A mobile robot control device that takes into account field navigation patrol and transportation of a load

Abstract

The application discloses a kind of mobile robot control device giving consideration to field navigation patrol and transport load, it is related to the field of robot control, comprising: sensor data acquisition module, forms multi-source sensor data;Environment understanding module, terrain risk and weather influence are predicted based on feature vector, and environment feature data and risk prediction are output;Navigation control module, navigation path is carried out based on environment feature data, and robot motion control instruction is generated;Load management module, the parameter of navigation control module is adjusted based on load state;Feedback adjustment module, control parameter is dynamically adjusted according to load state, and feedback is carried out to navigation control module to optimize and modify robot motion control instruction;Intelligent detection module detects environmental anomaly and dangerous event, and outputs alarm signal.The application has the advantages that: through multi-sensor layered feature extraction and attention weighted fusion, environment risk is accurately predicted, and motion parameters are dynamically adjusted in combination with moment balance model.

Description

Technical Field

[0001] This invention relates to the field of robot control, specifically to a mobile robot control device that combines field navigation and patrol with payload transport. Background Technology

[0002] With the widespread application of mobile robot technology in fields such as field exploration, material transportation, and environmental inspection, autonomous navigation and load transport capabilities have become critical requirements. However, existing mobile robot control systems suffer from several drawbacks in complex field environments. Their visual sensors fail in rainy or foggy weather, and lidar is insufficient in predicting the risk of collapse on soft surfaces (such as mud), leading to path planning that ignores environmental risk factors (sudden slope changes, collapse probability) and increases operational risks. Furthermore, the inability to adjust motion parameters (speed, radius of curvature) in real time during movement can easily cause the robot to tip over, especially on steep slopes or in strong winds. Summary of the Invention

[0003] To address the aforementioned technical problems, a mobile robot control device that combines field navigation and patrol with payload transport is provided. This technical solution solves the problems mentioned in the background art, such as one-sided environmental perception, poor adaptability to dynamic changes in payload, inefficient multi-source data fusion, and insufficient real-time response in complex field environments.

[0004] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0005] A mobile robot control device that combines field navigation and patrol with payload transport includes:

[0006] The sensor data acquisition module acquires visual data, lidar data, IMU data, and environmental sensor data from the field environment through visual sensors, lidar, IMU, and environmental sensors, forming multi-source sensor data;

[0007] The environmental understanding module is connected to the sensor data acquisition module. It receives and processes multi-source sensor data, fuses multi-sensor data through a deep learning model, extracts features and recognizes patterns, and predicts terrain risks and weather impacts based on feature vectors, outputting environmental feature data and risk predictions.

[0008] The navigation control module is connected to the environment understanding module. It determines the navigation path based on environmental feature data and generates robot motion control commands.

[0009] The load management module monitors the transport load status in real time and adjusts the parameters of the navigation control module based on the load status.

[0010] The feedback adjustment module is connected to the navigation control module and the load management module. It dynamically adjusts the control parameters according to the load status and feeds them back to the navigation control module to optimize and modify the robot's motion control commands.

[0011] The intelligent detection module is connected to the sensor data acquisition module and the environmental understanding module to detect environmental anomalies and dangerous events, and output alarm signals.

[0012] Preferably, the step of fusing multi-sensor data through a deep learning model, performing feature extraction and pattern recognition, and predicting terrain risk and weather impact based on feature vectors to output environmental feature data and risk prediction specifically includes:

[0013] Spatiotemporal synchronization of visual data, LiDAR data, IMU data, and environmental sensor data is performed, and heterogeneous data are unified into the robot's body coordinate system through a coordinate transformation matrix;

[0014] Hierarchical feature extraction is performed on the data inside the robot's body coordinate system to obtain feature data;

[0015] Based on the attention mechanism and dynamic weighted fusion step, the output multimodal feature vector is used to generate the joint feature vector of the environment;

[0016] The joint feature vector of the environment is input into the multilayer perceptron to output the terrain risk level and weather impact coefficient. The terrain risk level includes abrupt slope change and collapse probability, and the weather impact coefficient includes visibility reduction due to rain and fog and wind speed disturbance intensity.

[0017] Preferably, the step of performing hierarchical feature extraction on the data within the robot's body coordinate system to obtain feature data specifically includes:

[0018] After spatiotemporal synchronization, the multi-source sensor data is classified and processed according to sensor type to obtain visual data, lidar data, IMU data, and environmental sensor data.

[0019] Spatial texture features are extracted from visual data using convolutional neural networks;

[0020] Three-dimensional structural features are extracted from lidar data using point cloud networks;

[0021] Motion state features are extracted from IMU data using a time-series model;

[0022] Extracting statistical features from environmental sensor data;

[0023] The extracted and statistical features are hierarchically encoded according to sensor type to generate multimodal feature vectors.

[0024] Preferably, the dynamic weighted fusion step based on the attention mechanism, which outputs a multimodal feature vector to generate a joint environmental feature vector, specifically includes:

[0025] The three-dimensional structural feature vector extracted from the point cloud by lidar is used as the reference vector;

[0026] The attention weight coefficients of each modality feature are generated by calculating the cosine similarity between the spatial texture feature vector extracted by the computer vision sensor, the motion state feature vector extracted by the IMU sensor, and the statistical feature vector extracted by the environmental sensor and the reference vector.

[0027] The multimodal feature vectors are dynamically weighted and fused based on the attention weight coefficient to generate a joint environmental feature vector. The attention weight coefficient is determined by calculating the cosine similarity between the spatial texture feature vector extracted by the visual sensor, the motion state feature vector extracted by the IMU sensor, and the statistical feature vector extracted by the environmental sensor and the three-dimensional structural feature vector extracted by the lidar point cloud.

[0028] The fused joint environmental feature vector is output to the risk prediction module for predicting terrain risk level and weather impact coefficient.

[0029] Preferably, the step of generating a navigation path based on environmental feature data and generating robot motion control commands specifically includes:

[0030] Receive environmental characteristic data, terrain risk level, and weather impact coefficient output by the environmental understanding module;

[0031] A cost function that integrates environmental risks and motion constraints is constructed, with the robot's current position as the starting point, the target point as the ending point, and the grid map as the search space;

[0032] Nodes with a slope change risk quantification value greater than the maximum slope tolerance or a wind speed disturbance cost greater than the collapse probability safety threshold are removed to generate an initial navigation path. The maximum slope tolerance and collapse probability safety threshold are obtained through statistical analysis of historical experimental data.

[0033] The initial path is optimized locally in real time using the dynamic window method to generate motion control commands that include linear velocity and angular velocity.

[0034] The motion control commands are output to the robot chassis actuator, and the load status parameters fed back by the load management module are received simultaneously to dynamically adjust the path curvature radius threshold and maximum speed.

[0035] Preferably, the construction of the cost function that integrates environmental risk and motion constraints specifically includes:

[0036] ;

[0037] in, To integrate the cost function output value of environmental risk and motion constraints, The cost is the Euclidean distance between nodes. This is the quantified value of slope abrupt change risk in the medium terrain risk level. This represents the probability value of collapse within the medium terrain risk level. The price paid for reduced visibility due to rain and fog, The wind speed disturbance cost is represented by α, β, γ, and δ, which are attention weighting coefficients.

[0038] Preferably, the real-time monitoring of the transport load status and the adjustment of the navigation control module parameters based on the load status specifically includes:

[0039] The load distribution is detected in real time by a pressure sensor array, and the load center of gravity offset is calculated.

[0040] When the center of gravity offset exceeds the preset safety threshold, a deceleration command is sent to the navigation control module, and the path curvature radius threshold is dynamically reduced. The preset safety threshold is determined by statistically analyzing historical experimental data under different load distributions and combining the current terrain risk level and weather influence coefficient.

[0041] Based on the load weight distribution data, the robot chassis attitude angle is adjusted through the torque balance model, chassis tilt compensation command is generated and synchronized to motion control command.

[0042] Preferably, the dynamic adjustment of control parameters according to the load state specifically includes:

[0043] Obtain the center of gravity offset and load weight distribution data from the load management module;

[0044] Based on the proportional relationship between the center of gravity offset and the preset safety threshold, the calculation speed is reduced by a scaling factor.

[0045] Calculate the radius of curvature reduction factor based on the magnitude of the centroid offset;

[0046] Based on the load weight distribution data, the tilt compensation angle of the robot chassis is calculated using a torque balance model.

[0047] The speed reduction factor, curvature radius reduction factor, and tilt compensation angle are used as control parameter adjustment quantities and fed back to the navigation control module to adjust the linear velocity, angular velocity, and chassis attitude of the robot motion control commands in real time.

[0048] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0049] This invention proposes a mobile robot control device that combines field navigation and patrol with payload transportation. Using the 3D structural features of a lidar sensor as a reference vector, it calculates dynamic weights based on the cosine similarity of other modal features. These weighted features are then fused to generate a joint environmental feature vector, which is input into a multilayer perceptron (MLP). The output includes terrain risk levels (slope abrupt changes, collapse probability) and weather impact coefficients (visibility reduction in rain and fog, wind speed disturbance). This allows for automatic reduction of visual weights in adverse weather conditions, increasing the contributions of both lidar and environmental sensors. This reduces terrain risk prediction errors in rain and fog scenarios and improves the accuracy of collapse probability recognition. Simultaneously, the hierarchical processing and attention mechanism shortens feature extraction time, meeting real-time response requirements. Furthermore, the dynamic weighting mechanism based on lidar features effectively addresses sensor data conflicts.

[0050] This invention proposes a mobile robot control device that combines field navigation and patrol with payload transport. It uses a pressure sensor array to monitor load distribution in real time, calculates center of gravity offset and weight distribution, and presets a safety threshold by integrating historical overturning critical values, real-time terrain risk levels (slope abrupt changes), and weather influence coefficients (wind speed disturbances). When the center of gravity offset exceeds the threshold, a deceleration command is triggered, and the path curvature radius threshold is reduced. Simultaneously, based on load weight distribution data, a rigid body dynamics model is used to calculate the chassis tilt compensation angle, enabling multi-parameter coordinated adjustment. This allows for automatic tightening of the safety threshold when the slope abrupt change risk is ≥2, dynamically adapting to complex terrain. Linear speed reduction ensures stability, and exponential reduction of the curvature radius limits centrifugal force (e.g., when the offset reaches 120% of the threshold, the curvature radius shrinks to 80%). Furthermore, the tilt angle θ is calculated using a torque balance model (e.g., θ = 0.1 rad), actively adjusting the chassis attitude. Attached Figure Description

[0051] Figure 1 This is a schematic diagram of the process of the present invention;

[0052] Figure 2 This is a schematic diagram of the process in this invention that uses a deep learning model to fuse multi-sensor data, extract features and recognize patterns, and predict terrain risks and weather impacts based on feature vectors, outputting environmental feature data and risk predictions.

[0053] Figure 3 This is a schematic diagram of the process of extracting feature data from the internal data of the robot body coordinate system in this invention.

[0054] Figure 4 This is a schematic diagram of the process of generating a joint environmental feature vector from the multimodal feature vector output by the dynamic weighted fusion step based on the attention mechanism in this invention.

[0055] Figure 5This is a schematic diagram of the process in this invention for generating a navigation path based on environmental feature data and generating robot motion control commands.

[0056] Figure 6 This is a schematic diagram of the process in this invention for real-time monitoring of the transport load status and adjusting the parameters of the navigation control module based on the load status;

[0057] Figure 7 This is a schematic diagram of the process of dynamically adjusting control parameters according to load status in this invention. Detailed Implementation

[0058] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0059] Reference Figure 1 As shown, a mobile robot control device that combines field navigation and patrol with payload transport includes:

[0060] The sensor data acquisition module acquires visual data, lidar data, IMU data, and environmental sensor data from the field environment through visual sensors, lidar, IMU, and environmental sensors, forming multi-source sensor data;

[0061] The environmental understanding module is connected to the sensor data acquisition module. It receives and processes multi-source sensor data, fuses multi-sensor data through a deep learning model, extracts features and recognizes patterns, and predicts terrain risks and weather impacts based on feature vectors, outputting environmental feature data and risk predictions.

[0062] The navigation control module is connected to the environment understanding module. It determines the navigation path based on environmental feature data and generates robot motion control commands.

[0063] The load management module monitors the transport load status in real time and adjusts the parameters of the navigation control module based on the load status.

[0064] The feedback adjustment module is connected to the navigation control module and the load management module. It dynamically adjusts the control parameters according to the load status and feeds them back to the navigation control module to optimize and modify the robot's motion control commands.

[0065] The intelligent detection module is connected to the sensor data acquisition module and the environmental understanding module to detect environmental anomalies and dangerous events, and output alarm signals.

[0066] The sensor data acquisition module integrates visual sensors (such as RGB-D cameras), LiDAR (such as the Velodyne series), IMU (such as the MPU9250), and environmental sensors (temperature, humidity, barometer) to achieve synchronous acquisition of multi-source heterogeneous data. The environmental understanding module adopts a hierarchical feature extraction architecture: First, the heterogeneous data is spatiotemporally aligned to the robot's body coordinate system through a coordinate transformation matrix; second, different sensor data are processed separately through dedicated networks—spatial texture features are extracted from visual data using ResNet, 3D structural features are extracted from LiDAR data using PointNet++, motion state features are extracted from IMU data using LSTM, and statistical features are extracted from environmental sensors; then, multimodal feature vectors are dynamically weighted and fused based on an attention mechanism (using LiDAR features as a benchmark, calculating the cosine similarity between other modal features and the cosine as a weight coefficient) to generate a joint environmental feature vector; finally, the terrain risk level (including slope abrupt change quantization value and collapse probability value) and weather impact coefficient (including visibility attenuation due to rain and fog, and wind speed disturbance intensity) are output through a multilayer perceptron. The navigation control module can construct a cost function that incorporates environmental risks, directly eliminate path nodes to generate an initial navigation path, and then optimize the path in real time using a dynamic window method, outputting motion control commands containing linear velocity and angular velocity. The load management module detects load distribution through a pressure sensor array, calculates the center of gravity offset, and triggers a deceleration command and reduces the path curvature radius threshold when the offset exceeds a preset safety threshold (dynamically adjusted based on terrain risk and weather coefficients). Simultaneously, it calculates the robot chassis tilt compensation angle based on a torque balance model. The feedback adjustment module calculates the speed reduction factor, curvature radius reduction factor, and tilt compensation angle based on the ratio of the center of gravity offset to the safety threshold, adjusting the motion command parameters of the navigation control module in real time. The intelligent detection module uses anomaly pattern recognition algorithms (such as isolated forests) to jointly analyze the raw sensor data and the risk predictions output by the environmental understanding module, triggering environmental hazard alarms.

[0067] Thus, during its use, the environmental feature extraction error can be reduced and the terrain risk prediction accuracy can be improved through dynamic weighted fusion (cosine similarity weight allocation) via attention mechanism. At the same time, the dynamic adjustment mechanism of environmental risk weight coefficient in the cost function improves the success rate of path avoidance when the slope changes by more than 15° or the wind speed is greater than 8m / s, making the navigation scheme more adaptable to complex terrain in the wild. The chassis attitude compensation algorithm based on the torque balance model improves the load offset tolerance and reduces the number of emergency stops in the center of gravity offset handling scheme.

[0068] Reference Figure 2 As shown, the process of fusing multi-sensor data through a deep learning model, performing feature extraction and pattern recognition, and predicting terrain risk and weather impact based on feature vectors to output environmental feature data and risk prediction specifically includes:

[0069] Spatiotemporal synchronization of visual data, LiDAR data, IMU data, and environmental sensor data is performed, and heterogeneous data are unified into the robot's body coordinate system through a coordinate transformation matrix;

[0070] Hierarchical feature extraction is performed on the data inside the robot's body coordinate system to obtain feature data;

[0071] Based on the attention mechanism and dynamic weighted fusion step, the output multimodal feature vector is used to generate the joint feature vector of the environment;

[0072] The joint feature vector of the environment is input into the multilayer perceptron to output the terrain risk level and weather impact coefficient. The terrain risk level includes abrupt slope change and collapse probability, and the weather impact coefficient includes visibility reduction due to rain and fog and wind speed disturbance intensity.

[0073] A hardware timestamp alignment mechanism (such as the PTP protocol) is adopted to achieve synchronous acquisition of multi-source sensor data. The visual data (RGB-D camera), LiDAR point cloud, IMU pose and environmental sensor (temperature, humidity, barometer) data are uniformly transformed into the robot body coordinate system through a coordinate transformation matrix (homogeneous transformation matrix). Specifically: the LiDAR point cloud is transformed into the body coordinate system through an extrinsic parameter calibration matrix; the IMU data is aligned with the body coordinate system after pre-integration to compensate for motion distortion; and the visual data is projected into the same coordinate system in combination with camera intrinsic parameters and depth information. This allows for the following: Visual data is processed using convolutional neural networks (such as ResNet) to extract spatial texture features, capturing environmental details such as surface vegetation and rock distribution; LiDAR data is processed using point cloud networks (such as PointNet++) to extract 3D structural features, quantifying terrain undulations and obstacle outlines; IMU data is processed using temporal models (such as LSTM) to extract motion state features (acceleration, angular velocity changes); environmental sensor data is processed by calculating statistical features (mean, variance) to characterize dynamic changes in temperature, humidity, and air pressure; feature encoding allows the above features to be hierarchically encoded into multimodal feature vectors according to sensor type (e.g., visual features have a dimension of 512, LiDAR features have a dimension of 256), completing hierarchical feature extraction, preserving modality specificity (visual focuses on texture, LiDAR focuses on structure), and adaptive weighting through attention mechanisms (such as cosine similarity dynamically allocating weights) to reduce fusion errors. At the same time, hierarchical feature extraction compresses redundant information, reduces the input dimension of the MLP, and shortens inference time.

[0074] Subsequently, using the 3D structural feature vector of the LiDAR as the reference vector, the cosine similarity between the visual texture features, IMU motion features, environmental statistical features, and the reference vector is calculated to generate attention weight coefficients for each modality. Based on these weight coefficients, the multimodal features are dynamically weighted and fused to generate a joint environmental feature vector. This joint environmental feature vector is then input into a multilayer perceptron (MLP) to output the terrain risk level. The prediction layer employs a sigmoid activation function to normalize the risk value to the [0,1] interval. The similarity-based attention mechanism automatically reduces the visual weight and increases the weights of the LiDAR and environmental sensors in rainy or foggy weather, thus improving the accuracy of terrain risk prediction.

[0075] Reference Figure 3 As shown, the hierarchical feature extraction of the data within the robot's body coordinate system to obtain feature data specifically includes:

[0076] After spatiotemporal synchronization, the multi-source sensor data is classified and processed according to sensor type to obtain visual data, lidar data, IMU data, and environmental sensor data.

[0077] Spatial texture features are extracted from visual data using convolutional neural networks;

[0078] Three-dimensional structural features are extracted from lidar data using point cloud networks;

[0079] Motion state features are extracted from IMU data using a time-series model;

[0080] Extracting statistical features from environmental sensor data;

[0081] The extracted and statistical features are hierarchically encoded according to sensor type to generate multimodal feature vectors.

[0082] After achieving spatiotemporal synchronization of multi-source sensor data (through hardware timestamp alignment and coordinate transformation matrix unification to the robot's body coordinate system), the data is first categorized by sensor type into visual data, LiDAR data, IMU data, and environmental sensor data. For visual data (such as RGB-D images), convolutional neural networks (such as ResNet) are used to extract spatial texture features, capturing environmental details such as surface vegetation and rock distribution. For LiDAR point cloud data, point cloud networks (such as PointNet++) are used to extract 3D structural features, quantifying terrain undulations and obstacle contours. For IMU data (acceleration, angular velocity), temporal models (such as LSTM) are used to extract motion state features, characterizing the robot's own motion changes. For environmental sensor data (temperature, humidity, air pressure, etc.), statistical features (mean, variance) are directly calculated to reflect environmental dynamics. Finally, the extracted visual texture features (512 dimensions), LiDAR structural features (256 dimensions), IMU motion features, and environmental statistical features are hierarchically encoded according to sensor type to generate a unified multimodal feature vector.

[0083] During its construction, dedicated networks were strictly selected according to the physical characteristics of the sensors (CNN extracts spatial texture, point cloud network processes 3D structure, and LSTM captures temporal patterns), preserving the specificity of each modality (such as visual focusing on texture and laser focusing on geometric structure) and avoiding feature confusion. In its implementation, redundant information was compressed before layered encoding (such as CNN and point cloud network automatically filtering out invalid background), significantly reducing the feature dimension (compared to the original point cloud / image data), thereby reducing the input dimension of subsequent multimodal fusion and improving inference speed.

[0084] Subsequently, a highly targeted dedicated network can reduce the recognition error of visual texture by CNN, and the quantization accuracy of terrain undulation by point cloud network can reach the centimeter level. At the same time, hierarchical parallel processing (visual / laser / IMU features can be extracted simultaneously) shortens the time compared with serial fusion methods. Hierarchical encoding preserves the correlation of the original sensors, providing structured input for subsequent multimodal fusion based on attention mechanism (such as weighting based on laser features), thereby improving the accuracy of terrain risk prediction.

[0085] Reference Figure 4 As shown, the dynamic weighted fusion step based on the attention mechanism, which outputs a multimodal feature vector to generate a joint environmental feature vector, specifically includes:

[0086] The three-dimensional structural feature vector extracted from the point cloud by lidar is used as the reference vector;

[0087] The attention weight coefficients of each modality feature are generated by calculating the cosine similarity between the spatial texture feature vector extracted by the computer vision sensor, the motion state feature vector extracted by the IMU sensor, and the statistical feature vector extracted by the environmental sensor and the reference vector.

[0088] The multimodal feature vectors are dynamically weighted and fused based on the attention weight coefficient to generate a joint environmental feature vector. The attention weight coefficient is determined by calculating the cosine similarity between the spatial texture feature vector extracted by the visual sensor, the motion state feature vector extracted by the IMU sensor, and the statistical feature vector extracted by the environmental sensor and the three-dimensional structural feature vector extracted by the lidar point cloud.

[0089] The fused joint environmental feature vector is output to the risk prediction module for predicting terrain risk level and weather impact coefficient.

[0090] The three-dimensional structural feature vectors extracted from LiDAR point cloud data through a point cloud network (such as PointNet++) are used as reference vectors. After the LiDAR point cloud is unified to the robot body coordinate system through a coordinate transformation matrix, a 256-dimensional feature vector is extracted layer by layer through the point cloud network as a reference. The reference vector represents the geometric structure of the environment (such as terrain undulation and obstacle outline). Because it is less affected by lighting and weather, its stability is better than other modal features. The combined feature vector more comprehensively represents environmental risks (such as terrain collapse probability and wind speed disturbance).

[0091] After spatiotemporal synchronization, multi-source data is categorized and processed according to sensor type. Each modal feature is independently encoded, and the cosine similarity between the visual sensor (512-dimensional spatial texture features extracted by CNN), the IMU (motion state features extracted by LSTM), and the environmental sensor (statistical features such as mean / variance of temperature and humidity) and the baseline vector is calculated in real time. The multimodal feature vectors are then weighted and summed according to weight coefficients to generate a joint environmental feature vector, which is output to the risk prediction module. The system dynamically allocates weights using cosine similarity, automatically reducing the visual weight in adverse weather conditions, enhancing the contributions of laser and environmental sensors, and reducing feature fusion errors.

[0092] Reference Figure 5 As shown, the process of creating a navigation path based on environmental feature data and generating robot motion control commands specifically includes:

[0093] Receive environmental characteristic data, terrain risk level, and weather impact coefficient output by the environmental understanding module;

[0094] A cost function that integrates environmental risks and motion constraints is constructed, with the robot's current position as the starting point, the target point as the ending point, and the grid map as the search space;

[0095] Nodes with a slope change risk quantification value greater than the maximum slope tolerance or a wind speed disturbance cost greater than the collapse probability safety threshold are removed to generate an initial navigation path. The maximum slope tolerance and collapse probability safety threshold are obtained through statistical analysis of historical experimental data.

[0096] The initial path is optimized locally in real time using the dynamic window method to generate motion control commands that include linear velocity and angular velocity.

[0097] The motion control commands are output to the robot chassis actuator, and the load status parameters fed back by the load management module are received simultaneously to dynamically adjust the path curvature radius threshold and maximum speed.

[0098] The navigation control module receives real-time environmental feature data (including terrain texture, obstacle distribution, etc.), terrain risk level, and weather impact coefficient from the environment understanding module via a hardware bus (such as CAN or EtherCAT). This data is used to construct a cost function that integrates environmental risk and motion constraints. Before the A* algorithm traverses the grid map, thresholds (such as maximum slope tolerance and collapse probability safety threshold) are set based on historical experimental statistics. If a threshold meets the criteria, it is directly eliminated, generating an initial path without high-risk nodes. Using the initial path as a reference, feasible trajectories are sampled in the velocity space (linear velocity v, angular velocity ω). By evaluating the obstacle distance, path fit, and dynamic feasibility of the trajectory, smooth motion control commands are generated. When the motion commands are output to the chassis actuators (such as motor drivers), pressure sensor data from the load management module is received simultaneously for load status feedback and linkage.

[0099] During its use, by pre-screening high-risk nodes (such as areas with a slope change >15° or wind speed >8m / s), the path planning time can be reduced and the avoidance success rate can be improved. At the same time, the attention weight coefficients α, β, γ, δ are dynamically adjusted according to the confidence of multiple sensors (such as automatically increasing the weight of γ in rainy or foggy weather), so that the cost function can adapt to environmental changes in real time, improving the positioning accuracy compared with fixed weight models (such as pure distance optimization). The load state feedback mechanism calculates the robot chassis tilt compensation angle in real time through the torque balance model (such as θ=0.1rad), and adjusts the motion command parameters synchronously, reducing the number of emergency stops during heavy-load transportation and controlling the path tracking error within ±0.1m (traditional methods are prone to overturning because they ignore load offset).

[0100] The construction of the cost function that integrates environmental risks and motion constraints specifically includes:

[0101] ;

[0102] in, To integrate the cost function output value of environmental risk and motion constraints, The cost is the Euclidean distance between nodes. This is the quantified value of slope abrupt change risk in the medium terrain risk level. This represents the probability value of collapse within the medium terrain risk level. The price paid for reduced visibility due to rain and fog, The wind speed disturbance cost is represented by α, β, γ, and δ, which are attention weighting coefficients.

[0103] Reference Figure 6 As shown, the real-time monitoring of the transport load status and the adjustment of the navigation control module parameters based on the load status specifically include:

[0104] The load distribution is detected in real time by a pressure sensor array, and the load center of gravity offset is calculated.

[0105] When the center of gravity offset exceeds the preset safety threshold, a deceleration command is sent to the navigation control module, and the path curvature radius threshold is dynamically reduced. The preset safety threshold is determined by statistically analyzing historical experimental data under different load distributions and combining the current terrain risk level and weather influence coefficient.

[0106] Based on the load weight distribution data, the robot chassis attitude angle is adjusted through the torque balance model, chassis tilt compensation command is generated and synchronized to motion control command.

[0107] A high-density pressure sensor array (such as piezoresistive or piezoelectric sensors) is integrated at the bottom of the robot transport platform, forming a grid-like distribution. Each sensor node collects local pressure values ​​in real time and transmits them to the embedded processor of the load management module via a bus (such as CAN or SPI). A database of preset safety thresholds is established based on historical experimental data, storing the robot's tipping critical values ​​under different load distributions (such as eccentricity and off-center loading). This module combines the terrain risk level (such as slope abrupt change level and collapse probability) and weather influence coefficient (such as wind speed disturbance intensity) output in real time by the environmental understanding module to dynamically calculate the safety threshold under the current operating conditions. For example, when the slope abrupt change risk level is ≥2 and the wind speed disturbance coefficient is >0.5, the safety threshold is tightened to 80% of the historical critical value. The module can then use a built-in rigid body dynamics-based torque balance model, inputting load weight distribution data (generated from raw sensor array data after Kalman filtering and noise reduction), to calculate the torque distribution of each drive wheel of the robot chassis and output chassis tilt compensation angle commands.

[0108] This allows the robot to dynamically adjust thresholds by integrating terrain risk and weather coefficients during use, thereby reducing the risk of tipping over; at the same time, it can actively compensate for chassis tilt to offset tipping torque, reducing the number of emergency stops; and the load management module, navigation control module, and feedback adjustment module form a closed loop.

[0109] Reference Figure 7 As shown, the dynamic adjustment of control parameters based on load status specifically includes:

[0110] Obtain the center of gravity offset and load weight distribution data from the load management module;

[0111] Based on the proportional relationship between the center of gravity offset and the preset safety threshold, the calculation speed is reduced by a scaling factor.

[0112] Calculate the radius of curvature reduction factor based on the magnitude of the centroid offset;

[0113] Based on the load weight distribution data, the tilt compensation angle of the robot chassis is calculated using a torque balance model.

[0114] The speed reduction factor, curvature radius reduction factor, and tilt compensation angle are used as control parameter adjustment quantities and fed back to the navigation control module to adjust the linear velocity, angular velocity, and chassis attitude of the robot motion control commands in real time.

[0115] The preset safety threshold is calculated based on historical experiments and different load distributions to determine the robot's tipping critical value. It is then dynamically adjusted by integrating the terrain risk level (slope change level) and weather influence coefficient (wind speed disturbance intensity) output in real time from the environment understanding module (e.g., the threshold is tightened by 20% when the slope risk is ≥2). When the center of gravity offset exceeds the threshold, the speed reduction factor increases linearly according to the offset exceeding the limit (e.g., when the offset is 120% of the threshold, the speed drops to 80%). The curvature radius reduction factor decreases exponentially with the offset, limiting the centrifugal force of the robot when driving on curves. The above parameters (speed factor, curvature factor, tilt angle) are input into the motion command optimizer of the navigation control module in real time to dynamically adjust the linear velocity, angular velocity, and chassis servo angle, forming a "perception-computation-execution" closed loop.

[0116] The system simultaneously optimizes speed, steering curvature, and chassis attitude, actively counteracts overturning moments through a torque balance model, reducing the number of emergency stops during heavy-duty transport. It incorporates real-time environmental risks (such as collapse probability and wind speed disturbances) into preset safety thresholds, improving stability in complex terrain compared to fixed threshold methods (such as relying solely on historical data). Path tracking errors are controlled within ±0.1m, and the scaling factor is directly generated through linear / nonlinear functions, avoiding complex optimization algorithms (such as model predictive control, MPC). The processing latency on the embedded platform is <10ms, meeting the real-time response requirements in the field. Meanwhile, the tight coupling mechanism between load status and navigation control (such as dynamic reduction of the curvature radius threshold) improves the robot's center of gravity offset tolerance, significantly reducing the risk of rollover on sloping terrain.

[0117] In summary, the advantages of this invention are: it achieves accurate prediction of environmental risks through multi-sensor hierarchical feature extraction and attention-weighted fusion, and dynamically adjusts motion parameters by combining a torque balance model, thereby improving navigation robustness, load offset tolerance, and real-time response capability under complex terrain.

[0118] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A mobile robot control device that takes into account both field navigation patrol and transportation of a load, characterized by, The application relates to a robot navigation system, comprising: a sensor data acquisition module that acquires visual data, lidar data, IMU data, and environmental sensor data in an outdoor environment through visual sensors, lidar, IMUs, and environmental sensors, forming multi-source sensor data; an environmental understanding module connected to the sensor data acquisition module, receiving and processing the multi-source sensor data, fusing the multi-sensor data through a deep learning model, and predicting terrain risks and weather influences based on feature vectors through feature extraction and pattern recognition, and outputting environmental feature data and risk predictions; a navigation control module connected to the environmental understanding module, generating robot motion control instructions based on environmental feature data for navigation paths; a load management module that monitors the transport load state in real time and adjusts the parameters of the navigation control module based on the load state; a feedback adjustment module connected to the navigation control module and the load management module, dynamically adjusting the control parameters according to the load state and feeding back to the navigation control module to optimize and modify the robot motion control instructions; an intelligent detection module connected to the sensor data acquisition module and the environmental understanding module, detecting environmental abnormalities and dangerous events, and outputting alarm signals; The real-time monitoring of the transport load state and the adjustment of the parameters of the navigation control module based on the load state specifically comprises: detecting the load distribution state in real time through a pressure sensor array, and calculating the center of gravity offset; when the center of gravity offset exceeds a preset safety threshold, sending a speed reduction instruction to the navigation control module and dynamically reducing the path curvature radius threshold, the preset safety threshold being determined by statistically analyzing the robot overturning critical value under different load distributions through historical experimental data, and combining the current terrain risk level and the weather influence coefficient; based on the load weight distribution data, adjusting the robot chassis attitude angle through a moment balance model, generating a chassis tilt compensation instruction and synchronously feeding it to the motion control instruction; The dynamic adjustment of the control parameters according to the load state specifically comprises: obtaining the center of gravity offset and the load weight distribution data from the load management module; calculating the speed reduction ratio factor based on the proportion relationship between the center of gravity offset and the preset safety threshold; calculating the curvature radius reduction ratio factor based on the size of the center of gravity offset; calculating the robot chassis tilt compensation angle through a moment balance model based on the load weight distribution data; taking the speed reduction ratio factor, the curvature radius reduction ratio factor, and the tilt compensation angle as the control parameter adjustment amount, feeding them back to the navigation control module, and adjusting the linear speed, angular speed, and chassis attitude of the robot motion control instruction in real time.

2. The mobile robot control device for field navigation patrol and transportation load according to claim 1, wherein The deep learning model fuses multi-sensor data, extracts features, and identifies patterns based on feature vectors to predict terrain risks and weather influences, and outputs environmental feature data and risk predictions, specifically comprising: synchronizing the visual data, lidar data, IMU data, and environmental sensor data in space and time, and unifying the heterogeneous data to the robot body coordinate system through a coordinate transformation matrix; performing hierarchical feature extraction on the data in the robot body coordinate system to obtain feature data; outputting a multi-modal feature vector based on the attention mechanism dynamic weighting fusion step, and generating an environmental joint feature vector; The environment joint feature vector is input into a multi-layer perception to output a terrain risk level and a weather influence coefficient, wherein the terrain risk level includes slope mutation and collapse probability, and the weather influence coefficient includes rain and fog visibility attenuation and wind speed disturbance intensity.

3. The mobile robot control device for field navigation patrol and transportation load according to claim 2, wherein The hierarchical feature extraction on the robot body coordinate system internal data includes the following steps: The multi-source sensor data after spatio-temporal synchronization is classified and processed according to sensor types to obtain visual data, laser radar data, IMU data and environmental sensor data; The spatial texture features of the visual data are extracted by using a convolutional neural network; The three-dimensional structure features of the laser radar data are extracted by using a point cloud network; The motion state features of the IMU data are extracted by using a time series model; The statistical features of the environmental sensor data are extracted; The extracted features are hierarchically encoded according to sensor types to generate a multi-modal feature vector.

4. The mobile robot control device for field navigation patrol and transportation load according to claim 3, wherein The multi-modal feature vector output by the dynamic weighting fusion based on the attention mechanism includes the following steps: The three-dimensional structure feature vector extracted by the laser radar point cloud is taken as a reference vector; The cosine similarity of the spatial texture feature vector extracted by the visual sensor, the motion state feature vector extracted by the IMU sensor and the statistical feature vector extracted by the environmental sensor with the reference vector is calculated to generate the attention weight coefficients of each modal feature; The multi-modal feature vector is dynamically weighted and fused according to the attention weight coefficients to generate an environment joint feature vector, and the attention weight coefficients are determined by calculating the cosine similarity of the spatial texture feature vector extracted by the visual sensor, the motion state feature vector extracted by the IMU sensor and the statistical feature vector extracted by the environmental sensor with the three-dimensional structure feature vector extracted by the laser radar point cloud; The fused environment joint feature vector is output to a risk prediction module for prediction of the terrain risk level and the weather influence coefficient.

5. The mobile robot control device for field navigation patrol and transportation load according to claim 4, wherein The navigation path based on the environment feature data and the generation of the robot motion control instruction include the following steps: The environment feature data, the terrain risk level and the weather influence coefficient output by the environment understanding module are received; A cost function that fuses environment risk and motion constraints is constructed, taking the current position of the robot as the starting point, the target point as the end point and the grid map as the search space; Nodes with a slope mutation risk quantitative value greater than a maximum slope tolerance or a wind speed disturbance cost greater than a collapse probability safety threshold are removed to generate an initial navigation path, wherein the maximum slope tolerance and the collapse probability safety threshold are obtained by statistical analysis of historical experimental data; The initial path is locally optimized in real time by a dynamic window method to generate a motion control instruction containing linear velocity and angular velocity; The motion control instruction is output to the robot chassis actuator, and the load state parameters fed back by the load management module are synchronously received to dynamically adjust the path curvature radius threshold and the maximum speed.

6. The mobile robot control device for field navigation patrol and transportation load according to claim 5, wherein The construction of the cost function that fuses environment risk and motion constraints includes the following steps: ； wherein, is the cost function output value fusing the environmental risk and the motion constraints, is the inter-node Euclidean distance cost, is the slope abruptness risk quantification value in the terrain risk level, is the collapse probability value in the terrain risk level, is the rain and fog visibility attenuation cost, is the wind speed disturbance cost, and a, b, g, d are attention weight coefficients.

Images