to match the requirements of sampling sensitivity, direction discrimination capability, and spatial resolution at different stages such as wide-area exploration, directional probing, local verification, and near-source fine modeling. The gas sensor array can output raw response signals reflecting changes in target gas concentration, which, after preprocessing, serve as input for dynamic odor field modeling and source tracing decisions.

[0027] The environment and state perception unit is used to acquire information such as ambient wind speed, wind direction, platform position, platform attitude, platform movement speed, and sampling end orientation. Preferably, the environment and state perception unit includes wind speed and direction sensors, an inertial measurement unit, a position acquisition unit, and a sampling end attitude feedback unit. By uniformly introducing the ambient wind field, platform movement state, and sampling end orientation, the system can obtain the relative incoming flow information at the sampling port, providing an environmental compensation basis for subsequent direction-weighted modeling and path decision-making.

[0028] The dynamic odor field modeling unit is used to continuously construct and update the local odor field of the target gas in the inspection area based on multimodal sampling data. Preferably, the dynamic odor field modeling unit adopts a Gaussian decay mapping method based on probabilistic grids, and combines relative incoming flow direction, sampling end orientation, and asymmetric wind field correction to perform directional weighted updates on the local grid in the inspection area to generate a dynamic odor field model. Unlike simple interpolation or hotspot display based solely on concentration values, the dynamic odor field model in this invention can reflect local high-value areas and characterize diffusion direction, gradient migration trend, and local plume morphology.

[0029] The source tracing and localization unit is built upon a dynamic odor field model and is used to perform candidate source region extraction, priority ranking, path guidance, proximity confirmation, and source point locking. Preferably, the source tracing and localization unit extracts one or more candidate source regions based on local high-value areas, local gradient convergence, directional consistency, and multi-period recurring high-value features, and ranks them based on a comprehensive score according to peak intensity, spatial clustering degree, historical repetition frequency, and consistency with the upstream direction. For higher-priority candidate source regions, the source tracing and localization unit controls the robot platform to guide the approach to the candidate source region, and in the proximity phase, it uses methods such as headwind approach, local swirling approach, neighborhood surround sampling, or multi-directional comparison sampling to complete proximity confirmation and final locking.

[0030] The power supply and communication unit supplies power to the system and sends the dynamic odor field model, inspection trajectory, candidate source area information, suspected leak source location, final confirmed leak source location, and alarm information to the remote terminal. The remote terminal can be an edge computing unit, a remote monitoring center, or a smart park management platform, used to achieve real-time display, multi-machine collaboration, and emergency dispatch.

[0031] Please refer to Figures 3-5 A biomimetic sampling-based robotic dynamic odor field modeling and autonomous source tracing system, upon power-up, first completes platform attitude self-check, sampling end zero-position calibration, gas sensor background value correction, environmental and state perception unit time synchronization, and dynamic odor field model initialization. The inspection area is preferably divided into a spatial grid, with each grid cell recording the local odor field state value of the target gas at its corresponding location. A larger base grid size can be used for the wide-area inspection phase; for the near-source confirmation phase, a locally refined grid can be enabled in the neighborhood of the candidate source region to improve spatial resolution.

[0032] The operational steps of the robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling include: Initial inspection and multi-directional biomimetic sampling; During the initial inspection phase, the quadrupedal mobile platform moves along a preset fixed trajectory, which can be a reciprocating, grid-like, boundary-covering, or combination thereof. While the platform travels along the trajectory, the sampling end performs left-right sweeping, periodic deflection, or fan-shaped scanning, thereby acquiring active sampling data in different directions during continuous movement. The odor sensing and suction control unit employs a first suction flow rate range during the wide-area exploration phase to improve the ability to detect low-concentration, long-distance odors and quickly form a coarse-grained odor field profile for the target area. At any sampling time t, the system can record the following observation set: ; in, Sampling port location, This represents the current target gas concentration or sensor response value. For the platform velocity vector, For the platform attitude matrix, The angle at which the sampling tip faces. This represents the environmental wind field vector. Using the aforementioned observations, the system unifies spatial location, concentration response, wind field state, and sampling orientation as dynamic modeling inputs.

[0033] Relative inflow estimation and sampling validity assessment; To eliminate the influence of platform motion and attitude changes on incoming flow determination, the system calculates the relative incoming flow at the sampling port based on the ambient wind field, platform attitude, and platform velocity. Preferably, the relative incoming flow vector... It can be represented as: ; By using the relative incoming flow vector, the system can determine whether the current sampling is in a windward, leeward, or crosswind state, and further combine this with the orientation of the sampling end to determine the directional weight of the sampling data in subsequent modeling.

[0034] To improve modeling robustness, this implementation does not directly use the sampling data of each frame, but instead performs a reliability assessment on the current sampling data. Preferably, the sampling validity coefficient... It can be defined as: ; in, The relative inflow direction angle is determined by Together they calculated that, These are non-negative weighting coefficients. When the platform moves too fast, its attitude changes drastically, or the sampling end deviates significantly from the dominant incoming flow direction, the effectiveness of the current sampling decreases, and its contribution to model updates weakens accordingly.

[0035] Dynamic odor field modeling based on probabilistic grids and Gaussian decay mapping; The dynamic odor field modeling unit employs a probabilistic grid-based Gaussian decay mapping method to construct and update the local odor field in the inspection area online. Let the... The center position of each grid cell is Then the sampling time Update contribution to this grid cell for: ; in, This is the proportionality coefficient. These are the concentration values ​​after background correction. Let Gaussian distance decay function be used. The wind field direction weighting function is... The sampling end faces the weighting function. Preferably, the Gaussian distance decay function can be written as: ; in, This represents the spatial influence radius during the current sampling phase. During the wide-area exploration phase, Larger, to form a coarse-grained odor map; in the near-source fine modeling stage, Smaller size to improve local resolution.

[0036] To reflect the asymmetric diffusion characteristics of real plumes in the downwind and upwind directions, this embodiment introduces an asymmetric correction to the wind field direction weights. Preferably: ; in, and These represent the weighting coefficients for the downwind and upwind directions, respectively. This correction method allows the system to appropriately expand the influence range in the downwind direction and reduce the diffusion effect in the upwind direction, thereby improving the ability to depict the true plume morphology.

[0037] Preferably, the grid state value can be updated recursively: ; in, To update the coefficients, the above recursive relationship ensures that the current sampling contributes to the local odor field without completely covering historical observations, thus achieving dynamic and continuous modeling. As the robot's inspection progresses, all grid state values ​​collectively form a visualized dynamic spatial odor map.

[0038] Sampling phase switching and multi-scale modeling; like Figure 4As shown, to ensure that different sampling methods serve the modeling needs at different scales, this implementation divides the sampling process into a wide-area exploration stage, a directional exploration stage, a local verification stage, and a near-source fine modeling stage. Preferably, the system switches between stages based on indicators such as the concentration increase magnitude, local gradient consistency, frequency of high-value recurrence, wind field fluctuation degree, and signal continuity.

[0039] When the overall concentration level is low and a stable local high-value region has not yet formed, the system remains in the wide-area exploration phase, using a larger suction flow rate and a larger spatial influence radius to quickly form an initial odor field profile. When there are significant differences in the sampling response in different directions and a local gradient begins to form, the system switches to the direction probing phase, controlling the sampling end to adjust sequentially towards multiple candidate directions to enhance the direction comparison capability. When candidate directions have appeared but the local field structure still fluctuates, the system enters the local verification phase, suppressing false hotspots caused by transient disturbances through repeated sampling and model correction. When candidate regions appear continuously for multiple periods, the spatial range gradually converges, and the local high-value center tends to stabilize, the system switches to the near-source fine modeling phase, using a smaller spatial influence radius and a more refined path strategy to perform local high-resolution reconstruction of the candidate source region's neighborhood.

[0040] Candidate source region extraction and priority evaluation; like Figure 5 As shown, after the dynamic odor field model is initially formed, the source tracing and localization unit extracts candidate source regions from the model. Preferably, a candidate source region is defined as a connected region in the dynamic odor field model that exceeds the background increment threshold and has local peak characteristics. For each candidate source region, the system calculates features such as peak intensity, spatial clustering, gradient consistency, historical repetition frequency, and consistency with the upstream direction, and determines its priority through a comprehensive score. Preferably, the system sorts multiple candidate source regions according to the comprehensive score, and prioritizes the candidate source region with the higher score as the near-source confirmation search target.

[0041] Candidate source region center location The weighted centroid method can be used to obtain it:

[0042] This central location can serve as a reference point for subsequent path guidance and end-point orientation probing.

[0043] Candidate source regions guide approach and path generation; Once the priority candidate source region is determined, the robot switches from fixed-track inspection mode to candidate source region-guided approach mode. In this embodiment, the platform's movement direction is preferably determined jointly by the center direction of the candidate source region, the local gradient direction, and the headwind direction. ; in, The local gradient direction is a unit vector. The relative incoming flow vector, , where is a non-negative weighting coefficient. This expression means that the platform dynamically adjusts its approach path after integrating information from the candidate source region center, local gradient, and headwind, thereby reducing the probability of path deviation and entering false peak areas. During the approach process, the platform can simultaneously combine obstacle avoidance constraints, terrain information, or local access conditions to perform projection corrections or path smoothing on the movement direction, in order to meet the walking safety and stability requirements in engineering applications.

[0044] Near-source confirmation, false peak elimination, and source identification; Once the robot enters the neighborhood of the candidate source region, the system switches to near-source confirmation and search mode. At this time, the sampling head performs micro-angle high-resolution directional sampling, the air pump switches to the suction flow rate of the near-source fine modeling stage, and the platform selects at least one movement mode from the following based on the local odor field structure: headwind approach, local swirling approach, neighborhood surround sampling, or multi-directional comparison sampling. If the local high-value center is relatively stable, headwind approach is preferred; if the high-value center drifts, local swirling or neighborhood surround sampling is preferred; if there are multiple adjacent high-value peaks in the candidate region, multi-directional comparison sampling is preferred.

[0045] To avoid false peaks caused by local turbulence, backflow from obstacles, or instantaneous concentration fluctuations, this embodiment includes a mechanism to distinguish between true and false sources. Preferably, the center position drift, neighborhood attenuation consistency, and multi-cycle repetition stability are calculated for multiple consecutive sampling cycles of the priority candidate source region. The system determines that source point locking is complete when at least two of the following conditions are met simultaneously for multiple consecutive cycles of the candidate source region: 1) The center position drift is less than the preset threshold; 2) The central response of the candidate source region is consistently higher than that of adjacent regions; 3) The candidate source region exhibits a local attenuation characteristic that gradually decreases in the direction away from the center; 4) The responses under different sampling directions indicate that this region is the center of the strongest local response; 5) The relationship between the candidate source region and the upstream direction of the environmental wind field meets the preset consistency condition.

[0046] If a local high-value area appears only momentarily in a single sampling, has obvious location drift, or does not show reasonable attenuation in the surrounding area, it is determined to be a false peak and is not used as the final source point.

[0047] After the source point is located, the platform can stay near the located area for a short period of time and perform supplementary retests. At the same time, it can send the final leak source location, alarm level, local odor map and inspection trajectory information to the remote terminal through the communication unit.

[0048] Through the above implementation methods, this invention organically combines biomimetic multi-directional sampling, relative incoming flow estimation, probabilistic grid odor field modeling, Gaussian attenuation mapping, wind field asymmetric correction, gradient-guided path generation, candidate source region priority ranking, near-source confirmation, and source point locking to form a complete mobile autonomous source tracing closed loop. Compared with existing solutions that rely solely on single-point high-value detection, fixed sampling, or simple path tracking, this invention can more effectively improve the realism of odor field reconstruction, the efficiency of mobile source tracing, and the reliability of final location in complex environments.

[0049] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling, characterized in that, include: Robot platform, bionic sampling end unit, odor sensing and suction control unit, environment and state sensing unit, dynamic odor field modeling unit, source tracing and positioning unit, and power supply and communication unit; The robot platform is used to drive the system to move in uneven ground, narrow passages, obstacle environments or other complex scenarios. The robot platform can be any one of quadruped robots, wheeled robots, tracked robots, humanoid robots, composite mobile robots or other special robots with autonomous movement and task carrying capabilities. The bionic sampling end unit is mounted on the robot platform and includes a sampling end, a sampling port, and an attitude adjustment mechanism that drives the sampling end to adjust its posture. The attitude adjustment mechanism enables the sampling end to adjust its orientation in at least one degree of freedom to form active sampling in different directions. The odor sensing and suction control unit is connected to the sampling end and includes at least one gas sensor array and an adjustable flow pump for actively suctioning and sampling the gas in the target area and adjusting the suction flow rate according to the current sampling stage. The environment and state perception unit is used to acquire information such as environmental wind speed, wind direction, robot position, platform posture, platform movement speed, and sampling end orientation. The dynamic odor field modeling unit is connected to the bionic sampling end unit, the odor sensing and suction control unit, and the environment and state sensing unit, respectively. It is used to obtain the relative inflow information at the sampling port based on the sampling point location, sampling time, gas concentration information, sampling end orientation information, environmental wind field information, and platform motion state information. Based on the relative inflow information, it performs directional weighted updates on the local odor field in the inspection area to generate a dynamic odor field model. The source tracing and localization unit is connected to the dynamic odor field modeling unit and the robot platform. It is used to extract candidate source areas based on the dynamic odor field model, evaluate the priority of candidate source areas, and control the sampling end posture, suction flow rate and robot motion strategy to perform candidate source area guidance approach, near source confirmation search and leakage source locking. The power supply and communication unit is used to supply power to the system and send the dynamic odor field model, inspection trajectory, candidate source area information, suspected leak source location, confirmed leak source location and alarm information to the remote terminal.

2. The robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling according to claim 1, characterized in that: The attitude adjustment mechanism includes at least one of a pitch drive mechanism, a yaw drive mechanism, and a roll drive mechanism, enabling the sampling end to adjust the sampling orientation in the horizontal and vertical planes. The dynamic odor field modeling unit preferentially obtains sampling data in different directions by adjusting the posture of the sampling end head, so as to improve the directional resolution of dynamic odor field modeling; When the adjustable range of the sampling end is insufficient or the change in the direction of the local odor field exceeds the preset threshold, the robot is then controlled to perform overall turning or path switching.

3. The robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling according to claim 1, characterized in that: The environment and state perception unit includes a wind speed and direction sensor, an inertial measurement unit, a position acquisition unit, and a sampling end attitude feedback unit. The dynamic odor field modeling unit performs coordinate transformation and velocity compensation on the environmental wind vector based on the environmental wind field, robot posture, and platform motion velocity to obtain the relative inflow information at the sampling port. This relative inflow information is used as the basis for updating the dynamic odor field model, evaluating sampling effectiveness, and determining the direction of candidate source areas. The relative inflow information at the sampling port is determined based on the environmental wind vector, platform posture matrix, and platform velocity vector. When the sampling end has a local deflection posture, the relative inflow information is also transformed based on the local posture of the sampling end to obtain the relative inflow information in the local coordinate system of the sampling end.

4. The robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling according to claim 1, characterized in that: The dynamic odor field modeling unit uses at least one equivalent spatial distribution model, such as a probabilistic grid, a directional weighted grid, or a spatiotemporal concentration distribution map, to represent the dynamic odor field model. It also uses a probabilistic grid update method based on Gaussian decay mapping to transform discrete sampling points into continuous spatial influences on neighboring grid cells. The dynamic odor field modeling unit uses different update weights or influence ranges in the direction relative to the incoming flow, the direction opposite to the incoming flow, and the lateral direction to perform asymmetric correction of the wind field.

5. The robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling according to claim 1, characterized in that: The dynamic odor field modeling unit is also used to evaluate the sampling validity of the current sampling data based on the angle between the sampling end head and the relative incoming flow direction, the robot platform's movement speed, the platform's posture change, and the sampling end head's movement state; when the current sampling data is determined to be low-confidence sampling data, its influence weight on the dynamic odor field model update is reduced.

6. The robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling according to claim 1, characterized in that: The suction flow rate of the adjustable flow pump is switched according to the current sampling stage to match the data requirements of dynamic odor field modeling at different stages. The sampling phase includes at least the following phases, and at least two of the sampling phases correspond to different suction flow ranges: wide-area exploration phase, in which a first suction flow range is used to improve the ability to detect low-concentration, long-distance odors and form a coarse-grained odor field profile; In the directional exploration phase, a second suction flow rate range is used to improve the ability to compare concentrations in different directions; in the local verification phase, a third suction flow rate range is used to improve the consistency of local repeated sampling. In the near-source fine modeling stage, a fourth suction flow range is used to reduce sampling mixing effects and improve local high-resolution modeling capabilities. The dynamic odor field modeling unit and the source tracing and localization unit switch between the wide-area exploration stage, the directional trial stage, the local verification stage, and the near-source fine modeling stage according to the indicators. The criteria for determining the switching of the sampling phase include at least one or a combination of the following: the increase in the current gas concentration relative to the background value, the change in the gradient direction of the dynamic odor field model, the frequency of recurrence of local high-value areas, the degree of fluctuation of the environmental wind field, and the continuity of the odor signal.

7. The robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling according to claim 1, characterized in that: The source tracing and localization unit is used to extract one or more candidate source regions based on at least one of the following in the dynamic odor field model: high concentration accumulation region, local gradient convergence feature, directional consistency feature, and multi-period repeating high value region. The candidate source region is a connected region in the dynamic odor field model that exceeds the background increment threshold and has local peak features. The source tracing and localization unit prioritizes the candidate source regions based on at least one of the following: peak intensity, spatial clustering, gradient consistency, consistency with the upstream direction of the environmental wind field, and historical recurrence frequency. During the process of guiding the approach to the candidate source area, the source tracing and localization unit determines the motion direction of the robot platform based on at least two of the following: the center direction of the candidate source area, the local gradient direction, and the headwind direction.

8. The robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling according to claim 1, characterized in that: The source tracing and localization unit controls the sampling end to perform micro-angle high-resolution directional sampling during the near-source fine modeling stage, and controls the robot platform to perform at least one of headwind approach, local circling approach, neighborhood surround sampling or multi-directional comparison sampling to perform near-source confirmation search for candidate source regions. The source tracing and localization unit is also used to distinguish between true and false sources of local high peaks in the dynamic odor field model, and to exclude false peaks by at least one of the following methods: continuous time-series position offset comparison, neighborhood direction attenuation relationship comparison, or multi-cycle repeated modeling results. When at least two of the preset locking conditions are met simultaneously in multiple consecutive sampling periods of the candidate source region, the leakage source is determined to be locked.

9. A robot dynamic odor field modeling and autonomous tracing system based on biomimetic sampling, characterized in that, The system operation steps include: S1. Acquire environmental wind field information, robot position and posture information, motion speed information, and sampling end head orientation information; S2. Active suction sampling is performed through the odor sensing and suction control unit to obtain the current gas concentration information; S3. Based on the environmental wind field information, platform motion status information, and sampling end orientation information, the relative inflow information at the sampling port is obtained. Combined with the current gas concentration information, sampling point location, and spatial relationship, the local odor field is constructed and updated with directional weighting to generate a dynamic odor field model. S4. Control the sampling end to perform directional exploration sampling according to the current sampling stage, and adjust the suction flow rate of the air pump simultaneously; S5. Extract candidate source regions based on the dynamic odor field model and evaluate their priority. S6. For candidate source regions with higher priority, control the sampling head and robot platform to perform candidate source region guided approach and near-source confirmation search; S7. When the candidate source region meets the source point locking condition and the influence of false peak values ​​is eliminated, output the location of the leakage source and send the detection results and alarm information to the remote terminal.

Images