A method for guiding the operation of a shovel based on multi-sensor data fusion

Abstract

The application discloses a kind of based on multi-sensor data fusion's excavator operation guiding method, specifically related to engineering machinery technical field, by combining construction design data with field measured topographic state, utilize the topographic information of target operation unit of multi-source sensor data acquisition, and calculate perception credibility and topographic uncertainty, generate residual operation quantity interval and credible operation residual diagram.Based on the current pose of excavator and operation constraint generates multiple candidate bucket action primitives, and calculates confidence weighted residual reduction, predicted overbreak, low credible coverage and constraint conflict amount by bucket sweep matching evaluation, filters and optimizes target action primitive.After action execution, according to measured topographic change and residual operation quantity, revise evaluation parameter, realize closed loop dynamic update.This method can realize high-precision, intelligent excavator operation guiding in complex construction environment, improve construction efficiency and precision.

Description

Technical Field

[0001] This invention relates to the field of engineering machinery technology, and more specifically, to an excavator operation guidance method based on multi-sensor data fusion. Background Technology

[0002] With the continuous expansion of civil engineering construction, excavators are increasingly used in infrastructure construction, mining excavation, and foundation pit construction. However, traditional excavator operations rely on operator experience for work planning and control, resulting in low efficiency, difficulty in ensuring construction accuracy, and frequent over-excavation or under-excavation. Especially in complex terrain, slope areas, and confined spaces, limited visibility, incomplete terrain information, and operational uncertainties make it difficult for manual judgment to accurately select the optimal bucket action, leading to inconsistent construction quality. Furthermore, existing operational support systems typically rely on a single sensor or simple depth map comparison with the design model, lacking comprehensive consideration of sensor data uncertainties and actual terrain changes, making it difficult to form continuous, closed-loop operational optimization control.

[0003] Furthermore, traditional methods often fail to effectively quantify the potential operational benefits and risks of each candidate bucket action during motion planning. For example, the evaluation of the matching between the bucket coverage area and the target work unit is not refined enough, and over-digging risks, low-reliability coverage, and constraint conflicts are usually not quantifiable. This leads to the selection of optimal actions relying on empirical rules and lacking scientific basis. This not only increases construction deviations but also limits the excavator's ability to achieve high-precision and high-efficiency operations in complex environments.

[0004] Therefore, how to accurately acquire terrain information, quantify operational uncertainties, and scientifically guide bucket movements in a multi-source sensing environment has become a key technical challenge in improving the operational accuracy and construction efficiency of excavators. Summary of the Invention

[0005] To overcome the aforementioned deficiencies in the prior art, this invention provides an excavator operation guidance method based on multi-sensor data fusion. This method acquires the terrain state of the target operation unit through multi-source sensor fusion, quantifies the perception reliability and terrain uncertainty, generates a residual operation range and constructs a reliable operation residual map, and combines bucket sweeping matching evaluation to screen candidate actions and perform closed-loop correction, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for guiding excavator operation based on multi-sensor data fusion includes: converting construction design data into a set of target operation units in the on-site construction coordinate system; acquiring the measured terrain state corresponding to each target operation unit based on multi-source sensing data, determining the perception reliability of the measured terrain state, and propagating terrain uncertainty based on the perception reliability and the observation dispersion of the corresponding measured terrain state; combining the design terrain state, measured terrain state, and terrain uncertainty of the corresponding target operation unit to generate a reliable operation residual map containing the residual workload interval and constraint state; and based on the current excavator operation... Multiple candidate bucket motion primitives are generated based on the previous pose and operational constraints. The confidence-weighted residual cut, expected overcut, low-confidence coverage, and constraint conflict of each candidate bucket motion primitive are calculated based on the coverage relationship between the bucket sweep area of ​​each candidate bucket motion primitive and the reliable operational residual map. The candidate bucket motion primitives are screened and optimized based on the calculation results to determine the target bucket motion primitive and output its guiding parameters. After the target bucket motion primitive is executed, the evaluation parameters used for screening and optimization are corrected based on the updated measured terrain conditions and the changes in residual workload before and after execution.

[0007] As a further aspect of the present invention, the step of converting construction design data into a set of target work units in the on-site construction coordinate system includes: dividing construction drawings, three-dimensional design models, or input work parameters into multiple target work units; establishing a coordinate transformation relationship between the design coordinate system and the on-site construction coordinate system; and mapping each target work unit to the on-site construction coordinate system; wherein each target work unit is associated with at least two of the following: target elevation, target slope, target depth, allowable error, and work boundary.

[0008] As a further aspect of the present invention, the terrain uncertainty is obtained by propagating the perceived credibility, including: determining the perceived credibility based on at least three of the following: positioning solution quality, attitude stabilization quality, point cloud quality, image recognition quality, time synchronization quality, and occlusion status; and determining the elevation uncertainty and slope uncertainty of the corresponding target operation unit by combining the observation dispersion of the corresponding measured terrain status.

[0009] As a further aspect of the present invention, the residual workload range is determined based on the target elevation, measured elevation, elevation uncertainty, area of ​​the target work unit, and allowable error of the target work unit; when the upper limit of the residual workload range is not greater than the preset completion threshold, the corresponding target work unit is marked as a completed unit.

[0010] As a further aspect of the present invention, the generation of multiple candidate bucket action elements includes: determining the reachable range of the bucket teeth tip based on the current position of the excavator, the structural parameters of the working device, and the joint limits; selecting target work units from the reliable work residual map whose lower limit of the residual work volume interval is greater than a preset threshold and whose perceived reliability is not lower than a preset reliability threshold as candidate work units; and generating multiple candidate bucket action elements based on the boundary of the candidate work unit, the target slope, the current machine position, and the reachable range of the bucket teeth tip.

[0011] As a further aspect of the present invention, each candidate bucket action element includes at least an entry point, an exit point, a cutting depth, an entry posture, and a tooth tip trajectory; the bucket sweeping area is determined based on the posture envelope formed by the tooth tip trajectory, the bucket width, and the changes in the bucket posture.

[0012] As a further aspect of the present invention, the calculation of the confidence-weighted residual reduction, expected over-excavation, low-confidence coverage, and constraint conflict of each candidate bucket action element includes: determining the lower and upper limits of the confidence-weighted residual reduction based on the overlap relationship between the bucket sweeping area and the covered target work unit, the residual work volume range, the perceived confidence level, and the effective coefficient of the bucket sweeping; determining the upper limit of the expected over-excavation based on the portion of the bucket sweeping area that exceeds the design terrain boundary or exceeds the allowable error range; determining the low-confidence coverage based on the proportion of the covered target work unit with a perceived confidence level lower than a preset threshold; and determining the constraint conflict based on the overlap ratio between the bucket sweeping area and the restricted area, obstacle area, or work device constraint boundary.

[0013] As a further aspect of the present invention, the step of screening and selecting candidate bucket action elements based on the calculation results includes: firstly, eliminating candidate bucket action elements whose constraint conflict exceeds a first threshold or whose low confidence coverage exceeds a second threshold; then, prioritizing the selection of candidate bucket action elements with the largest lower limit of confidence-weighted residual reduction among the remaining candidate bucket action elements; when multiple candidate bucket action elements have the same lower limit of confidence-weighted residual reduction, sequentially selecting the candidate bucket action elements with the smallest upper limit of expected over-excavation and the smallest action path cost.

[0014] As a further aspect of the present invention, the step of correcting the evaluation parameters based on the updated measured terrain state and the change in residual workload before and after execution includes: obtaining the measured terrain state after the execution of the target bucket action element; when the measured residual reduction amount after execution deviates from the predicted residual reduction amount range before execution, at least one of the soil correction coefficient, bucket sweeping effective coefficient and uncertainty propagation coefficient is corrected according to the degree of deviation.

[0015] As a further aspect of the present invention, after correcting the evaluation parameters, the confidence-weighted residual cutting amount, expected overcut amount, low confidence coverage amount, and constraint conflict amount of the candidate bucket action primitives that have not been executed are recalculated; when the optimal candidate bucket action primitives obtained by recalculation are inconsistent with the target bucket action primitives to be executed determined before the update, the entry point, exit point, cutting depth, and bucket attitude guidance parameters after the switch are output.

[0016] The technical effects and advantages of this invention, a method for guiding excavator operations based on multi-sensor data fusion, are as follows: This invention achieves intelligent guidance for excavator operations by introducing multi-source sensing data fusion and operational uncertainty quantification. Furthermore, by assessing the reliability of the terrain state of the target work unit and propagating it to uncertainty, this invention comprehensively reflects the impact of measurement errors, occlusion, and complex local terrain on the operation, thus improving the scientific nature of the operation planning. In the generation and evaluation process of candidate bucket action elements, this invention simultaneously considers residual workload, reliability, over-excavation risk, and constraints, providing a quantitative basis for action selection and ensuring the accuracy and safety of the construction process.

[0017] Furthermore, this invention establishes a closed-loop correction mechanism, which automatically adjusts the evaluation parameters of subsequent actions by acquiring terrain feedback after the action is executed in real time, thereby achieving dynamic optimization in continuous excavation scenarios. This mechanism not only reduces the reliance on manual intervention, but also effectively reduces rework caused by over-excavation or under-excavation, thus improving construction efficiency.

[0018] This invention enables high-precision and high-reliability operation guidance in complex construction environments, providing operable and quantifiable technical support for intelligent construction by excavators, and also providing an scalable reference framework for the future automated control of construction machinery. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the excavator operation guidance method based on multi-sensor data fusion according to the present invention.

[0020] Figure 2 This is a schematic diagram of the residual workload interval-bucket sweeping matching evaluation model based on the propagation of perceived uncertainty in this invention.

[0021] Figure 2 In the diagram, 21 represents the target work unit, 22 represents the low-confidence unit, 23 represents the restricted area, 24 represents the tooth tip trajectory, 25 represents the bucket sweeping area, 26 represents the entry point, 27 represents the exit point, 28 represents the confidence-weighted residual reduction, 29 represents the expected over-excavation, 30 represents the low-confidence coverage, 31 represents the constraint conflict, and 32 represents the target bucket action primitive. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1

[0023] like Figure 1 As shown, this embodiment is applied to a layered excavation scenario for a foundation pit. The area to be constructed is a strip-shaped working surface with a length of 18m and a width of 6m. The designed bottom elevation is -2.400m, the allowable elevation error is ±0.030m, the longitudinal target slope is 0.8%, and the transverse target slope is 0.5%. Slope constraints and no-entry boundaries are set near the slope side. Before construction begins, the construction design data is imported into the on-board computing unit and divided into 432 target working units according to a grid size of 0.5m × 0.5m. Each target working unit is associated with target elevation, target slope, target depth, allowable error, and working boundary information. The on-site construction coordinate system is established through three reference points, located at the two ends and the middle control point of the working area. The coordinate transformation relationship between the design coordinate system and the on-site construction coordinate system is established using a combination of planar rigid transformation and elevation offset, so that each target working unit in the design model can be mapped to its corresponding position in the on-site construction coordinate system. After mapping is completed, the construction design data is converted into a set of target work units that can be directly used for terrain comparison, work evaluation, and dynamic updates.

[0024] The multi-source sensing data used in this embodiment includes positioning data, attitude data, point cloud data, and image data. Positioning data is provided by a dual-antenna high-precision positioning receiver and a body position calculation module; attitude data is provided by an inertial measurement unit, a working device angle sensor, and a rotation angle detection module; point cloud data is provided by an airborne laser scanning unit; and image data is provided by a forward-facing binocular vision unit. The vehicle-mounted computing unit refreshes data every 200ms, with the body pose and working device attitude updated at 50ms intervals, point cloud data updated at 10Hz, and image data updated at 15Hz. After installation, each sensor undergoes extrinsic parameter calibration and time synchronization calibration, and is uniformly projected onto the on-site construction coordinate system. For any target work unit, the vehicle-mounted computing unit fuses the point cloud elevation samples, local slope fitting results, image boundary information, body pose, and working device attitude of the corresponding area within the current evaluation period to obtain the measured terrain state corresponding to the target work unit. The measured terrain conditions include at least the measured elevation, measured slope, point cloud height dispersion, local surface normal dispersion, image sharpness, update time, and occlusion status.

[0025] In this embodiment, perceived reliability is used as the core input for subsequent uncertainty propagation and action optimization. For any target work unit i, the positioning solution quality score q is first calculated. g,i Attitude stability quality score q a,i Point cloud quality score q l,i Image recognition quality score q v,i Time synchronization quality score q t,i And the score q for occlusion status o,i Each score is normalized to between 0 and 1, with higher values ​​indicating more reliable measured terrain conditions. Specifically, the positioning solution quality is determined by the positioning solution status, plane variance, and elevation variance; attitude stabilization quality is determined by the aircraft's pitch angle fluctuation, rotation angle fluctuation, and working device angle jitter; point cloud quality is determined by the number of effective points within the unit, the proportion of point cloud holes, and height dispersion; image recognition quality is determined by image sharpness, boundary discernibility, and target surface visibility; time synchronization quality is determined by the timestamp deviation of each sensor; and occlusion status is determined by the proportion of occlusion of the unit's field of view by the bucket, boom, soil pile, and obstacles. The overall perception reliability Qi is calculated using the following formula: In the formula, Q i The overall perceived credibility of target task unit i; q g,i To calculate the quality score for the location solution; q a,i The attitude stability quality score; q l,i Point cloud quality score; q v,i Image recognition quality score; q t,i For time synchronization quality score; q o,i The occlusion status is scored. The weights in the formula are pre-set based on the contribution of terrain measurement and motion guidance to various types of data. The weight of point cloud quality is 0.30, the combined weight of localization solution quality and attitude stabilization quality is 0.40, and the combined weight of image recognition quality, time synchronization quality, and occlusion status is 0.30.

[0026] Taking the target work unit i in the middle of the aforementioned work area as an example, its q g,i =0.96, q a,i =0.87, q l,i =0.79, q v,i =0.76, q t,i =0.97, q o,i =0.81, substituting this into the equation yields Qi = 0.858. Another target work unit j, located near the slope, is affected by soil accumulation and the obstruction of the bucket pole; its q... g,j =0.91, q a,j =0.82, q l,j =0.52, q v,j=0.61, q t,j =0.95, q o,j =0.41, substituting this into the equation gives Q. j =0.676. Further stipulation: when Q... i When Q is not less than 0.75, the corresponding target job unit is marked as a high-confidence unit; when Q... i When Q is between 0.55 and 0.75, the corresponding target job unit is marked as an available unit; when Q... i When the value is less than 0.55, the corresponding target job unit is marked as a low-confidence unit and will participate in the calculation of low-confidence coverage in subsequent evaluations.

[0027] After obtaining the perceived confidence level, the onboard computing unit propagates the terrain uncertainty based on the perceived confidence level. This embodiment adopts a unit-level propagation method, first calculating the height observation dispersion d based on the point cloud elevation samples within the target operation unit. z,i Then, based on the normal fluctuation of the locally fitted slope, the slope observation dispersion d is calculated. β,i For any target work unit i, the elevation uncertainty σ z,i and slope uncertainty σ β,i The following rules are used to determine σ: z,i =d z,i [1+κ(1-Q i )],σ β,i =d β,i [1+κ(1-Q i In the formula, κ is the uncertainty propagation coefficient, used to adjust the amplification of the uncertainty of the unit terrain by the perceived credibility and the degree of observation dispersion. In this embodiment, the initial κ is 0.35. When the target work unit is located at the slope turning point, the perceived credibility decreases within two consecutive evaluation cycles, or there is aggravation of local occlusion, κ is allowed to increase to 0.40 to 0.45 in this local area to improve the sensitivity of the upper limit of the subsequent residual work volume range to adverse working conditions.

[0028] To facilitate the unified application of elevation uncertainty and slope uncertainty in volume interval calculations, this embodiment further converts them into an equivalent height uncertainty u. i , where u i From σ z,i The sum of the elevation projection error caused by slope uncertainty at the half-unit scale is used to determine this. For a square target work unit with a side length of 0.5m, if σ z,i =0.021m, σ β,i =1.6°, then u i Approximately 0.028m; if the Q of a certain target work unit i If the value decreases to 0.49, and both the point cloud height dispersion and the slope observation dispersion increase simultaneously, then u iThis can be increased to 0.061m. Therefore, the perceived confidence level is propagated into terrain uncertainty that can subsequently be used for volumetric interval calculations.

[0029] This embodiment employs a residual workload interval-bucket sweep matching evaluation model based on perceived uncertainty propagation. The model takes the designed terrain state, measured terrain state, and multi-source perceived quality parameters of the target work unit as input. First, it obtains the unit-level terrain uncertainty through perceived confidence propagation, and then obtains the residual workload interval of the corresponding target work unit from the terrain uncertainty. Subsequently, it maps the bucket sweep area corresponding to the candidate bucket action element to a confidence operation residual map, calculates the effective reduction benefit, over-excavation risk, low confidence risk, and constraint conflict risk of the candidate bucket action element for the covered target work unit, and determines the target bucket action element based on screening and optimization rules. After the target bucket action element is executed, the soil correction coefficient, bucket sweep effectiveness coefficient, uncertainty propagation coefficient, and action evaluation weight coefficient in the model are rewritten and corrected based on the measured results after execution, to update the evaluation results of subsequent unexecuted candidate bucket action elements. In this model, perceived confidence, elevation uncertainty, slope uncertainty, equivalent height uncertainty, residual work range, bucket sweep area, confidence-weighted residual reduction, expected over-excavation, low-confidence coverage, and constraint conflict constitute an evaluation chain with sequential coupling.

[0030] For any target work unit, the residual workload range is determined based on the target elevation, measured elevation, elevation uncertainty, target work unit area, and allowable error. This embodiment uses the difference Δh between the target elevation and the measured elevation. i The unit elevation residual is expressed in terms of allowable error ε. i Indicates the completion tolerance, with the target work unit area A. i and soil correction factor η i As a volume conversion parameter, A i =0.25m², η i The initial value is 0.93. The residual workload range is determined using the following formula: In the formula, This represents the lower limit of the amount of work to be reduced in target work unit i. The upper limit of the amount of work to be reduced for target work unit i. The excess of the measured elevation of target work unit i relative to the target elevation. The allowable error for this target work unit. This is the equivalent height uncertainty obtained by converting elevation uncertainty and slope uncertainty. The target work unit area. This is the soil correction factor. When a target work unit is located in a sloping area, the soil correction factor is first adjusted based on the target slope and the measured slope. After correcting for the slope normal, substitute the values ​​into the above formula. Taking target work unit i in the middle of the work area as an example, its target elevation is -2.400m, and the measured elevation is -2.210m, then Δh i =0.190m, allowable error ε i =0.030m, u i =0.028m, substituting this into the equation, we obtain the residual workload range for this target work unit as approximately [0.030m³, 0.043m³]. The measured elevation of the other corner target work unit j is -2.160m. Although its nominal elevation residual is larger, due to u i Reaching 0.061m indicates a significantly wider residual workload range, suggesting higher uncertainty in terrain perception at that location. This embodiment specifies that when the upper limit of the residual workload range for a target work unit is no greater than 0.008m³, the target work unit is marked as a completed unit; when the upper limit is greater than 0.008m³ but the lower limit is no greater than 0, the target work unit is marked as a refinement unit; and when the lower limit is greater than 0, the target work unit is marked as a unit awaiting work. The residual workload ranges and constraint states of all target work units together constitute a reliable work residual map. The constraint states include whether the target work unit is adjacent to a slope, whether it is adjacent to a restricted area, whether it is affected by an underground protected area, and whether it is located in a low-confidence area.

[0031] After obtaining the reliable operational residual map, the onboard computing unit begins to generate multiple candidate bucket motion primitives. This embodiment uses a 22t tracked hydraulic excavator with a boom length of 5.70m, a stick length of 2.90m, a bucket width of 1.20m, and a rated bucket capacity of 1.0m³. The current machine position center coordinates are (10.40m, 6.80m), and the machine's orientation forms an angle of 18° with the longitudinal direction of the site. The onboard computing unit determines the reachable range of the bucket teeth based on the excavator's current position, working device structural parameters, and joint limits. The working range of the boom angle is set to 18° to 63°, the stick angle to 22° to 138°, and the bucket angle to 28° to 142°. Simultaneously, a safety distance of no less than 0.40m from the edge of the pit and a distance of no less than 0.30m from the outer edge of the underground protection zone are set. Subsequently, target work units with a residual work volume lower limit greater than 0.040 m³ and a perceived confidence level of not less than 0.55 are selected as candidate work units from the reliable work residual map. For target work units adjacent to completed units, slopes, or restricted areas, their allowed entry and exit directions are further defined. Based on the current machine position, this embodiment generates a total of 12 candidate bucket motion elements. Each candidate bucket motion element includes at least an entry point, an exit point, a cutting depth, an entry posture, and a tooth tip trajectory. Taking candidate bucket motion element P2 as an example, its entry point is (11.00 m, 7.20 m), its exit point is (12.05 m, 7.75 m), its cutting depth is 0.31 m, its entry posture is 41° relative to the horizontal plane, its tooth tip trajectory is connected by a quintic spline curve with a trajectory length of 1.42 m, and its cutting direction has an angle of 12° with the design boundary.

[0032] like Figure 2As shown, each candidate bucket motion primitive forms a posture envelope of a complete operation based on the tooth tip trajectory, bucket width, and bucket attitude changes, and the corresponding bucket sweep area is obtained from this posture envelope. After mapping the bucket sweep area to the reliable operation residual map, the set of target operation units covered by the candidate bucket motion primitive can be obtained. Subsequently, the confidence-weighted residual reduction, expected overcut, low-confidence coverage, and constraint conflict are calculated respectively. The confidence-weighted residual reduction is determined by summing "unit coverage ratio × unit residual work range × unit perceived confidence level × bucket sweeping effectiveness coefficient" unit by unit, retaining both a lower limit and an upper limit (the bucket sweeping effectiveness coefficient is used to characterize the proportion of theoretical sweeping volume converted into actual effective reduction volume); the expected over-excavation volume is determined by the upper limit of the volume portion of the bucket sweeping area that exceeds the design terrain boundary or the allowable error range; the low-confidence coverage is determined by the proportion of the covered target work unit with a perceived confidence level below 0.55; the constraint conflict is determined by the overlap ratio between the bucket sweeping area and the restricted area, obstacle area, and working device constraint boundaries. The motion path cost is used to characterize the smoothness and efficiency of the candidate bucket motion element. It is obtained by weighting the changes in slewing angle, boom angle, stick angle, and tooth tip trajectory length. The weight of the slewing angle change is 0.35, the weight of the boom angle change is 0.20, the weight of the stick angle change is 0.20, and the weight of the tooth tip trajectory length is 0.25. The smaller the value, the better the smoothness of the execution and the energy utilization of the candidate bucket motion element.

[0033] To illustrate the screening and selection process, Table 1 presents the evaluation results of four representative candidate bucket motion primitives.

[0034] Table 1 Evaluation results of candidate bucket motion primitives Candidate bucket motion primitive Confidence-weighted residual reduction range / m³ Expected over-excavation limit / m³ Low Trust Coverage Constraint Conflict Quantity Action path cost P1 [0.47,0.61] 0.05 0.09 0.00 1.34 P2 [0.47,0.58] 0.03 0.07 0.00 1.26 P3 [0.50,0.73] 0.04 0.31 0.00 1.22 P4 [0.44,0.55] 0.02 0.06 0.08 1.19 In this embodiment, the screening and selection rules are executed in the following order: First, candidate bucket action elements with constraint conflict exceeding the first threshold of 0.02 or low confidence coverage exceeding the second threshold of 0.25 are eliminated; then, among the remaining candidate bucket action elements, the candidate bucket action element with the largest confidence-weighted residual reduction lower limit is selected first; when multiple candidate bucket action elements have the same confidence-weighted residual reduction lower limit, the candidate bucket action element with the smallest expected over-excavation upper limit and the smallest action path cost is selected in sequence. As shown in Table 1, although the confidence-weighted residual reduction lower limit of P3 reaches 0.50 m³, its low confidence coverage is 0.31, exceeding the second threshold of 0.25, and therefore it is eliminated; the constraint conflict of P4 is 0.08, exceeding the first threshold of 0.02, and therefore it is eliminated. Both P1 and P2 have the same lower limit for the confidence-weighted residual cut, 0.47 m³. Then, comparing the upper limit of the expected overcut and the action path cost, P2 has a smaller upper limit for the expected overcut and a lower action path cost; therefore, P2 is determined as the target bucket action element. After the target bucket action element is determined, the onboard computing unit outputs its guidance parameters, including the entry point, exit point, cutting depth, bucket posture, and tooth tip guide trajectory. Simultaneously, it outputs the target work unit number covered by the action, low-confidence area boundary warnings, and no-entry boundary warnings. In the cab interface, the entry point is displayed as a green marker, the exit point as a blue marker, the tooth tip guide trajectory as a continuous curve, and the low-confidence area is indicated as a yellow semi-transparent area.

[0035] After the target bucket action element is executed, the onboard computing unit re-acquires the measured terrain state of the work area and recalculates the perception reliability, terrain uncertainty, and residual work volume range of the target work element in that area. In this embodiment, the predicted residual reduction range before P2 execution is [0.47m³, 0.58m³]. After execution, the measured residual reduction is calculated as 0.43m³ by updating the volume difference of the terrain surface before and after execution and the sweep coverage relationship. Since 0.43m³ is lower than the lower limit of the predicted residual reduction range of 0.47m³, the onboard computing unit determines that the actual reduction effect deviates negatively from the predicted range before execution. Subsequently, based on the degree of deviation of the measured residual reduction after execution from the predicted residual reduction range before execution, the evaluation parameters are rewritten and corrected. Specifically, the soil correction coefficient η is corrected from 0.93 to 0.89, the effective coefficient of bucket sweeping ξ is corrected from 0.86 to 0.82, and the uncertainty propagation coefficient κ is corrected from 0.35 to 0.39. The above corrections are applied first to local areas with the same soil type, similar slope conditions, and similar shading patterns as the currently executed actions, in order to improve the consistency between the evaluation results of subsequent actions and the actual operational effects.

[0036] After revising the evaluation parameters, the onboard computing unit recalculates the confidence-weighted residual reduction, expected over-excavation, low-confidence coverage, and constraint conflict of the unexecuted candidate bucket action units. Taking the target bucket action unit P6 and another unexecuted candidate bucket action unit P7, which were determined before the update, as examples, the lower limit of the confidence-weighted residual reduction of P6 before the revision was 0.45 m³, and the upper limit of the expected over-excavation was 0.04 m³. After the revision, due to the increased uncertainty of adjacent target work units and the decrease in the effective coefficient of bucket sweeping, the lower limit of its confidence-weighted residual reduction decreased to 0.39 m³. The lower limit of the confidence-weighted residual reduction of P7 before the revision was 0.44 m³, and the upper limit of the expected over-excavation was 0.03 m³. After the revision, since its coverage area is mainly located in high-confidence target work units, the lower limit of the revised confidence-weighted residual reduction remains at 0.41 m³.

[0037] After re-screening and optimization, P7 becomes the new optimal candidate bucket action primitive. When P7 is inconsistent with the previously determined target bucket action primitive P6, the system outputs the entry point, exit point, cutting depth, and bucket attitude guidance parameters after the switch, and determines P7 as the next action to be executed. Thus, the entire processing flow forms a complete closed loop, from constructing the target work unit set, acquiring measured terrain conditions, determining perceived reliability, propagating terrain uncertainty, generating residual work volume intervals, constructing a reliable work residual map, generating candidate bucket action primitives, evaluating bucket sweep matching, screening and optimizing candidate actions, executing target actions, writing back execution deviations, and recalculating unexecuted candidate actions.

[0038] In this embodiment, the perceived reliability is determined by at least three of the following: positioning solution quality, attitude stabilization quality, point cloud quality, image recognition quality, time synchronization quality, and occlusion status. Elevation uncertainty and slope uncertainty are obtained through perceived reliability and observation dispersion propagation. The residual workload range is generated jointly by the target elevation, measured elevation, elevation uncertainty, target work unit area, and allowable error. The confidence-weighted residual reduction, expected over-excavation, low-reliability coverage, and constraint conflict are determined by the coverage relationship between the bucket sweep area and the reliable work residual map. Evaluation parameters are corrected by writing back the deviation of the measured residual reduction after execution from the predicted residual reduction range before execution. This processing flow has clear parameter meanings and a clear input-output relationship, enabling stable output of the entry point, exit point, cutting depth, and bucket attitude guidance parameters for the next scoop action in continuous excavation scenarios, and continuously correcting the evaluation results of subsequent actions based on the actual execution results.

[0039] Comparative Example 1 To verify the technical effect of the present invention in propagating terrain uncertainty based on perception credibility and observation discreteness and generating residual workload range, Comparative Example 1 was set up under the same construction scenario, machine parameters, sensor configuration, evaluation cycle and operation constraints as Example 1.

[0040] In Comparative Example 1, the construction design data is also converted into a set of target work units in the on-site construction coordinate system. The multi-source sensing data is also used to obtain the measured terrain status corresponding to the target work unit, generate candidate bucket action elements, and perform bucket sweeping matching evaluation. The difference is that Comparative Example 1 directly determines the residual work amount of the unit based on the difference between the target elevation and the measured elevation of the target work unit, and selects and optimizes the candidate bucket action elements based on the single residual work amount.

[0041] For low-confidence areas, obscured areas, and slope transition areas, only a prompt is displayed on the interface, but these are not further converted into elevation uncertainty, slope uncertainty, or equivalent height uncertainty, nor are they included in the calculation of the upper and lower limits of the residual workload range. Apart from this, the generation method of candidate bucket action elements, the calculation methods for confidence-weighted residual reduction, expected over-excavation, low-confidence coverage, and constraint conflict in Comparative Example 1 are consistent with those in Example 1. Specifically, the confidence-weighted residual reduction in Comparative Example 1 is determined by cumulatively adding the unit coverage ratio, the unit's single-value residual workload, and the unit's perceived confidence level unit by unit.

[0042] Under this comparative model, since the terrain perception of the target work unit participates in the action evaluation with single-value results, it cannot reflect the impact of occlusion, point cloud voids, slope turning and local dispersion on the judgment of the unit's residual work volume. As a result, some candidate bucket action units are still judged as high-profit actions in the area near the slope and in local low confidence areas, which in turn increases the risk of local over-excavation and the probability of rework.

[0043] Comparative Example 2 To verify the technical effect of the deviation write-back correction mechanism in this invention on the stability of the continuous excavation process action optimization, Comparative Example 2 was set up under the same construction scenario, machine parameters, sensor configuration, operation constraints, and perception credibility and terrain uncertainty propagation rules as Example 1.

[0044] In Comparative Example 2, the perceived reliability is also calculated based on the positioning solution quality, attitude stabilization quality, point cloud quality, image recognition quality, time synchronization quality, and occlusion status. Elevation uncertainty, slope uncertainty, and equivalent height uncertainty are obtained based on the observation dispersion, thereby generating the residual workload interval and the reliable operation residual map. Similarly, the confidence-weighted residual reduction, expected over-excavation, low-reliability coverage, and constraint conflict are calculated based on the coverage relationship between the bucket sweeping area and the reliable operation residual map. The target bucket action element is selected according to the same rules as in Example 1. The difference lies in that, after the target bucket action element is executed, Comparative Example 2 does not rewrite and correct the soil correction coefficient, bucket sweeping effectiveness coefficient, uncertainty propagation coefficient, and action evaluation weight coefficient based on the deviation of the measured residual reduction after execution from the predicted residual reduction interval before execution. Nor does it recalculate the evaluation results for unexecuted candidate bucket action elements. Instead, it keeps the initial evaluation parameters unchanged and continues to execute subsequent actions according to the original parameters.

[0045] In this comparative example, although the initial screening accuracy of the candidate actions is better than that of Comparative Example 1, when the soil condition, local occlusion degree and actual reduction effect change during continuous excavation, the evaluation parameters are not updated in a closed loop with the execution results. As a result, the benefit judgment and risk judgment of the subsequent candidate bucket action elements will gradually deviate from the actual working conditions, leading to a decrease in the effective reduction of a single bucket, a deterioration in the stability of continuous operation, and an increase in the number of local reworks.

[0046] To verify the effectiveness of the method of the present invention, under the same construction scenario, machine parameters, sensor configuration and operation constraints as in Example 1, continuous excavation statistics were performed on the method of the present invention, Comparative Example 1 and Comparative Example 2 respectively. The average absolute error of the final elevation, the over-excavation volume ratio, the average effective reduction per shovel and the number of rework were used as evaluation indicators. The statistical results are shown in Table 2.

[0047] Table 2 Comparison of the results of different methods

[0048] As shown in Table 2, the average absolute error of the final elevation of the method of the present invention is 0.017m, which is significantly less than 0.034m of Comparative Example 1 and 0.025m of Comparative Example 2; the over-excavation volume ratio of the method of the present invention is 3.8%, which is significantly lower than 9.6% of Comparative Example 1 and 6.7% of Comparative Example 2; the average effective reduction per shovel of the method of the present invention reaches 0.44m³, which is higher than 0.39m³ of Comparative Example 1 and 0.41m³ of Comparative Example 2; the number of reworks is only 3, which is less than 8 times of Comparative Example 1 and 5 times of Comparative Example 2.

[0049] The above results show that, in Comparative Example 1, because the perceived credibility and observation dispersion were not propagated into terrain uncertainty, a residual workload interval reflecting the unit measurement risk could not be formed. This resulted in the benefit judgment of candidate bucket action elements being insufficiently sensitive to low-confidence areas and slope transition areas, easily amplifying the risk of local over-excavation and rework. The method of this invention, by sequentially propagating perceived credibility, observation dispersion, elevation uncertainty, slope uncertainty, and equivalent height uncertainty to the residual workload interval and the credible operation residual map, enables the screening and optimization process of candidate bucket action elements to simultaneously consider operational benefits and perceived risks, thus significantly reducing elevation error and over-excavation volume ratio.

[0050] Meanwhile, although Comparative Example 2 retained the perception credibility and terrain uncertainty propagation process, and its initial action selection results were better than Comparative Example 1, the lack of back-writing corrections to the soil correction coefficient, bucket sweeping effectiveness coefficient, uncertainty propagation coefficient, and action evaluation weight coefficient based on the measured residual reduction after execution led to a gradual deviation of the action evaluation parameters from the actual working conditions during continuous excavation. This resulted in insufficient stability in the subsequent selection results of candidate bucket action elements, manifested as a decrease in the average effective reduction per shovel and an increase in the number of rework operations. The method of this invention achieves consistency correction between candidate action evaluation results and actual operational effects by back-writing the execution deviation to the evaluation process of unexecuted candidate bucket action elements within areas of similar soil layers, similar slope conditions, and similar occlusion patterns. Therefore, it exhibits higher operational accuracy and better dynamic adaptability in continuous excavation scenarios.

[0051] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0052] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for guiding excavator operation based on multi-sensor data fusion, characterized in that, include: Convert the construction design data into a set of target work units in the on-site construction coordinate system; Based on multi-source sensing data, the measured terrain state corresponding to each target work unit is obtained, the perception credibility of the measured terrain state is determined, and the terrain uncertainty is obtained by propagation based on the perception credibility and the observation dispersion of the corresponding measured terrain state. Combining the design terrain state, measured terrain state, and terrain uncertainty of the corresponding target work unit, a credible work residual map containing the residual work volume range and constraint state is generated. Based on the current pose of the excavator and the work constraints, multiple candidate bucket action elements are generated. Based on the coverage relationship between the bucket sweeping area of ​​each candidate bucket action element and the credible work residual map, the confidence-weighted residual reduction, expected over-excavation, low-credibility coverage, and constraint conflict of each candidate bucket action element are calculated. Based on the calculation results, the candidate bucket action elements are screened and optimized to determine the target bucket action element and output its guidance parameters. After the target bucket action element is executed, the evaluation parameters used for screening and selection are corrected based on the updated measured terrain conditions and the changes in residual workload before and after execution.

2. The excavator operation guidance method based on multi-sensor data fusion according to claim 1, characterized in that, The process of converting construction design data into a set of target work units in the on-site construction coordinate system includes: dividing construction drawings, three-dimensional design models, or input work parameters into multiple target work units; establishing a coordinate transformation relationship between the design coordinate system and the on-site construction coordinate system; and mapping each target work unit to the on-site construction coordinate system. Each target work unit is associated with at least two of the following: target elevation, target slope, target depth, allowable error, and work boundary.

3. The excavator operation guidance method based on multi-sensor data fusion according to claim 1, characterized in that, The terrain uncertainty is obtained by propagating the perceived credibility, including: determining the perceived credibility based on at least three of the following: positioning solution quality, attitude stabilization quality, point cloud quality, image recognition quality, time synchronization quality, and occlusion status; and determining the elevation uncertainty and slope uncertainty of the corresponding target operation unit by combining the observation dispersion of the corresponding measured terrain status.

4. The excavator operation guidance method based on multi-sensor data fusion according to claim 3, characterized in that, The residual workload range is determined based on the target elevation, measured elevation, elevation uncertainty, target work unit area, and allowable error of the target work unit; when the upper limit of the residual workload range is not greater than the preset completion threshold, the corresponding target work unit is marked as a completed unit.

5. The excavator operation guidance method based on multi-sensor data fusion according to claim 1, characterized in that, The generation of multiple candidate bucket action primitives includes: determining the reachable range of the bucket teeth tip based on the excavator's current position, working device structural parameters, and joint limits; selecting target work units from the reliable work residual map whose lower limit of the residual work volume interval is greater than a preset threshold and whose perceived reliability is not lower than a preset reliability threshold as candidate work units; and generating multiple candidate bucket action primitives based on the boundary of the candidate work unit, target slope, current machine position, and the reachable range of the bucket teeth tip.

6. The excavator operation guidance method based on multi-sensor data fusion according to claim 5, characterized in that, Each candidate bucket motion element includes at least an entry point, an exit point, a cutting depth, an entry posture, and a tooth tip trajectory; the bucket sweeping area is determined based on the posture envelope formed by the tooth tip trajectory, the bucket width, and the changes in the bucket posture.

7. The excavator operation guidance method based on multi-sensor data fusion according to claim 1, characterized in that, The calculation of the confidence-weighted residual reduction, expected over-excavation, low-confidence coverage, and constraint conflict for each candidate bucket action element includes: determining the lower and upper limits of the confidence-weighted residual reduction based on the overlap between the bucket sweep area and the covered target work unit, the residual work volume range, the perceived confidence level, and the effective coefficient of the bucket sweep; determining the upper limit of the expected over-excavation based on the portion of the bucket sweep area that exceeds the design terrain boundary or the allowable error range; determining the low-confidence coverage based on the proportion of the covered target work unit with a perceived confidence level lower than a preset threshold; and determining the constraint conflict based on the overlap ratio between the bucket sweep area and the restricted area, obstacle area, or work device constraint boundary.

8. The excavator operation guidance method based on multi-sensor data fusion according to claim 7, characterized in that, The process of screening and selecting candidate bucket action elements based on the calculation results includes: firstly, eliminating candidate bucket action elements whose constraint conflict exceeds a first threshold or whose low confidence coverage exceeds a second threshold; then, prioritizing the selection of candidate bucket action elements with the largest lower limit of confidence-weighted residual reduction among the remaining candidate bucket action elements; and when multiple candidate bucket action elements have the same lower limit of confidence-weighted residual reduction, selecting the candidate bucket action elements with the smallest upper limit of expected over-excavation and the smallest action path cost in sequence.

9. The excavator operation guidance method based on multi-sensor data fusion according to claim 1, characterized in that, The step of correcting the evaluation parameters based on the updated measured terrain state and the changes in residual workload before and after execution includes: obtaining the measured terrain state after the execution of the target bucket action element; when the measured residual reduction amount after execution deviates from the predicted residual reduction amount range before execution, at least one of the soil correction coefficient, bucket sweeping effective coefficient and uncertainty propagation coefficient is corrected according to the degree of deviation.

10. The excavator operation guidance method based on multi-sensor data fusion according to claim 9, characterized in that, After correcting the evaluation parameters, the confidence-weighted residual cut, expected overcut, low confidence coverage, and constraint conflict of the candidate bucket action primitives that were not executed are recalculated. When the recalculated optimal candidate bucket action primitive is inconsistent with the target bucket action primitive determined before the update, the entry point, exit point, cutting depth, and bucket attitude guidance parameters after the switch are output.

Images