Quadruped robot system for forest autonomous inspection

Abstract

The invention relates to the technical field of intelligent sensing and control equipment manufacturing, and discloses a quadruped robot system for forest autonomous inspection, which comprises a multi-dimensional sensing module for synchronously acquiring environmental geometric point cloud, visual images, body postures and joint driving motor electrical parameters and angular velocity data; the terrain impedance resolving module is used for stripping a motor current gravity load component to obtain a dynamic resistance moment, and generating a terrain impedance coefficient by combining gradient and vibration frequency fusion; the energy cost mapping module is used for constructing a metabolic cost map containing nonlinear energy cost based on the terrain impedance coefficient and the gradient, and distinguishing rigid obstacles from passable flexible vegetation; the dynamic energy planning module is used for planning a minimum energy consumption path and triggering return flight based on dynamic return energy consumption predication.According to the method, the terrain impedance is inversed through the motor current, and the technical problems that passable vegetation is misjudged as obstacles and energy consumption estimation fails in the forest soft medium environment through traditional geometric navigation are solved.

Description

Technical Field

[0001] This invention relates to a quadruped robot system for autonomous forest inspection, belonging to the field of intelligent sensing and control equipment manufacturing technology. Background Technology

[0002] Currently, automated data collection and inspection of large-scale unstructured forest areas is an industry trend. Traditional forestry surveys rely on manual foot patrols or fixed-point deployment of infrared cameras. Manual patrols are labor-intensive and inefficient, and pose personal safety risks in complex mountainous terrain or extreme weather. Infrared cameras have limited coverage and data acquisition is delayed, requiring manpower for regular maintenance, which is difficult to meet the needs of real-time, full-area perception of dynamic changes in forest areas.

[0003] Quadruped robot-based mobile platforms are increasingly being introduced into field inspections. Current mainstream solutions utilize general-purpose mobile robot navigation and control architectures, focusing on improving the mechanical reliability and modular maintenance efficiency of the robot body. Hardware optimization is used to adapt to field operation needs. For example, Chinese invention patent CN120697872A discloses a quadruped robot that uses a clamp-type connecting component to achieve rapid assembly and disassembly of the joint motors and limb mechanisms. Precise matching of positioning slots and protrusions solves the problem of cumbersome assembly, improving equipment assembly efficiency and maintenance convenience. Quadruped robot-based mobile platforms are gradually being introduced into field inspections. Current mainstream solutions utilize general-purpose mobile robot navigation and control architectures, relying on LiDAR combined with a global navigation satellite system to perceive the environment and plan paths. The core logic constructs a geometric grid map, treating objects higher than the ground as rigid obstacles, and planning routes based on the principle of shortest geometric distance. The system determines the remaining power status based on battery voltage or a preset power percentage threshold, triggering a return-to-base charging. However, this control strategy based on geometric information and static thresholds has adaptability limitations in complex, unstructured, high-impedance environments like forests.

[0004] Therefore, how to overcome the limitations of single geometric visual perception, improve the robot's ability to identify the physical properties of the environment, and build a path planning and dynamic energy management system based on physical work mechanisms has become the technical problem to be solved by this invention. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A quadruped robot system for autonomous forest inspection, the system comprising a multi-dimensional perception module, a terrain impedance calculation module, an energy cost mapping module, and a dynamic energy planning module:

[0006] The multi-dimensional sensing module is used to simultaneously collect environmental geometric point cloud data, visual image data, fuselage inertial attitude data, phase current data of each joint drive motor, and joint angular velocity data.

[0007] The terrain impedance calculation module is used to extract the gravity load component from the phase current data using the Kalman filter algorithm to obtain the dynamic drag torque, and to perform weighted fusion with the ground slope and vibration frequency represented by the fuselage inertial attitude data, and output the terrain impedance coefficient corresponding to the current position coordinates in real time.

[0008] The energy cost mapping module is used to construct a navigation grid map and execute energy cost assignment rules based on the terrain impedance coefficient: for areas where echoes are detected based on environmental geometric point cloud data but identified as vegetation based on visual image data, and the terrain impedance coefficient is lower than the preset rigidity threshold, a low-damping energy cost is given to allow passage; for areas where the terrain impedance coefficient is higher than the preset softness threshold or the ground slope is higher than the preset angle, the energy cost is nonlinearly increased based on the product of the resistance coefficient and the sine of the slope, and a metabolic cost map is generated.

[0009] The dynamic energy planning module is used to plan the travel path with the minimum cumulative energy cost on the metabolic cost map using a heuristic search algorithm, and calculates the expected return trip energy consumption along the travel path to the resupply base station in real time. The return trip command is triggered only when the difference between the current remaining energy and the expected return trip energy consumption is less than the safety redundancy threshold.

[0010] Preferably, when generating the metabolic cost map, the energy cost mapping module calculates the single-step travel energy cost for each grid cell. Execute the following quantization calculation logic: ,in, and The preset weighting coefficients, This is the baseline power for the robot to walk on a flat, hard surface. The estimated time to pass through this grid cell. This is the dynamic drag torque corresponding to the terrain impedance coefficient. The joint angular velocity, The gravitational potential energy influencing factor. For the overall quality, It is the acceleration due to gravity. The slope is represented by the ground slope. The dynamic energy planning module is used to input the energy cost of a single-step journey as a weight into the cost function of the A algorithm in order to search for the topology path with the lowest global total power consumption.

[0011] Preferably, the terrain impedance calculation module includes: a gravity compensation subunit, used to calculate in real time the static torque components generated by the weight of each joint based on the fuselage inertial attitude data and the robot linkage dynamics model; a disturbance observation subunit, used to map the phase current data into the total electromagnetic torque, and subtract the static torque components and inertial torque components to obtain the dynamic drag torque characterizing the interaction of the surface medium; and a feature fusion subunit, used to time-domain align the amplitude change rate of the dynamic drag torque with the high-frequency vibration components in the fuselage inertial attitude data, and determine the road surface as gravel when a low drag torque and high vibration frequency feature is detected, and determine the road surface as soft mud when a high drag torque and low vibration frequency feature is detected, and generate the terrain impedance coefficient accordingly.

[0012] Preferably, the energy cost mapping module assigns the low-damping energy cost according to the following logic: when the environmental geometric point cloud data indicates that there is a geometric obstacle in front and the visual image data indicates that the texture features of the obstacle match the shrub or tall grass category, the system temporarily designates the area as a potential passage zone; if the instantaneous contact resistance fed back by the terrain impedance calculation module is less than the preset flexible crossing threshold when the quadruped robot contacts the area, the passage status of the area in the navigation grid map is corrected from unreachable to reachable, and the energy cost of the area is set to a preset multiple of the baseline flat land cost, which is positively correlated with the vegetation density.

[0013] Preferably, the dynamic energy planning module includes a dynamic backhaul prediction subunit, which is used to perform the following operations: during the robot's inspection task, based on the currently generated metabolic cost map, it searches backwards at a preset frequency for the minimum energy consumption path from the current coordinate point to the nearest supply base station; it integrates and sums the energy costs of all grid cells on the minimum energy consumption path to obtain the expected backhaul energy consumption; it monitors the real-time state of charge output by the battery management system, and when the difference between the remaining energy corresponding to the real-time state of charge and the expected backhaul energy consumption is lower than the safety redundancy threshold, it interrupts the task and controls the robot to return along the minimum energy consumption path.

[0014] Preferably, the multi-dimensional perception module also includes a visual topology positioning subunit, which is used to take over the positioning function when the satellite positioning signal is lost. The visual topology positioning subunit is used to extract tree trunk textures, rock outlines and forest window light distribution features in the surrounding environment as visual landmarks and construct a topology node map. When the multi-dimensional perception module detects that the satellite signal accuracy factor is lower than the preset standard, this subunit is used to perform feature matching between the currently acquired visual features and key frames in the historical database, calculate the relative pose transformation matrix to correct the odometer drift, until a high-precision satellite signal is reacquired or a resupply base station is reached.

[0015] Preferably, the system also includes a power data base station deployed in the forest area. The power data base station includes a photovoltaic power generation module, a battery pack, a wireless charging interface, and a network relay module. The quadruped robot also includes a power receiving module adapted to the wireless charging interface. When the quadruped robot executes the return command and arrives at the docking range of the power data base station, the quadruped robot uses visual servo to align with the wireless charging interface for non-contact power replenishment, and at the same time uploads the inspection data through the network relay module.

[0016] Preferably, the feature fusion subunit also includes slip detection logic: when a sharp increase in phase current data is detected while the acceleration integral displacement in the fuselage inertial attitude data does not change proportionally, it is determined that the current state is a high slip rate state; in response to the high slip rate state, the terrain impedance calculation module adjusts the terrain impedance coefficient of the current position to infinity and marks the corresponding area as a high-risk energy consumption trap zone in the metabolic cost map, prohibiting subsequent path planning from crossing the area.

[0017] Preferably, when planning a path, the dynamic energy planning module follows the following anisotropic cost rule for handling terrain slope: for the same grid area, the energy cost along the upward slope direction is set to a preset multiple of the energy cost along the downward slope direction; when the planned path involves traversing a slope, the dynamic energy planning module introduces a lateral stability penalty factor, which increases non-linearly with the tangent of the slope angle, in order to suppress the robot from selecting traversing paths that have lower energy consumption but pose a risk of rollover.

[0018] Preferably, the system also includes a data management platform, which includes a task scheduling module and a data analysis module. The task scheduling module is used to divide the forest area into multiple inspection sub-areas based on the geographical information of the forest area, and update the global energy consumption topology network based on the metabolic cost map returned by each quadruped robot. The data analysis module is used to receive the data collected by the quadruped robots, use deep learning algorithms to identify animal and plant traces, and associate the identification results with the terrain impedance coefficient of the collection location to generate a comprehensive forestry electronic map containing biological distribution and terrain accessibility features.

[0019] Compared with the prior art, the beneficial effects of the present invention are:

[0020] 1. In a quadruped robot autonomously patrolling a forest, real-time current and back electromotive force data of the robot's joint drive motors are collected. Kalman filtering is used to separate the dynamic drag torque generated by the interaction of the ground medium. Combined with spatiotemporal registration of geometric point cloud data from visual sensors, the navigation control system subdivides the visual obstacle feature area into impassable rigid obstacles and passable flexible vegetation based on actual contact feedback. The path planning weights are adjusted to plan paths through flexible, high-damping areas such as shrubs or tall grasses while ensuring the stability of the robot's dynamics. This solves the problem of ineffective detours or path planning logic deadlock caused by relying solely on visual geometric information to distinguish the physical properties of obstacles. This improves the autonomous passage capability and task coverage efficiency in unstructured complex terrain.

[0021] 2. The terrain slope, ground flatness, and dielectric impedance coefficient are mapped to a gridded energy passage cost, constructing a metabolic energy cost map that reflects the physical work characteristics of the environment. The navigation controller performs a global path search with minimum cumulative mechanical work, guiding the mobile platform to avoid areas with short geometric distances such as mud and steep slopes, which cause the motor to operate in inefficient and high-energy-consuming conditions. It optimizes energy distribution from the source of physical work for solid, flat, and low-energy-consumption paths, reducing energy consumption per unit mileage of a single task under the boundary condition of limited battery physical capacity, extending the effective operating radius of the equipment, and solving the problem of reduced range caused by the mismatch between the path planning model and the actual energy consumption characteristics of complex terrain.

[0022] 3. Construct a dynamic return-to-base decision-making closed loop that incorporates terrain energy consumption characteristics. Calculate in real time the estimated physical work required to return to the nearest supply base station along the path with the minimum energy cost from the current location. Use this as a dynamic safety threshold and compare it with the current remaining energy in real time. This replaces the open-loop control logic that triggers return-to-base based on static voltage percentage. Automatically compensate for additional energy consumption caused by severe terrain undulations or deteriorating return route conditions. Ensure that the return-to-base energy is accurately reserved to cover the actual physical work demand when encountering extreme conditions such as continuous uphill or high ground resistance. Eliminate the risk of power outages and equipment loss in the field due to insufficient return energy consumption estimation, and ensure the reliability of the fully autonomous unmanned inspection system's cyclic operation. Attached Figure Description

[0023] Figure 1 This is a flowchart of the system control logic for terrain impedance inversion in this invention;

[0024] Figure 2 This is a graph showing the nonlinear effect of terrain slope on the energy cost of passage according to the present invention.

[0025] Figure 3 This is a schematic diagram of the multi-dimensional functional module architecture of the forest autonomous inspection system of the present invention. Detailed Implementation

[0026] To enable those skilled in the art to better understand the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below. 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 should fall within the scope of protection of the present invention.

[0027] This invention discloses a quadruped robot system for autonomous forest patrol. At the architecture level, it mainly consists of four core logical units: a multi-dimensional perception module, a terrain impedance calculation module, an energy cost mapping module, and a dynamic energy planning module. These units interact via a high-speed CAN bus and Ethernet. The top-level logical flow of the system is as follows: the multi-dimensional perception module captures raw data streams containing the robot's kinematic state and environmental geometry in real time; the terrain impedance calculation module decouples the physical properties of the surface medium based on electrical feedback from the motor; the energy cost mapping module integrates the physical properties with the geometric terrain to construct a metabolic cost map representing the mechanical work loss; the dynamic energy planning module generates a globally optimal energy consumption path based on this map and monitors the return energy consumption in a closed loop to trigger a safe return; and the system is designed for the high canopy density and unstructured nature of forest environments. Addressing the engineering challenge of ground surface degradation leading to the failure of single-vision or radar perception, a multi-dimensional perception module constructs a global information acquisition network. This module integrates LiDAR, a binocular depth camera, an inertial measurement unit (IMU), and current sensors and encoders distributed within the actuators of each joint of the quadruped robot. During the system's operation cycle, the LiDAR scans the environment at a frequency of at least 10Hz to generate geometric point cloud data, which is used to construct the basic occupancy grid. The binocular depth camera simultaneously captures RGB images and extracts vegetation texture features. Meanwhile, the underlying control unit simultaneously reads the three-phase current data, rotor angular position, and joint angular velocity data of each joint drive motor at a high-frequency sampling rate of 200Hz to 1kHz. Combined with the triaxial acceleration and Euler angle data output by the IMU, this provides accurate physical quantity inputs for subsequent dynamic parameter identification.

[0028] Traditional geometric navigation often fails to distinguish between impassable rigid rocks and passable flexible shrubs, leading to frequent robot detours or deadlocks. This system addresses this pain point by employing a terrain impedance calculation module to perform ontology-based medium property identification. Internally, this module runs a dynamic observation algorithm based on the Extended Kalman Filter (EKF). This algorithm reads the robot's link dynamics model parameters and calculates the theoretical torque components generated by gravity, Coriolis force, and centrifugal force in real time based on joint angular positions and angular velocities. The measured phase current data is then processed through a torque constant... This is mapped to a total electromagnetic torque, from which the aforementioned theoretical torque components and joint friction torque are subtracted, thereby separating the dynamic drag torque generated solely by the interaction between the foot and the ground. Based on this, the module correlates the amplitude of this dynamic drag torque with the ground slope calculated by the IMU. The system performs weighted fusion of vertical vibration frequencies of the aircraft fuselage to output a normalized terrain impedance coefficient. For example, when high-frequency vibration is detected and the drag torque fluctuates sharply, it is identified as a gravel road and a medium impedance coefficient is output. When the drag torque is stable but its value increases non-linearly with the depth of subsidence, it is identified as soft mud and a high impedance coefficient is output. When the system visually identifies vegetation and the contact drag torque is less than a preset flexible crossing threshold, a low impedance coefficient is output, establishing the passability of the area. The terrain impedance calculation module executes a non-linear normalization model calculation procedure to generate the terrain impedance coefficient. The processor reads the dynamic drag torque. and vertical vibration amplitude of the fuselage The maximum-minimum normalization operation is performed using boundary threshold parameters pre-stored in non-volatile memory, and the dimensionless normalized torque value is output. With normalized vibration value The system follows the formula Calculate the comprehensive stagnation characteristic value ,in , and Preset constants to characterize torque, vibration, and slope weights. To calculate the ground slope angle for the inertial measurement unit, a negative-weighted vibration term is introduced. Distinguish between high-resistance, low-vibration, loose mud conditions and low-resistance, high-vibration, crushed stone conditions.

[0029] Comprehensive blocking characteristic value Input Logistic mapping function Output terrain impedance coefficient in the range of 0 to 1 In the formula The sensitivity coefficient is used to control the steepness of the function curve. The offset of the center for determining medium properties, and the weighting constant. , , and function parameters , Derived from offline calibration procedures: Controlling a quadruped robot to traverse standard peat, gravel, and grassland areas in a controlled experimental environment, and recording various working conditions. and The data distribution is used to iteratively solve for the above parameters using the least squares method, resulting in the model output. To minimize the mean square error of the preset impedance label values ​​for each standard area, the fitted parameter set is solidified as a system operating constant. To address the issue that the shortest geometric path often corresponds to high-energy-consumption steep slopes or high-resistance muddy areas, thus reducing actual driving range, the energy cost mapping module constructs a metabolic cost map containing nonlinear energy consumption characteristics. This module initializes a navigation grid map with a resolution of 0.1 to 0.2 meters. For each grid cell, an energy cost assignment rule is applied based on the terrain impedance coefficient: for areas where LiDAR detects echoes but are visually identified as vegetation and whose terrain impedance coefficient is below a preset rigidity threshold, the system marks them as implicitly passable areas and assigns a low-damping energy cost for passage; for areas where the terrain impedance coefficient is above a preset softness threshold or the ground slope is above a preset angle (e.g., 25 degrees), the passage weight of the grid cell is nonlinearly increased based on the product of the resistance coefficient and the sine of the slope, assigning a single-step passage energy cost to each grid cell. Strictly adhere to the following quantitative calculation logic: ,in, and These are preset dimensionless weighting coefficients used to balance the influence of baseline power consumption and dynamic power consumption. This represents the baseline power measured when the robot walks at a standard speed on a flat, hard surface. The estimated time to pass through this grid cell, calculated based on the current planned speed. The dynamic drag torque is mapped to the terrain impedance coefficient output by the aforementioned terrain impedance calculation module. The average magnitude of the joint angular velocity. This is the gravitational potential energy influence factor, used to correct for differences in work efficiency under different slopes. For the overall quality, It is the acceleration due to gravity. For the ground slope determined by the fuselage attitude data, this mapping mechanism transforms the complex physical work process into an intuitive cost value on the map, enabling the path planning algorithm to naturally tend to select the terrain corridor with the lowest energy consumption.

[0030] Based on the aforementioned metabolic cost map, the dynamic energy planning module is responsible for performing global path optimization and task safety closure. This module uses an improved A* algorithm to calculate the energy cost of a single-step journey. As the edge weight input cost function, a topological path, rather than the geometric shortest path, is searched to connect the current position and the target point with the minimum cumulative mechanical work. Throughout the robot's inspection task, the dynamic backhaul prediction subunit embedded in this module searches in real time at a frequency of 1Hz for the minimum energy consumption path from the current coordinate point to the nearest supply base station, and performs a backhaul analysis on all grids along this path. By integrating and summing, we can obtain the estimated return energy consumption. The system monitors the State of Charge (SOC) and current total remaining energy output by the Battery Management System (BMS) in real time. Once detected If the difference is less than a preset safety redundancy threshold, such as the energy value corresponding to 5% to 10% of the total battery capacity, the system interrupts the current inspection task, triggers a return command, and controls the robot to return to the base station for resupply along the planned minimum energy consumption path. This physically avoids the risk of running out of power due to rough terrain on the return journey. In addition, to address the boundary condition where satellite positioning signals are easily blocked by tree canopy in deep forests, a visual topology positioning subunit is integrated into the multi-dimensional perception module. When the system detects that the satellite signal accuracy factor DOP is lower than a preset standard, such as HDOP being greater than 2.5, this subunit is automatically activated. It uses a binocular camera to extract tree trunk textures, rock outlines, and forest window light distribution features from the surrounding environment as visual landmarks, constructs a local topology node map, and performs feature matching between the currently collected visual feature descriptors and keyframes in the historical database. The relative pose transformation matrix is ​​calculated to correct the cumulative drift of the odometer until the satellite signal is restored or the power data base station with the visual target is reached. In scenarios involving special geological disaster early warning, the feature fusion subunit in the terrain impedance calculation module further includes slip detection logic. The system monitors the phase current change rate of the drive motor and the acceleration integral displacement output by the IMU in real time. When the phase current data is detected to rise sharply by more than 30% of the rated value within 100ms while the corresponding acceleration integral displacement does not change proportionally, it is determined that the current contact surface is in a high slip rate state. In response to this state, the terrain impedance calculation module adjusts the terrain impedance coefficient of the current position to infinity and marks the area as a high-risk energy consumption trap area in the metabolic cost map, prohibiting subsequent path planning from crossing it. This achieves active avoidance of slippery moss or quicksand areas at the algorithm level.

[0031] Example 1: In a typical coniferous forest inspection scenario involving slippery steep slopes and dense shrubs, conventional navigation logic often leads the robot to take long, geometrically safe detours because it cannot distinguish between visual obstacles and actual physical obstacles. While these paths are geometrically collision-free, they require frequent traversal of high-energy-consuming areas with slopes exceeding 25 degrees, causing the motors to operate in a high-torque, low-efficiency range for extended periods, resulting in a significant reduction in actual driving range. This example addresses this by using a multi-dimensional perception module and a terrain impedance calculation module in tandem. A basic geometric map is constructed using LiDAR, while the joint motor phase current data is monitored in real time. When the robot attempts to traverse what appears to be a dense shrub area, the terrain impedance calculation module separates the gravity and inertial terms from the current feedback. The calculated dynamic resistance torque is below a preset rigidity threshold, indicating that the area is a flexible, passable medium. The energy cost mapping module then marks this area as an implicitly passable zone and utilizes... The formula calculates energy costs by incorporating the topographic impedance coefficient and the sine of the ground slope to quantify the mechanical work loss per unit distance. For high-energy-consuming areas with steep slopes, even with short geometric distances, the calculated energy cost is higher. The value will also increase. The dynamic energy planning module uses the A* algorithm to search for the path with the minimum cumulative energy cost on the metabolic cost map. This path automatically avoids energy consumption traps with high slopes and chooses to traverse shrub corridors with less resistance and gentle slopes. This not only resolves the contradiction between geometric safety and energy efficiency, but also achieves adaptive traversal of unstructured terrain through physical energy consumption perception. During the execution of the task, the system continuously reverse-calculates the expected return trip energy consumption to the base station along the optimal energy consumption path. and real-time comparison with the current remaining energy. The system compares the two values, and when the difference reaches the safety redundancy threshold, it triggers a forced return command to ensure that the robot still has enough energy to return to the base station in extreme terrain.

[0032] Example 2: To verify the energy consumption optimization and maneuverability performance of the quadruped robot system of the present invention in complex forest terrain, a comprehensive test platform with various surface media and slope variations was constructed. The platform was 200 meters long and consisted of a flat, hard ground area, a variable slope area of ​​20 to 35 degrees, and a soft medium area covered with a mixture of fallen leaves and peat moss with a thickness of 20 to 40 centimeters. The test object was a quadruped robot prototype equipped with the control system of the present invention. The overall mass of the robot was... Weighing 15kg, with a single-leg joint drive motor rated torque of 25Nm, a control group was set up where terrain impedance calculation and energy cost mapping modules were not enabled, and A* path planning relied solely on the geometric occupancy grid map. During the experiment, baseline power consumption was calibrated on a flat, hard surface. The robot traversed the hard area at a speed of 0.5m / s, and the average input power of each joint motor was recorded to determine... Reference power in the calculation formula With a power output of 180W, the robot enters a variable slope zone. Under a slope set at 30 degrees, the prototype utilizes an onboard IMU to perceive the fuselage pitch angle in real time. The gravity load component is calculated using the Kalman filter algorithm, and the system is based on the formula... Calculate the energy cost of the current grid, where the weighting coefficients are... and The gravitational potential energy influence factors were set to 0.6 and 0.4 respectively. Set to 1.2, to address the challenges caused by steep slopes. The dynamic energy planning module automatically planned a zigzag path along contour lines to climb the slope, rather than forcibly climbing in a straight line. In contrast, the control group robot still chose a straight path, causing the hind leg joint motors to remain saturated with current in the middle of the climb, and shutting down due to overheat protection after several attempts. Next, a passability test was conducted in a soft medium area. When the robot came into contact with the mixed soft ground, the terrain impedance calculation module detected that the impact torque at the moment the foot touched the ground was small, and the support phase resistance torque was small. The system exhibits a non-linear increase in the depth of depression, thereby identifying the high impedance characteristics of the region and dynamically increasing the passage cost in the metabolic cost map. Table 1 shows a comparison of key performance data under different path selections in the region.

[0033] Table 1: Comparison of Passage Performance in Soft Media Areas

[0034]

[0035] As shown in Table 1, the control group, lacking awareness of the physical properties of the medium, forcibly traversed the center of the soft area, resulting in severe foot sinking and energy consumption per unit distance reaching as high as 980 J / m. Ultimately, the task failed due to insufficient torque. In contrast, the sample group of this invention, by constructing a metabolic cost map, identified a channel in relatively firm hard soil at the edge of the soft area, although the geometric distance was longer (increased by 15%). The planned path, although detouring, had an energy consumption of only 420 J / m per unit distance. It not only successfully traversed the area but also reduced overall energy consumption by more than 57%. Furthermore, regarding the verification of the return trip energy consumption prediction function, when the robot's remaining battery power was 30%, a return command was manually triggered. The system, based on the constructed metabolic cost map, calculated the estimated return trip energy consumption from the current position to the starting point through inverse integration. It is 1.2 J. During the actual return journey, although it encountered sudden gusts of wind resistance as an uncontrollable environmental disturbance, the system, relying on the reserved safety redundancy threshold set at 10% of the total energy, finally arrived safely at the starting point with 3% of the remaining power. This verifies the stability and reliability of the dynamic energy planning module in dealing with non-ideal environmental factors. This invention solves the technical problems of energy consumption estimation failure and poor passability of traditional geometric navigation in complex forest environments by introducing a physical mapping mechanism of terrain impedance and energy cost.

[0036] Example 3: This example combines Figures 1 to 3 A description of a quadruped robot system for autonomous forest patrol, such as... Figure 1As shown, the control logic flow of this system is as follows: the multi-dimensional sensing module receives environmental geometric point cloud, phase current and angular velocity, and fuselage inertial attitude signals. After synchronous processing, it outputs the raw data stream to the next stage. The terrain impedance calculation module receives this data stream, uses the Kalman filter algorithm to remove the gravity load component to obtain the dynamic drag torque, and combines the fusion characteristics of slope and vibration frequency to output the terrain impedance coefficient. Then, the energy cost mapping module constructs a metabolic cost map based on the terrain impedance coefficient and distinguishes between rigid obstacles and passable flexible vegetation. Finally, the dynamic energy planning module performs minimum energy consumption path planning and return decision based on the map. After generating dynamic return energy consumption prediction data, it outputs the minimum energy consumption path command or return command.

[0037] like Figure 2 As shown, the horizontal axis of this energy cost curve represents the slope in degrees (°), and the vertical axis represents the energy cost in J (J). The solid line in the graph represents the energy cost in the upward slope direction, showing a non-linear upward trend as the slope value increases. The dashed line represents the energy cost in the downward slope direction, with a relatively gentle increase in energy cost with slope change. Furthermore, a preset slope threshold line corresponding to 25° is set in the graph. This threshold line intersects with the upward slope energy cost curve, indicating that when the uphill angle exceeds this threshold, the system's energy cost will be higher than the preset baseline level. Figure 3 As shown, the main branch points to the forest autonomous inspection quadruped robot system. The upper left branch is the multi-dimensional perception module, which includes environmental geometric point cloud acquisition, motor phase current and angular velocity monitoring, and visual topology localization under GPS-free conditions. The lower left branch is the terrain impedance calculation module, which covers Kalman filtering to remove gravity terms, obtaining dynamic drag torque, and slippage detection logic. The upper right branch is the energy cost mapping module, which involves distinguishing between rigid and flexible vegetation, constructing metabolic cost maps, and nonlinearly increasing the cost of high-resistivity areas. The lower right branch is the dynamic energy planning module, which integrates return safety commands, dynamic return energy consumption prediction, and minimum cumulative energy consumption path planning based on the A* algorithm.

[0038] Example 4: In the verification of the energy management strategy for a quadruped robot under extreme climatic conditions, a comprehensive test scenario simulating high-altitude cold and strong wind environments was constructed. The ambient temperature in this scenario was set to -20°C. The system introduces lateral gusts with wind speeds up to 15 m / s to expose the suppressive effect of low temperatures on battery discharge performance and the nonlinear impact of strong wind resistance on the power consumption of the joint motor. For these extreme conditions, this embodiment performs targeted adaptive parameter correction and boundary condition completion on the general energy cost mapping model in the original document. To address the issue of effective capacity decay caused by increased battery internal resistance in low-temperature environments, the system introduces a temperature compensation coefficient based on the Arrhenius equation in the dynamic energy planning module. This coefficient corrects the state of charge (SOC) reported by the battery management system (BMS) in real time, adjusting the nominal remaining energy. Adjusted to actual usable energy Specifically, the discharge curves of the battery at different temperatures were measured through offline calibration experiments, and a temperature-capacity mapping table was established. (The last sentence appears to be incomplete and possibly refers to a specific temperature range.) Under operating conditions, the system automatically assigns the data based on this mapping table. Revised to This is 65% of the original range, avoiding the risk of overestimating the driving range due to low temperatures.

[0039] Secondly, to address the dynamic impact of strong crosswinds on the robot's drag, the terrain impedance calculation module is configured to further integrate the unbalanced components of the body's lateral acceleration and joint torque. When the IMU detects continuous lateral acceleration and a corresponding increase in lateral joint torque, the system determines that it has encountered strong crosswind interference. At this point, the energy cost mapping module calculates the energy cost for a single step of travel. At that time, an additional wind resistance power consumption term is introduced. This parameter is calculated by estimating the aerodynamic drag coefficient in real time, which is proportional to the cube of the relative wind speed. When encountering a 15 m / s crosswind, this mechanism increases the grid cost of the windward path, enabling the planning algorithm to automatically search for the leeward side or a low-drag attitude, thus achieving proactive adaptation to environmental disturbances at the algorithm level. Finally, in the above-mentioned high-altitude and strong wind scenario, by integrating temperature compensation and wind resistance sensing mechanisms, the system successfully predicted the additional energy consumption caused by the harsh environment and the expected return energy consumption. The calculation accuracy is improved by 28%. When the actual remaining energy reaches the temperature-corrected safety threshold, the system triggers the return command in advance to ensure that the robot can safely return to the base station under extreme conditions where the battery output power is limited and the travel resistance is doubled, thus avoiding power outage and freezing accidents in the field caused by parameter black box.

[0040] Example 5: To ensure that the dynamic energy planning module maintains the expected prediction accuracy in forest environments with different latitudes and vegetation types, the system incorporates a standardized offline parameter calibration procedure to generate an energy consumption coefficient lookup table adapted to specific forest areas. This procedure controls the quadruped robot to move at a gradient speed of 0.2 m / s to 0.8 m / s on a horizontal hard reference surface, and the reference power is determined through linear regression fitting. and speed weighting coefficient Based on the baseline values, the robot was placed on a standard test ramp with a known inclination angle and a simulated soft soil trough with calibrated shear strength to perform climbing and traversing actions, respectively. The system recorded the difference between the actual motor power consumption and the theoretical mechanical work done, and used the least squares method to iteratively solve for the gravitational potential energy influence factor. With dynamic resistance torque weighting coefficient The optimal solution is obtained by encapsulating the above calibration data into an unmodifiable configuration file and storing it in the non-volatile memory of the underlying controller, which serves as the sole arithmetic benchmark for the subsequent generation of the metabolic cost map.

[0041] To address the cold start initialization issue of the multi-dimensional perception module in unexplored areas, the system executes a pre-deployment calibration process to establish a local environmental baseline. Before formally performing the inspection task, the quadruped robot performs a self-check of its three-axis posture in place and a closed-loop alignment walk with a radius of 5 meters. During this process, the terrain impedance calculation module collects the current noise baseline of the unloaded swing phase and the support phase, automatically compensating for the sensor zero-point deviation caused by temperature drift. Simultaneously, the visual topology positioning subunit collects the static texture features of the surrounding environment and the light distribution pattern of the forest window at high frequency during this process, constructing an initial topology node map containing at least 50 keyframes. This map is used as the absolute reference coordinate system for subsequent odometry drift correction, thereby ensuring that the system has readily available relative navigation capabilities in satellite signal blind zones.

[0042] Example 6: In this example, the system executes a standardized multi-sensor joint calibration and initialization procedure. After the robot leaves the factory or undergoes major repairs, the robot body is placed on a precision optical platform with a levelness error of less than 0.1 degrees. IMU zero-bias calibration is performed. The system collects raw data streams from the accelerometer and gyroscope in a stationary state for no less than 300 seconds. The zero-bias stability and random walk coefficients of each axis are calculated through Allan variance analysis, and the error compensation matrix inside the IMU is updated accordingly. External parameter joint calibration of the lidar and binocular camera is performed within a range of 2 to 5 meters in front of the robot. A standard checkerboard calibration board with known geometric features is placed, and the robot is controlled to perform pitch and yaw movements from multiple angles. The system simultaneously acquires laser point cloud data and visual images, uses the PnP algorithm to calculate the relative pose of the camera and the calibration board, and determines the position of the calibration board in the radar coordinate system based on the point cloud feature extraction algorithm. By constructing an objective function that minimizes the reprojection error, the rotation and translation matrix of the laser radar coordinate system relative to the camera coordinate system is iteratively solved until the average reprojection error is less than 1 pixel. The calibration result is solidified into the system parameter file and serves as the basis for the reference space transformation of multi-source data fusion.

[0043] To address the accuracy issue of battery state estimation in the dynamic energy planning module, the system incorporates a self-learning calibration process for battery capacity. When the robot performs a charging task, once the battery voltage reaches the full charge threshold, the system automatically records the current cumulative charge and uses it as the baseline capacity for this cycle. During discharge, the system integrates current data in real time and corrects the estimated state of charge (SOC) value by combining it with the open-circuit voltage curve. When the discharge reaches the cutoff voltage, the system compares the actual discharged capacity with the baseline capacity and uses a Kalman filter algorithm to dynamically update the battery's aging factor and internal resistance model parameters. This ensures that the energy management strategy is always based on the battery's current true state of health (SOH), thereby avoiding deviations in range prediction caused by battery aging.

[0044] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0045] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A quadruped robot system for autonomous forest patrol, characterized in that, The system includes a multi-dimensional sensing module, a terrain impedance calculation module, an energy cost mapping module, and a dynamic energy planning module. The multi-dimensional sensing module is used to simultaneously collect environmental geometric point cloud data, visual image data, fuselage inertial attitude data, phase current data of each joint drive motor, and joint angular velocity data. The terrain impedance calculation module is used to extract the gravity load component from the phase current data using the Kalman filter algorithm to obtain the dynamic drag torque, and to perform weighted fusion with the ground slope and vibration frequency represented by the fuselage inertial attitude data, and output the terrain impedance coefficient corresponding to the current position coordinates in real time. The energy cost mapping module is used to construct a navigation grid map and execute energy cost assignment rules based on the terrain impedance coefficient: for areas where echoes are detected based on environmental geometric point cloud data but identified as vegetation based on visual image data, and the terrain impedance coefficient is lower than the preset rigidity threshold, the energy cost is set as a low-damping energy cost that allows passage; for areas where the terrain impedance coefficient is higher than the preset softness threshold or the ground slope is higher than the preset angle, the energy cost is nonlinearly increased based on the product of the resistance coefficient and the sine of the slope, and a metabolic cost map is generated. The dynamic energy planning module is used to plan the travel path with the minimum cumulative energy cost on the metabolic cost map using a heuristic search algorithm, and calculates the expected return trip energy consumption along the travel path to the resupply base station in real time. The return trip command is triggered only when the difference between the current remaining energy and the expected return trip energy consumption is less than the safety redundancy threshold.

2. The quadruped robot system for autonomous forest patrol according to claim 1, characterized in that, When generating the metabolic cost map, the energy cost mapping module calculates the single-step travel energy cost for each grid cell. Execute the following quantization calculation logic: ,in, and The preset weighting coefficients, This is the baseline power for the robot to walk on a flat, hard surface. The estimated time to pass through this grid cell. This is the dynamic drag torque corresponding to the terrain impedance coefficient. The joint angular velocity, The gravitational potential energy influencing factor. For the overall quality, It is the acceleration due to gravity. The slope is represented by the ground slope. The dynamic energy planning module is used to input the energy cost of a single-step journey as a weight into the cost function of the A* algorithm in order to search for the topology path with the lowest global total power consumption.

3. A quadruped robot system for autonomous forest patrol according to claim 1, characterized in that, The terrain impedance calculation module includes: a gravity compensation subunit, used to calculate the static torque components of each joint due to its own gravity in real time based on the fuselage inertial attitude data and the robot linkage dynamics model; a disturbance observation subunit, used to map the phase current data into the total electromagnetic torque and subtract the static torque components and inertial torque components to obtain the dynamic drag torque characterizing the interaction of the surface medium; and a feature fusion subunit, used to time-domain align the amplitude change rate of the dynamic drag torque with the high-frequency vibration components in the fuselage inertial attitude data. When a low drag torque and high vibration frequency feature is detected, it is determined to be a gravel road surface; when a high drag torque and low vibration frequency feature is detected, it is determined to be a soft mud road surface, and the terrain impedance coefficient is generated accordingly.

4. A quadruped robot system for autonomous forest patrol according to claim 1, characterized in that, The energy cost mapping module assigns low-damping energy cost according to the following logic: when the environmental geometric point cloud data indicates that there is a geometric obstacle in front and the visual image data indicates that the texture features of the obstacle match the shrub or tall grass category, the system temporarily designates the area as a potential passage area. If the instantaneous contact resistance reported by the terrain impedance calculation module is less than the preset flexible crossing threshold when the quadruped robot comes into contact with the area, the accessibility status of the area in the navigation grid map is corrected from unreachable to reachable, and the energy cost of the area is set to a preset multiple of the baseline flat land cost, which is positively correlated with the vegetation density.

5. A quadruped robot system for autonomous forest patrol according to claim 1, characterized in that, The dynamic energy planning module includes a dynamic backhaul prediction subunit, which performs the following operations: during the robot's inspection task, it searches backwards from the current coordinate point to the nearest supply base station based on the currently generated metabolic cost map at a preset frequency; it integrates and sums the energy costs of all grid cells on the minimum energy consumption path to obtain the expected backhaul energy consumption; it monitors the real-time state of charge output by the battery management system, and when the difference between the remaining energy corresponding to the real-time state of charge and the expected backhaul energy consumption is lower than the safety redundancy threshold, it interrupts the task and controls the robot to return along the minimum energy consumption path.

6. A quadruped robot system for autonomous forest patrol according to claim 1, characterized in that, The multidimensional perception module also includes a visual topology positioning subunit, which takes over the positioning function when the satellite positioning signal is lost. The visual topology positioning subunit is used to extract tree trunk textures, rock outlines and forest window light distribution features in the surrounding environment as visual landmarks and construct a topology node map. When the multidimensional perception module detects that the satellite signal accuracy factor is lower than the preset standard, the visual topology positioning subunit is used to perform feature matching between the currently acquired visual features and key frames in the historical database, calculate the relative pose transformation matrix to correct the odometer drift, until a high-precision satellite signal is reacquired or a resupply base station is reached.

7. A quadruped robot system for autonomous forest patrol according to claim 1, characterized in that, The system also includes a power data base station deployed in the forest area. The power data base station includes a photovoltaic power generation module, a battery pack, a wireless charging interface, and a network relay module. The quadruped robot also includes a power receiving module adapted to the wireless charging interface. When the quadruped robot executes the return command and arrives at the docking range of the power data base station, the quadruped robot uses visual servo to align with the wireless charging interface for non-contact power replenishment, and at the same time uploads the inspection data through the network relay module.

8. A quadruped robot system for autonomous forest patrol according to claim 3, characterized in that, The feature fusion subunit also includes slip detection logic: when a sharp increase in phase current data is detected while the acceleration integral displacement in the fuselage inertial attitude data does not change proportionally, it is determined that the current state is in a high slip rate state; in response to this high slip rate state, the terrain impedance calculation module adjusts the terrain impedance coefficient of the current position to infinity and marks the corresponding area as a high-risk energy consumption trap zone in the metabolic cost map, prohibiting subsequent path planning from crossing the area.

9. A quadruped robot system for autonomous forest patrol according to claim 1, characterized in that, When planning a path, the dynamic energy planning module follows the following anisotropic cost rule for handling terrain slope: For the same grid area, the energy cost along the upward slope direction is set to a preset multiple of the energy cost along the downward slope direction; when the planned path involves traversing a slope, the dynamic energy planning module introduces a lateral stability penalty factor, which increases non-linearly with the tangent of the slope angle.

10. A quadruped robot system for autonomous forest patrol according to claim 1, characterized in that, The system also includes a data management platform, which comprises a task scheduling module and a data analysis module. The task scheduling module is used to divide the forest area into multiple inspection sub-regions based on the geographical information of the forest area, and to update the global energy consumption topology network based on the metabolic cost map returned by each quadruped robot. The data analysis module is used to receive the data collected by the quadruped robots, use deep learning algorithms to identify animal and plant traces, and correlate the identification results with the topographic impedance coefficient of the collection location to generate a comprehensive forestry electronic map containing biological distribution and topographic accessibility features.

Images