Intelligent sorting robot dynamic obstacle avoidance system based on multi-sensor fusion

Abstract

The application discloses a dynamic obstacle avoidance system of an intelligent sorting robot based on multi-sensor fusion and belongs to the technical field of robot safety control and obstacle avoidance.The application obtains environmental sensing information and odometer information, generates a target list and a local environment occupancy grid, obtains actuator posture state information, determines shape envelope parameters such as an overhanging length and an overhanging direction angle, performs directional obstacle inflation on the local environment occupancy grid according to the shape envelope parameters, obtains a posture envelope occupancy grid, adaptively adjusts a hierarchical safety domain threshold according to the shape envelope parameters, and determines a warning domain, a deceleration domain and a parking domain, and outputs a mobile chassis speed / brake control instruction and an actuator locking / recovery linkage instruction.The application can dynamically adjust a safety domain with changes in the actuator posture, reduce false triggering and unnecessary parking, and improve obstacle avoidance safety and work efficiency in a complex environment.

Description

Technical Field

[0001] This invention belongs to the field of robot safety control and obstacle avoidance technology, specifically relating to a dynamic obstacle avoidance system for intelligent sorting robots based on multi-sensor fusion. Background Technology

[0002] When mobile chassis sorting robots perform tasks such as package pushing and guiding within the sorting area, they typically rely on visual sensors, distance sensors, and odometer information to achieve obstacle perception and avoidance control. In existing technologies, the local environment occupancy representation and safe domain threshold are mostly established based on the fixed shape envelope of the vehicle body. When the pushing / guiding actuator is in an extended, swinging, or multi-pose state, the robot's shape envelope changes accordingly. The occupancy representation and threshold determination under the fixed envelope are inconsistent with the actual collision boundary, leading to the following objective problems: (1) The safety domain triggering conditions are not adjusted synchronously with the changes in the shape of the actuator, and there is a risk of delay in deceleration or stopping triggering; (2) The occupied grid does not reflect the sweep boundary introduced by the attitude of the actuator, and the obstacle avoidance planning space does not adequately express the constraints of the real shape; (3) In a dynamic obstacle-dense environment, when the fixed threshold and fixed envelope are used together, there is insufficient collision margin or excessive conservatism that may cause false stops, resulting in insufficient stability of the operation cycle.

[0003] To address the problem that changes in the robot's shape envelope caused by changes in the posture of the push / guide actuator lead to deviations in the local occupancy representation and the safety domain threshold determination from the actual collision boundary, this invention establishes a closed loop of actuator posture, shape envelope parameters, grid expansion, threshold tightening, hierarchical safety domain, and linkage control. This ensures that the safety domain determination is simultaneously constrained by both the posture envelope occupancy grid and the posture-related threshold, and outputs linkage control commands between the chassis and the actuator. Summary of the Invention

[0004] The purpose of this invention is to provide a dynamic obstacle avoidance system for intelligent sorting robots based on multi-sensor fusion, which solves the technical problem in the prior art where changes in the posture of the pushing / guiding actuator cause changes in the robot's shape envelope, thereby causing the local occupancy expression and the safety domain threshold determination to deviate from the actual collision boundary.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A dynamic obstacle avoidance system for intelligent sorting robots based on multi-sensor fusion, wherein the sorting robot is used to perform pushing / guiding operations on packages, including: The perception fusion unit is used to generate a target list of dynamic obstacles and a local environment occupancy grid based on environmental information from at least one visual sensor and at least one distance sensor, combined with odometer information from the mobile chassis. The attitude envelope parameter unit is used to obtain the attitude state information of the push / guide actuator and determine the shape envelope parameters accordingly. The attitude-related safety domain unit is used to perform obstacle expansion processing on the local environment occupancy grid according to the shape envelope parameters to obtain the attitude envelope occupancy grid, and adjust the hierarchical safety domain threshold based on the shape envelope parameters, and then determine the warning domain, deceleration domain and parking domain according to the attitude envelope occupancy grid and the adjusted threshold. The linkage control unit is used to output speed limit commands and / or braking commands for the mobile chassis according to the graded safety domain, and to output locking commands and / or retraction commands for the push / guide actuator. The hierarchical safety domain threshold tightens as the shape envelope parameter increases, and the expansion amount of the obstacle expansion process is determined by the shape envelope parameter, so that the hierarchical safety domain determination is simultaneously constrained by both the attitude envelope occupancy grid and the attitude association threshold.

[0006] Furthermore, the shape envelope parameters include at least the extension length and extension direction angle of the push / guide actuator, and optionally include the extension side and the shape envelope expansion amount; the attitude state information includes the encoder reading and / or limit switch status of the actuator, and the shape envelope parameters are obtained by converting the encoder reading and / or by selecting a preset parameter table through the limit switch status.

[0007] Furthermore, the obstacle expansion process is a morphological expansion of the grid, and the structural element used is a directional ellipse, whose major axis direction is consistent with the outward extension direction angle, and the expansion amount along the outward extension direction is greater than the expansion amount perpendicular to the outward extension direction, and the length of the major axis increases with the increase of the outward extension length.

[0008] Furthermore, the graded safety domain threshold includes at least a deceleration threshold and a stopping threshold. The threshold is calculated based on the time to collision TTC and / or minimum safe distance, and satisfies the following: as the overhang length increases, at least one of the deceleration threshold and the stopping threshold increases monotonically; the threshold adjustment is implemented using a linear function or a piecewise linear function, and the maximum and minimum values ​​of the threshold are constrained by a preset safety range.

[0009] Furthermore, the linkage control unit outputs a deceleration mechanism locking command when it determines that it has entered the deceleration zone; when it determines that it has entered the parking zone, it executes the following sequence: first, it outputs a moving chassis braking command to reduce the chassis speed to below a preset low speed threshold, and then outputs a deceleration mechanism retraction command; and only after the deceleration mechanism has been retracted or a preset safe posture has been reached, it is allowed to switch from the parking zone to the deceleration zone or the warning zone and apply a slow start speed limit.

[0010] A dynamic obstacle avoidance method for intelligent sorting robots based on multi-sensor fusion, wherein the sorting robot is used to perform pushing / guiding operations on packages, includes the following steps: S1. Acquire environmental information from at least one visual sensor and at least one distance sensor, and acquire odometer information from the mobile chassis; S2. Based on the environmental information and combined with the odometer information, perform fusion processing to generate a target list of dynamic obstacles and generate a local environment occupancy grid. S3. Obtain the attitude status information of the push / guide actuator and determine the shape envelope parameters accordingly; S4. Perform obstacle dilation processing on the local environment occupancy grid according to the shape envelope parameters to obtain the attitude envelope occupancy grid; S5. Adjust the threshold of the graded safety domain based on the shape envelope parameters, and determine the warning domain, deceleration domain and parking domain according to the attitude envelope occupying the grid and the adjusted threshold. S6. Output speed limit command and / or braking command for the mobile chassis according to the warning domain, deceleration domain and parking domain, and output locking command and / or retraction command for the push / guide actuator; The hierarchical safety domain threshold tightens as the shape envelope parameter increases, and the expansion amount of the obstacle expansion process is determined by the shape envelope parameter.

[0011] Further, step S4 includes: using a directional elliptical structural element to morphologically expand the occupied grid, wherein the major axis of the structural element is aligned with the outward extension direction angle, and the expansion amount along the outward extension direction is greater than the expansion amount perpendicular to the outward extension direction, and the length of the major axis increases with the increase of the outward extension length.

[0012] Furthermore, the graded safety domain threshold in step S5 includes at least a deceleration threshold and a stopping threshold. The threshold is calculated based on the time to collision TTC and / or minimum safe distance, and satisfies that at least one of the deceleration threshold and the stopping threshold increases monotonically as the overhang length increases. The threshold adjustment is achieved through a linear function or a piecewise linear function and is constrained by a preset safety range.

[0013] Furthermore, the target list includes a target category field, and in step S5, a more conservative threshold adjustment coefficient and / or a larger expansion amount are applied to pedestrian targets than to dropped packages, so that the conditions for pedestrian targets to trigger deceleration or stopping domains are more stringent.

[0014] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention obtains the attitude state information of the actuator and determines the shape envelope parameters, so that the robot's safety protection boundary changes in real time as the actuator extends / swings; and generates an attitude envelope occupancy grid and hierarchical safety domain based on the shape envelope parameters; thereby avoiding the collision blind zone generated by the existing fixed shape / fixed safety distance scheme when the mechanism extends, and significantly improving the safety and risk coverage integrity of the variable configuration robot in complex scenarios.

[0015] 2. This invention expands the directional barrier of the local environment grid based on the shape envelope parameters, so that high-risk directions such as outward extension directions can obtain a more reasonable space margin; at the same time, the thresholds of the warning domain / deceleration domain / stopping domain are adaptively adjusted in association with the shape envelope parameters, and can be combined with spatial constraints and time constraints for joint judgment; this mechanism can provide earlier warning and deceleration / stopping in high-risk postures, and can also suppress false alarms caused by noise and dynamic targets, thereby reducing unnecessary stops and frequent emergency stops, and improving traffic efficiency and operational continuity.

[0016] 3. This invention, while outputting chassis speed / braking commands, also coordinates with the output actuator to perform locking / retraction actions: in the deceleration zone, it restricts the chassis and locks the mechanism to prevent further extension and expansion of risks; in the parking zone, it triggers braking and retracts the mechanism when conditions are met to quickly reduce the shape and collision cross-section; thereby significantly reducing secondary risks such as scraping and entanglement, shortening the recovery time from a dangerous state to a safe state, and reducing the impact of sudden stop-start through graded smooth control, reducing mechanical wear, and improving system stability and service life. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a general block diagram of the dynamic obstacle avoidance system for intelligent sorting robots based on multi-sensor fusion according to the present invention; Figure 2 This is a flowchart of the dynamic obstacle avoidance method for intelligent sorting robots based on multi-sensor fusion according to the present invention; Figure 3 This is a schematic diagram illustrating the structural principle of the intelligent sorting robot of the present invention; Figure 4 This is a schematic diagram illustrating the grid expansion principle of the present invention; Figure 5 This is a schematic diagram of the security domain determination and action mapping of the present invention; Figure 6This is the timing diagram for the linkage control of the present invention. Detailed Implementation

[0019] 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.

[0020] Example 1 See Figures 1-3 As shown, the intelligent sorting robot dynamic obstacle avoidance system based on multi-sensor fusion in this embodiment is installed on the mobile chassis sorting robot, which is used to perform package pushing / guiding operations.

[0021] Specifically, the intelligent sorting robot dynamic obstacle avoidance system based on multi-sensor fusion includes a perception fusion unit, an attitude envelope parameter unit, an attitude-related safety domain unit, and a linkage control unit. The perception fusion unit acquires environmental information output by at least one visual sensor and at least one distance sensor, and combines it with mobile chassis odometer information to generate a dynamic obstacle target list and a local environment occupancy grid. The attitude envelope parameter unit acquires the attitude state information of the push / guide actuator and determines the shape envelope parameters. The attitude-related safety domain unit expands the local environment occupancy grid according to the shape envelope parameters to obtain the attitude envelope occupancy grid, and tightens and adjusts the hierarchical safety domain threshold according to the shape envelope parameters, thereby determining the warning domain, deceleration domain, and stopping domain under the joint constraints of the attitude envelope occupancy grid and the adjusted threshold. The linkage control unit outputs chassis speed limit commands and / or braking commands according to the hierarchical safety domain, and outputs actuator locking commands and / or retraction commands to achieve linkage obstacle avoidance control between the chassis and the actuator.

[0022] To ensure that perception fusion, threshold calculation, and safety domain determination have a consistent reference benchmark, a robot base coordinate system is established in this embodiment. The observation results from the vision sensor and distance sensor are transformed through extrinsic parameter calibration to... The following expression indicates that each sensor output carries a timestamp.

[0023] Preferably, the sensing fusion unit aligns multi-source observations to the same reference time. Furthermore, odometry information is used to compensate for pose differences caused by different sampling times, ensuring that the target list and the local environment occupying the grid are updated at the same reference time. The aforementioned extrinsic parameter calibration, time alignment, and motion compensation are necessary conditions for fusion processing, preventing systematic deviations in subsequent obstacle expansion and TTC / distance threshold determination due to coordinate or time inconsistencies.

[0024] In this embodiment, the perception fusion unit completes the construction and updating of the target list and the local environment occupancy grid in each update cycle. The perception fusion unit can project the center point or bounding box region of the visual target onto the grid coordinate system and generate a dynamic occupancy layer G in the local environment occupancy grid. dyn The final occupied grid cell G can be determined by G = max(G static G dyn ) or G=G static ∪G dyn We obtain and set the update period Δt and cost decay strategy to suppress instantaneous noise.

[0025] Regarding the target list, the visual sensor provides target category information (e.g., pedestrians, dropped packages, or other categories) and confidence level, while the distance sensor provides obstacle spatial geometry information. The system maintains target identification consistency through cross-frame data association and outputs a target list {ID, p} containing at least the target location field. i}, where p i For the goal The target position (m). The target list further includes the target velocity v. i (m / s), class field i With the confidence field conf i This is to support subsequent differentiated security strategies.

[0026] Cross-frame data association can be achieved by: calculating the cost matrix (such as 1-IoU or Euclidean distance) for the detection results of adjacent frames, and then using the Hungarian algorithm to solve for the minimum cost matching; or by using Kalman filtering to predict the target state and performing nearest neighbor matching with gated distance, thereby generating stable target IDs and velocity estimates for the target list.

[0027] Regarding the local environment occupancy grid, the distance sensor point cloud or distance echo is discretized onto the grid plane through rasterization or ray projection, and the grid cells are assigned an occupancy state or value according to the grid resolution r (m / grid) to represent the passable space and obstacle space near the robot.

[0028] See Figure 3As shown, the shape envelope parameters include at least the outward extension length L (m) and outward extension direction angle θ (°) of the push / guide actuator, and optionally include the outward extension side S and the shape envelope expansion amount Δ (m), used to characterize the direction and extent of the actuator's attitude expansion on the robot's shape envelope. Attitude state information is obtained from encoder readings E (counts) and / or limit switch states K.

[0029] The extension length L is obtained by converting the encoder count E, which is a common model for linear calibration of displacement-count in engineering metrology: when the pitch, transmission ratio and encoder resolution are fixed, the displacement and count are approximately linearly related within the working range.

[0030] When using encoder readings, the conversion method for the extension length L according to the linear proportional calibration model is as follows: ; Where k E E0 is the displacement proportionality coefficient (m / count) and zero-count; the dimensions satisfy m / count × count = m. When the limit switch is in use, K∈{retracted, half-expanded, expanded} is used to select the preset parameter table and output (L,θ,Δ).

[0031] When the encoder reading E is inconsistent with or missing the limit switch state K, a safety backoff strategy can be adopted to ensure the conservatism of the safety domain determination. For example, the upper limit of L and Δ can be taken or the corresponding parameters of the expansion range can be used to tighten the threshold and maintain a more conservative safety margin with grid expansion.

[0032] See Figure 4 As shown, the obstacle expansion processing for the local environment occupying the grid is based on the expansion operation in mathematical morphology. It expands the obstacle set using structuring elements, thereby mapping the shape envelope changes introduced by the actuator's posture onto the grid constraints. The structuring element is a directional elliptic kernel, whose continuous space definition originates from the analytic geometric elliptic inequality: ; Where a is the major semi-axis (m), b is the minor semi-axis (m), and x′ and y′ are the coordinates after rotation and alignment (m). Both sides of the equation are dimensionless and have consistent dimensions.

[0033] The derivation based on the classic elliptic kernel is that the direction of the elliptic kernel is rotated and aligned with the outward extension angle θ, and the size of the elliptic kernel increases monotonically with the outward extension length L, thereby achieving a more conservative anisotropic expansion along the outward extension direction.

[0034] Rotation alignment uses a two-dimensional rotation matrix: ; Where θ is measured in degrees, and trigonometric functions use the same angular unit; matrix elements are dimensionless, and both input and output coordinates are in m, with consistent dimensions.

[0035] To reflect the safety logic that the greater the extension, the greater the expansion, the core size is defined as a linear function with boundaries: ; ; Where a0 and b0 are the reference dimensions (m), k a k b It is a proportionality constant (dimensionless) and satisfies k a >k b ≥0, thus ensuring that the expansion along the outward direction is greater than the expansion in the vertical direction.

[0036] At the raster implementation level, a(L) and b(L) are discretized into the number of grid cells according to the raster resolution r (m / grid): A = ceil(a(L) / r), B = ceil(b(L) / r); Where A and B are dimensionless integer cell numbers, and ceil(·) is the floor function to ensure that the safety margin is not underestimated after discretization. Based on this, a discrete structuring element set S is constructed, and the occupied grid G ​​is dilated to obtain the attitude envelope occupied grid G. infl : ; Where ⊕ represents the dilation operation; in engineering implementation, this is equivalent to: for each occupied cell, marking the cells within the coverage area of ​​S as occupied or increasing their cost value, thereby obtaining the pose envelope occupied grid G. infl This serves as a spatial constraint for subsequent security domain determination.

[0037] Example 2 See Figures 5-6 As shown, the graded safety domain threshold is calculated based on the time-to-collision (TTC) and / or the minimum safe distance. TTC is a classic safety indicator in relative kinematics and traffic engineering; under the assumption that relative distance changes approximately linearly with approach speed, TTC equals the relative distance divided by the approach speed. Graded safety domain determination includes spatial determination and threshold determination: spatial determination uses the attitude envelope to occupy the grid G. infl The occupancy status is used as the basis; the threshold determination is based on the warning / deceleration / stop threshold after the shape envelope parameter is adjusted; when target i meets the spatial determination condition and the corresponding threshold determination condition, target i is determined to enter the corresponding safety domain.

[0038] Assume the relative position vector between the robot and target i is r. i (m), relative velocity vector is v i(m / s), relative distance is: ; Define the approach velocity along the direction of the line (retaining only the approach component): ; Where r i ⊤ v i The dimension of m 2 / s, divided by ||r i ||(m) gives m / s, which has the same dimensions.

[0039] ; Where ε is the lower limit velocity (m / s), and its dimensions satisfy m / (m / s)=s.

[0040] The key modification in this application is that the grading threshold is set as a function of the extension length L and satisfies monotonically tightening. When the TTC threshold is used: ; ; Among them, T slow,0 T stop,0 k is the baseline threshold (s). s k t The unit is s / m, so that k s L, k t The unit of L is seconds, thus ensuring dimensional consistency and satisfying the condition that L monotonically increases without decreasing and is subject to upper and lower bounds. If a distance threshold is used, D can also be isomorphically defined. slow (L), D stop (L) (m).

[0041] The safety domain determination adopts a common constraint approach: on the one hand, the attitude envelope occupies the grid G. infl It provides spatial constraints; on the other hand, TTC or distance-to-threshold comparison provides time / distance constraints.

[0042] Spatial constraints can be valid based on any of the following criteria: (1) The position p of target i i Mapped to the pose envelope occupying grid G infl Grid cell (qx) i ,qy i If G infl (qx i ,qy i If )≥Gocc ​​(occupancy / cost threshold), then the space constraint holds; (2) Taking the current velocity direction vector u of the chassis as a reference, if p iIf u≥0 and |pi×u|≤W / 2 (W is the preset bandwidth), then the spatial constraint holds.

[0043] Those skilled in the art may choose to use one or a combination of them depending on the implementation complexity.

[0044] Taking TTC as an example: when there exists a target i that satisfies TTC i ≤T stop (L) and when the spatial constraint is met, it is determined that the vehicle enters the parking area; when there is a target satisfying T stop (L) <TTC i ≤T slow (L) When the spatial constraint is met, it is determined to enter the deceleration domain; when the outermost warning threshold range is met and the spatial constraint is met, it is determined to enter the warning domain. By associating the threshold with the attitude parameters and using it together with the attitude envelope occupancy grid, the triggering time of the actuator extension is moved forward and can be verified.

[0045] In this embodiment, the linkage control unit uses a hierarchical safety domain as input and output for coordinated control of the chassis and actuators. The linkage control unit employs a state switching method involving warning, deceleration, stopping, and recovery states. Specifically, when entering the deceleration domain, a speed limit command is output to the chassis and a lock command is output to the actuator; when entering the stopping domain, control is executed sequentially: first, a braking command is output to reduce the chassis speed to below a preset low-speed threshold v. low (m / s), and then output the actuator recovery command to reduce the secondary risks introduced by the recovery action at high speed.

[0046] After the actuator has completed recovery or reached the preset safe posture, it is allowed to switch from the parking zone to the deceleration zone or warning zone, and a slow start speed limit v is applied to the chassis. restart,max (m / s) to suppress the risk of sudden acceleration during the recovery phase. The safe attitude is determined by the position switch or by a geometric threshold, such as satisfying L≤L safe (m) and |θ|≤θ safe (°).

[0047] In this embodiment, to improve the safety margin of the human-machine shared area, the target list includes a target category field. i And adopt a differentiated conservative strategy for different categories.

[0048] For example, a threshold adjustment factor k is introduced for pedestrian targets. ped >1, to make the conditions for pedestrians to trigger the deceleration or stopping zone more stringent; for example, on the TTC threshold side: ; ; Where k pedDimensionless, with the threshold dimension still being s; as an equivalent alternative, the same coefficient can also be used on the expansion side to enlarge the size of the elliptic kernel: ; Where a ped b ped The units for a(L) and b(L) are all in meters (m). A smaller coefficient k can be used for dropped packages. pkg and set k ped >k pkg This leads to a more conservative and verifiable realization for pedestrians.

[0049] Example 3 Expansion kernel size and discretization calculation Given a grid resolution r = 0.05 m / square, an actuator extension length L = 0.35 m, and an extension direction angle θ = 30°, the kernel size mapping parameters are: a0 = 0.15 m, b0 = 0.10 m, k a =0.8, k b =0.3, and set boundary a min =0.15m, a max =0.60m, b min =0.10m, b max =0.35m, then: ; ; The number of discretized lattices is: , ; Based on this, a discrete set of structural elements S is constructed, and directional elliptic expansion is performed on the occupied grid.

[0050] TTC and Security Domain Determination Calculation Let the relative distance d between target i and target i be... i =1.20m, approximate velocity v app,i =0.60m / s, take ε=0.05m / s, then: ; Let the threshold function parameter be: T slow,0 =1.5s, k s =1.0s / m; T stop,0 =0.8s, k t =0.5s / m, and let T slow ∈[1.2,2.8]s、T stop ∈[0.5,1.5]s; When L = 0.35m: T slow (L) = 1.85s, T stop(L) = 0.975 s; Since TTC i = 2.0 s > T slow (L), the deceleration zone / stop zone is not triggered; if the warning threshold T warn (L) = 2.4 s, the warning zone is triggered when the spatial constraint is satisfied. For pedestrian targets, when k ped = 1.3: T slow,ped (L) = 1.3 × 1.85 = 2.405 s; Thus, an earlier deceleration determination is triggered under the same TTCi condition.

[0051] Calculation example of parking state linkage timing Let the low-speed threshold v low = 0.15 m / s, and the current chassis speed v = 0.60 m / s when entering the parking area. The linkage control unit first outputs a braking command to reduce the speed to v = 0.12 m / s (satisfying v < vlow) and then outputs a recovery command; when the received signal is true or L ≤ L safe = 0.05 m, it enters the recovery stage and applies a slow start speed limit v restart,max = 0.30 m / s.

[0052] Example 4 Refer to Figures 1-2 As shown, this example provides an intelligent sorting robot dynamic obstacle avoidance method based on multi-sensor fusion within each update period Δt, including the following steps: Step 1, collect multi-source observations: read visual detection results (target center / box, category class i , confidence conf i ), distance sensor point cloud / echo, and odometer pose increment; Step 2, time alignment and motion compensation: align the observations of each sensor to the reference time t​​​​​​​​​​​​​​​​​Step 4, Cross-frame association and velocity estimation: Construct a cost matrix C (e.g., Euclidean distance or 1−IoU), use the Hungarian algorithm to solve for matching and update the target ID, and use Kalman filtering for state prediction / updating of successfully matched targets to obtain the velocity v. i ; Step 5, attitude parameter calculation: (L, θ, Δ) are obtained from the encoder / limiter. If the data is missing, a conservative backtracking is adopted (taking the upper limit or unfolded parameters). Step 6, construct the structuring element S: calculate a(L), b(L) and discretize them into A and B; for each candidate offset (d x ,d y Determine if the following conditions are met: ; in , Those who meet the requirements are added to set S; Step 7, dilatation to obtain Ginfl: Perform morphological dilatation on G to obtain Ginfl=G⊕S, which serves as the basis for spatial constraints; Step 8, Calculate the threshold and determine the domain state: Calculate T from L. stop (L), T slow (L), calculate TTC per target i By combining spatial constraint criteria, the warning / deceleration / stopping domain status is obtained; Step 9, Linkage Output and State Machine: Output chassis speed limiting / braking and actuator locking / recovery commands based on the domain state; retract the chassis only after v ≤ vlow is satisfied when entering the parking domain; enter the recovery state and apply v after recovery is complete or the safe posture is met. restart,max .

[0053] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0054] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A dynamic obstacle avoidance system for an intelligent sorting robot based on multi-sensor fusion, the sorting robot being used for performing push-sorting / diverting work on parcels, characterized in that, include: The perception fusion unit is used to generate a target list of dynamic obstacles and a local environment occupancy grid based on environmental information from at least one visual sensor and at least one distance sensor, combined with odometer information from the mobile chassis. The attitude envelope parameter unit is used to obtain the attitude state information of the push / guide actuator and determine the shape envelope parameters accordingly. The shape envelope parameters include at least the extension length and extension direction angle of the push / guide actuator; The attitude-related safety domain unit is used to perform obstacle expansion processing on the local environment occupancy grid according to the shape envelope parameters to obtain the attitude envelope occupancy grid, and adjust the hierarchical safety domain threshold based on the shape envelope parameters, and then determine the warning domain, deceleration domain and parking domain based on the attitude envelope occupancy grid and the adjusted hierarchical safety domain threshold. The linkage control unit is used to output speed limit commands and / or braking commands for the mobile chassis according to the graded safety domain, and to output locking commands and / or retraction commands for the push / guide actuator. The graded safety domain threshold includes at least a deceleration threshold and a stopping threshold. The graded safety domain threshold is calculated based on the time to collision (TTC) and / or the minimum safe distance, and satisfies the following: when the overhang length increases, at least one of the deceleration threshold and the stopping threshold increases monotonically, and the expansion amount of the obstacle expansion process is determined by the shape envelope parameter, so that the graded safety domain determination is simultaneously constrained by both the attitude envelope occupancy grid and the graded safety domain threshold. The obstacle expansion process is a morphological expansion of the grid, and the structural element used is a directional ellipse. Its major axis direction is consistent with the outward extension direction angle, and the expansion amount along the outward extension direction is greater than the expansion amount perpendicular to the outward extension direction. Furthermore, the length of the major axis increases with the increase of the outward extension length.

2. The dynamic obstacle avoidance system for intelligent sorting robots based on multi-sensor fusion according to claim 1, characterized in that, The shape envelope parameters also include the extension side and the shape envelope expansion amount; the attitude state information includes the encoder reading of the actuator and / or the limit switch state, and the shape envelope parameters are obtained by converting the encoder reading and / or by selecting a preset parameter table through the limit switch state.

3. The dynamic obstacle avoidance system for intelligent sorting robots based on multi-sensor fusion according to claim 1, characterized in that, The hierarchical security domain threshold adjustment is implemented using a linear function or a piecewise linear function, and the maximum and minimum values ​​of the hierarchical security domain threshold are constrained by a preset security range.

4. The dynamic obstacle avoidance system for intelligent sorting robots based on multi-sensor fusion according to claim 1, characterized in that, When the linkage control unit determines that it has entered the deceleration zone, it outputs a locking command for the guide mechanism; when it determines that it has entered the parking zone, it executes the following sequence: first, it outputs a braking command for the moving chassis to reduce the chassis speed to below a preset low speed threshold, and then outputs a command for the guide mechanism to retract; and only after the guide mechanism has retracted completely or reached a preset safe posture is it allowed to switch from the parking zone to the deceleration zone or the warning zone and apply a slow start speed limit.

5. A dynamic obstacle avoidance method for intelligent sorting robots based on multi-sensor fusion, applicable to the dynamic obstacle avoidance system for intelligent sorting robots based on multi-sensor fusion as described in any one of claims 1-4, wherein the sorting robot is used to perform pushing / guiding operations on packages, characterized in that... Includes the following steps: S1. Acquire environmental information from at least one visual sensor and at least one distance sensor, and acquire odometer information from the mobile chassis; S2. Based on the environmental information and combined with the odometer information, perform fusion processing to generate a target list of dynamic obstacles and generate a local environment occupancy grid. S3. Obtain the attitude status information of the push / guide actuator and determine the shape envelope parameters accordingly; S4. Perform obstacle dilation processing on the local environment occupancy grid according to the shape envelope parameters to obtain the attitude envelope occupancy grid; S5. Adjust the threshold of the graded safety domain based on the shape envelope parameters, and determine the warning domain, deceleration domain and parking domain according to the grid occupied by the attitude envelope and the adjusted graded safety domain threshold. S6. Output speed limit command and / or braking command for the mobile chassis according to the warning domain, deceleration domain and parking domain, and output locking command and / or retraction command for the push / guide actuator; The hierarchical safety domain threshold tightens as the shape envelope parameter increases, and the expansion amount of the obstacle expansion process is determined by the shape envelope parameter.

6. The method for dynamic obstacle avoidance of an intelligent sorting robot based on multi-sensor fusion according to claim 5, characterized in that, Step S4 includes: using a directional elliptical structural element to morphologically expand the occupied grid, wherein the major axis of the structural element is aligned with the outward extension direction angle, and the expansion amount along the outward extension direction is greater than the expansion amount perpendicular to the outward extension direction, and the length of the major axis increases with the increase of the outward extension length.

7. The method for dynamic obstacle avoidance of an intelligent sorting robot based on multi-sensor fusion according to claim 5, characterized in that, In step S5, the graded safety domain threshold includes at least a deceleration threshold and a stopping threshold. The graded safety domain threshold is calculated based on the time to collision TTC and / or minimum safe distance, and satisfies that at least one of the deceleration threshold and the stopping threshold increases monotonically as the overhang length increases. The graded safety domain threshold adjustment is achieved through a linear function or a piecewise linear function and is constrained by a preset safety range.

Images