An automatic obstacle-avoiding unmanned road sweeper

Abstract

The application provides an automatic obstacle-avoiding unmanned road sweeper, and belongs to the technical field of unmanned environmental sanitation equipment. The automatic obstacle-avoiding unmanned road sweeper comprises a vehicle-mounted heterogeneous multi-sensor sensing unit, a main controller, a chassis, a linkable cleaning operation unit and a safety redundancy unit; the main controller is internally provided with a coverage rate calibration module, a two-dimensional sensing processing module, a coverage rate priority obstacle-avoiding decision module, a binding type path planning module and a synchronous linkage control module which are sequentially connected by signals. The automatic obstacle-avoiding unmanned road sweeper can simultaneously consider high-definition scanning coverage rate and high obstacle-avoiding safety, linkage of obstacle-avoiding driving and cleaning operation, and has a multi-stage safety redundancy design.

Description

Technical Field

[0001] This application relates to the field of unmanned sanitation equipment technology, specifically to an unmanned road sweeper with automatic obstacle avoidance. Background Technology

[0002] With the commercialization of autonomous driving technology, driverless sweepers have gradually replaced manual sweeping and become the core equipment for urban sanitation operations. The core performance indicators of driverless sweepers are twofold: first, obstacle avoidance safety, preventing collisions between the vehicle / sweeping equipment and obstacles; and second, sweeping coverage, ensuring that the road surface is swept without blind spots or missed areas.

[0003] Existing road sweepers all follow a design philosophy of "prioritizing obstacle avoidance safety and then considering sweeping coverage," placing collision safety as the sole priority for obstacle avoidance, with sweeping coverage only as a subsequent constraint. This leads to an imbalance between safety and coverage, resulting in situations where excessive detours for safety cause large areas of missed sweeping, or where reduced avoidance distances for coverage lead to vehicle and brush collisions. This fails to simultaneously meet the operational requirements of high coverage and high safety. Furthermore, existing obstacle avoidance control systems only control the vehicle's path, while the sweeping operation system operates independently without coordinated control. During obstacle avoidance, the sweeping brushes may not rise or retract synchronously, easily colliding with obstacles and causing equipment damage. Conversely, premature brush retraction results in large areas of missed sweeping along the detour route, failing to guarantee sweeping coverage during obstacle avoidance. Consequently, there is no deep linkage between vehicle obstacle avoidance and sweeping operations, making it impossible to simultaneously address collision prevention and missed sweeping. In addition, existing road sweepers only have a single emergency braking redundancy, lack a graded safety protection mechanism, and do not have hardware-level fallback protection. When the perception system fails, it is very easy to cause safety accidents, and there are safety hazards in extreme scenarios.

[0004] Therefore, there is an urgent need to develop an unmanned road sweeper with automatic obstacle avoidance to solve the above-mentioned technical problems.

[0005] It should be noted that the above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0006] The aim is to solve the above-mentioned technical problems and provide an unmanned road sweeper that can simultaneously achieve high-definition sweeping coverage and high obstacle avoidance safety, link obstacle avoidance driving and sweeping operation, and has a multi-level safety redundancy design for automatic obstacle avoidance.

[0007] This application relates to an unmanned road sweeper with automatic obstacle avoidance, characterized in that it includes a body, a chassis, a sweeping operation unit, an on-board heterogeneous multi-sensor perception unit, a main controller and a safety redundancy unit, wherein the body is mounted on the chassis; The cleaning unit is installed on the bottom and side of the chassis. The cleaning unit includes dual side brushes, a roller brush and a vacuum suction cup. The cleaning unit is equipped with an electric actuator with vehicle bus communication. The vehicle-mounted heterogeneous multi-sensor sensing unit is fixed to the chassis by a mounting bracket and is used to collect full-dimensional data of the working environment. The signal output terminal of the vehicle-mounted heterogeneous multi-sensor sensing unit is connected to the signal input terminal of the main controller via vehicle Ethernet. The main controller is fixedly installed in the sealed electrical compartment of the chassis. The main controller is equipped with a coverage calibration module, a two-dimensional perception processing module, a coverage priority obstacle avoidance decision module, a bound path planning module, and a synchronous linkage control module that are connected in sequence by signals. The first signal output terminal of the synchronous linkage control module is communicatively connected to the execution unit of the chassis through the vehicle bus, and the second signal output terminal is communicatively connected to the electric actuator through the vehicle bus. The safety redundancy unit is a safety controller that operates independently of the main controller. The safety redundancy unit is connected to the vehicle heterogeneous multi-sensor perception unit, the main controller and the chassis through hard wiring and dual channels of the vehicle bus. The coverage calibration module of the main controller is used to pre-calibrate the target cleaning coverage threshold for this operation, divide the allowable coverage loss range bound to the target coverage threshold, pre-store the obstacle avoidance decision rules, path planning constraint parameters and cleaning operation linkage parameters corresponding to the allowable coverage loss range, and store them in the local storage unit of the main controller. The dual-dimensional perception processing module is used to perform time synchronization, spatial calibration and fusion processing on the data collected by the vehicle-mounted heterogeneous multi-sensor perception unit, perform dual-dimensional calibration on obstacles, and output the collision risk level of the obstacle and the cleaning coverage loss value corresponding to the obstacle to the coverage priority obstacle avoidance decision module. The coverage-priority obstacle avoidance decision module takes the sweeping coverage loss value not exceeding the allowable range as the first priority constraint, matches the corresponding obstacle avoidance decision rule, and generates an obstacle avoidance decision instruction. The bound path planning module is used to call the path planning constraint parameters bound to the obstacle avoidance decision rule, correct the cost function of the path planning algorithm in real time, and generate a collision-free obstacle avoidance driving path that meets the coverage requirements. The synchronous linkage control module is used to synchronously call the cleaning operation linkage parameters bound to the obstacle avoidance decision rules. While the chassis is executing the obstacle avoidance driving path, it synchronously controls the operation status of the cleaning operation unit to complete the closed-loop obstacle avoidance cleaning operation.

[0008] In one specific implementation scheme, the vehicle-mounted heterogeneous multi-sensor sensing unit includes multiple sensors and a multi-sensor hardware synchronization box. The multiple sensors include a main lidar, a blind spot lidar, a millimeter-wave radar, a binocular camera, and an ultrasonic radar. The main lidar is fixedly installed on the top of the vehicle body, and the blind spot lidars are fixedly installed on the front, rear and left and right sides of the vehicle body respectively, for covering the near blind spots of the vehicle body and detecting low obstacles. The millimeter wave radar is fixedly installed on the left and right sides of the front and rear of the vehicle body respectively. The binocular camera is fixedly installed on the front, rear and left and right sides of the vehicle body respectively. The ultrasonic radar is arranged along the circumference of the vehicle body. The trigger signal terminals of the multiple sensors are all connected to the multi-sensor hardware synchronization box, and the time synchronization of the multi-sensor data is achieved through a high-precision time synchronization protocol. The output terminal of the multi-sensor hardware synchronization box is communicatively connected to the dual-dimensional perception processing module of the main controller. The front end of the lens of each of the multiple sensors is equipped with an automatic high-pressure blowing device, a wiper device, and a heating device that are linked to the main controller.

[0009] In one specific implementation scheme, the dual-dimensional perception processing module includes a multi-sensor raw-level pre-fusion submodule, a three-dimensional environment modeling submodule, and a sanitation scene semantic segmentation submodule; The multi-sensor raw-level pre-fusion submodule is used to perform time synchronization and spatial calibration of lidar point cloud, camera image, and millimeter-wave radar echo data, and to filter dust noise and complete image enhancement in rain, fog, and backlight conditions. The 3D environment modeling submodule is used to construct a real-time 3D environment occupancy grid model based on the fused sensor data. The semantic segmentation submodule for sanitation scenarios is equipped with a dedicated dataset segmentation network for sanitation scenarios. This network is used to distinguish between cleanable obstacles and obstacles that need to be avoided, and to perform a two-dimensional calibration of the collision risk level and coverage loss value of obstacles.

[0010] In a specific implementation scheme, the coverage-priority obstacle avoidance decision module sets up 5 levels of obstacle avoidance decision rules, each corresponding to an allowable coverage loss interval. The allowable coverage loss interval is divided into a single-avoidance allowable coverage loss interval and a total cumulative allowable coverage loss interval. The 5 levels of obstacle avoidance decision rules, from lowest to highest, are as follows: Level 1 Sweepable Straight-Ahead Rule: The allowable coverage loss for a single avoidance is 0, and the total allowable coverage loss throughout the process does not exceed the total loss limit corresponding to the target coverage threshold. When matching sweepable obstacles such as fallen leaves and light garbage, a decision instruction is generated to maintain the original sweeping path and continue straight-ahead operation without executing avoidance actions. Level 2 edge-keeping operation rules: The allowable coverage loss for a single avoidance is the first preset value, and the cumulative allowable coverage loss throughout the process is less than or equal to the first proportion of the target total loss limit. Matching static fixed obstacles such as curb stones and fixed trash cans, an edge-keeping cleaning decision instruction is generated to maintain a preset safe distance from the obstacles. Level 3 minor detour rule: The allowable coverage loss for a single avoidance is the second preset value, and the cumulative allowable coverage loss throughout the entire process is less than or equal to the second proportion of the target total loss limit. It matches static scattered obstacles such as loose stones and mineral water bottles on the road surface and generates a minor detour decision instruction with a maximum lateral deviation of no more than the first value. Level 4 Large Detour or Waiting Rule: The allowable coverage loss for a single avoidance is the third preset value, and the total allowable coverage loss is ≤ the third proportion of the target total loss limit. When encountering large obstacles such as vehicles occupying the road or dense pedestrians, and when a small detour cannot avoid the collision, the OCC occupancy grid model is used to determine whether there is compliant detour space. If there is detour space, a large detour decision instruction is generated with the maximum lateral deviation not exceeding the second value. If there is no detour space, a stop and wait decision instruction is generated, and the operation is resumed after the obstacle is removed. Level 5 Emergency Obstacle Avoidance Rules: No coverage loss limit. When the collision time is less than or equal to the emergency collision threshold, it matches a suddenly intruding dynamic target. When triggered, it prioritizes the safety of personnel and vehicles, generates emergency obstacle avoidance decision instructions that coordinate steering and braking, and automatically returns to the original operation path after the emergency scenario is resolved.

[0011] In one specific implementation, the bound path planning module incorporates an improved DWA path planning algorithm, the cost function of which is: Cost(v,ω)=α Speed_cost(v,ω)+β Obstacle_cost(v,ω)+γ Path_deviation_cost(v,ω)+δ Brush_cover_cost(v,ω); Where Cost(v,ω) is the total cost of the predicted trajectory corresponding to the speed combination, Speed_cost(v,ω) is the speed cost term, Obstacle_cost(v,ω) is the obstacle cost term, Path_deviation_cost(v,ω) is the sweeping path deviation penalty term, and Brush_cover_cost(v,ω) is the brushing operation coverage constraint term. v is the vehicle linear velocity, ω is the vehicle angular velocity, α, β, γ, and δ are the weight coefficients of the corresponding sub-cost items, all of which are positive real numbers in the interval [0,1]. Each weight coefficient is a preset value, and the weight coefficients are bound one by one to the 5-level obstacle avoidance decision rules. The weight of the cleaning path deviation penalty item is negatively correlated with the obstacle avoidance level.

[0012] In a specific feasible implementation, the synchronous linkage control module has built-in cleaning operation linkage parameters that are bound one-to-one with the 5-level obstacle avoidance decision rules, specifically: When executing the Level 1 cleanable straight-line rule, the dual side brushes, the roller brush, and the suction cup are controlled to maintain the rated operating state. When executing the Level 2 edge-applying operation rules, the double-sided brushes on the edge-applying side are controlled to automatically adjust their extension length and rotation speed to maintain a preset safe distance from the curb. When executing the Level 3 small detour rule, the double-sided brushes on the detour side are controlled to automatically raise to a preset height, while the double-sided brushes on the non-detour side remain in their rated operating state. When executing the Level 4 large detour or waiting rule, the dual side brushes and the roller brush are controlled to retract to the vehicle body area, the suction cups pause operation, and automatically resume rated operation after returning to the cleaning path; When executing the Level 5 emergency obstacle avoidance rule, control all the cleaning operation units to immediately retract and stop operation, and simultaneously trigger the emergency braking of the chassis.

[0013] In one specific implementation scheme, the chassis is equipped with a load sensor and an adaptive dynamics control unit. The load sensor is used to collect load data of the garbage bin, clean water tank, and sewage tank in real time, and output the real-time total mass m to the adaptive dynamics control unit. The adaptive dynamics control unit is communicatively connected to the synchronous linkage control module and is used to adjust the PID control parameters of steering and braking in real time according to the real-time total mass m and the vehicle speed, so as to realize the tracking control of the planned obstacle avoidance path. The adjustment formula for the PID control parameters is as follows:

[0014] in , , These are the incremental PID stability reference parameters that have been calibrated under vehicle unloaded preparation state, preset rated operating speed, and standard operating road conditions. Let be the vehicle's unloaded curb weight, and m be the vehicle's real-time gross weight after first-order low-pass filtering. (clamp(x, ...)) , ) is the parameter limiting function.

[0015] In one specific implementation scheme, the main controller is further equipped with a multi-source fusion positioning module, which includes a Beidou dual-mode positioning unit, an inertial measurement unit, a laser SLAM unit, and a visual odometry unit, for real-time positioning in complex scenarios; the main controller is also equipped with a vehicle-road cooperative communication unit, for acquiring obstacle information in visual blind spots and realizing obstacle avoidance beyond line of sight.

[0016] In one specific implementation scheme, the safety redundancy unit includes a perception and warning submodule, a deceleration and avoidance submodule, an emergency braking submodule, and a mechanical collision switch submodule arranged sequentially; the mechanical collision switch is arranged along the circumference of the vehicle body and connected to the braking actuator of the chassis, and immediately cuts off the power output and performs lock-up braking after being triggered.

[0017] In one specific implementation scheme, the main controller further includes a coverage closed-loop compensation module, which is signal-connected to the coverage calibration module, the dual-dimensional perception processing module, the bound path planning module, and the synchronous linkage control module. The execution steps of the coverage closed-loop compensation module are as follows: First, the cumulative actual coverage loss value of the entire operation process is calculated in real time. The cumulative actual coverage loss value = (sum of missed areas caused by all avoidance actions in the entire operation process ÷ total planned cleaning area of ​​this operation) × 100%; Secondly, when the cumulative actual coverage loss value reaches the preset cumulative loss warning threshold, the coverage constraint priority of subsequent obstacle avoidance decisions is automatically increased, and the triggering permission of obstacle avoidance rules with high coverage loss at level 3 and above is locked. Furthermore, when the cumulative actual coverage loss value exceeds the upper limit of the total allowable loss corresponding to the target coverage threshold, based on the environmental grid data output by the dual-dimensional perception processing module, the missed areas are automatically identified and a backtracking and re-sweeping path instruction is generated and sent to the bound path planning module and the synchronous linkage control module to control the vehicle to complete the re-sweeping operation of the missed areas.

[0018] The unmanned road sweeper with automatic obstacle avoidance provided in this application has at least the following beneficial effects: 1. Before operation, the target cleaning coverage threshold is pre-calibrated through the coverage calibration module, and the allowable coverage loss range bound to the threshold is divided. The range is then bound to the obstacle avoidance decision rules, path planning constraint parameters, and cleaning operation linkage parameters. When making obstacle avoidance decisions, the obstacle avoidance strategy is matched with the first priority constraint that the cleaning coverage loss value does not exceed the allowable range, rather than using collision risk as the sole decision-making basis. Through the linkage design of the coverage calibration module and the coverage priority obstacle avoidance decision module built into the main controller, the cleaning coverage is used as a rigid pre-condition constraint for the entire obstacle avoidance process, rather than a post-optimization target after obstacle avoidance is completed. Through the closed-loop design of the pre-condition constraint of coverage, a stable cleaning coverage is achieved in all scenarios, while reducing the risk of obstacle collision. 2. Through the coverage calibration module, dual-dimensional perception processing module, coverage priority obstacle avoidance decision module, binding path planning module, and synchronous linkage control module connected in sequence within the main controller, the obstacle avoidance decision rules, path planning constraint parameters, cleaning operation linkage parameters and coverage loss intervals are bound one by one. The output of the previous module directly serves as the sole input reference for the next module, forming a closed loop of coverage calibration - dual-dimensional perception - coverage priority decision - binding path planning - synchronous linkage control. Each link takes the previous coverage constraint as the core reference and cannot be used separately. This constructs a complete closed-loop control architecture, thereby realizing deep linkage between vehicle obstacle avoidance and cleaning operation, and can simultaneously take into account collision prevention and missed sweeping. 3. Pre-set cleaning operation linkage parameters that are bound to the 5-level obstacle avoidance decision rules. While the vehicle performs the corresponding level of obstacle avoidance action, the working status of the dual side brushes, roller brushes and vacuum nozzles is synchronously controlled through the vehicle bus. The driving control of the chassis and the execution control of the cleaning operation unit are deeply coupled through the synchronous linkage control module. This avoids the risk of equipment damage caused by the collision between the cleaning operation unit and obstacles during obstacle avoidance, and also reduces the missed cleaning range during obstacle avoidance, thus ensuring cleaning coverage while avoiding obstacles. 4. By using load sensors and adaptive dynamic control units installed on the chassis, load data of the garbage bin, clean water tank, and sewage tank are collected in real time. The PID control parameters of steering and braking are adjusted in real time according to load changes, so that the parameters are adaptively matched. This solves the problem of reduced control accuracy caused by load fluctuations, realizes the tracking control of obstacle avoidance path under low-speed conditions, and ensures the accurate execution of obstacle avoidance path. 5. A safety redundancy unit that operates independently of the main controller is set up, and a four-level safety protection mechanism is constructed, which includes perception and early warning, deceleration and avoidance, emergency braking and mechanical collision backup. The mechanical collision switch is connected to the chassis braking actuator and is not affected by the operating status of the main controller and perception system. In extreme scenarios, it can directly trigger lock-up braking to ensure operational safety and eliminate safety hazards in extreme scenarios such as perception failure and main controller downtime. Attached Figure Description

[0019] Figure 1 : A schematic diagram of the structure of an unmanned road sweeper with automatic obstacle avoidance according to an embodiment of this application; Figure 2 This application provides an embodiment of an automatic obstacle avoidance unmanned sweeper's automatic obstacle avoidance sweeping operation flowchart.

[0020] Reference numerals: 1. Chassis; 2. Sweeping unit; 21. Dual side brushes; 22. Roller brush; 23. Vacuum suction cup; 3. Vehicle-mounted heterogeneous multi-sensor sensing unit; 31. Main lidar; 32. Blind spot lidar; 33. Millimeter wave radar; 34. Binocular camera; 35. Ultrasonic radar; 4. Main controller. Detailed Implementation

[0021] Preferred embodiments of this application are described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of this application and are not intended to limit the scope of protection of this application. Those skilled in the art can make adjustments as needed to adapt to specific application scenarios.

[0022] It should be noted that, in the description of this application, the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," and "circumferential," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this application. Furthermore, features defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and should not be construed as indicating or implying relative importance. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0023] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "setting" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0024] The following is in conjunction with the appendix Figures 1 to 2 This application will be described in further detail.

[0025] This application provides an unmanned road sweeper with automatic obstacle avoidance, including a body, chassis 1, sweeping operation unit 2, on-board heterogeneous multi-sensor perception unit 3, main controller 4 and safety redundancy unit, with the body mounted on chassis 1; The cleaning unit 2 is installed on the bottom and side of the chassis 1. The cleaning unit 2 includes dual side brushes 21, roller brushes 22 and vacuum suction cups 23. The cleaning unit 2 is equipped with an electric actuator with vehicle bus communication. The vehicle-mounted heterogeneous multi-sensor sensing unit is fixed to the chassis 1 by a mounting bracket and is used to collect full-dimensional data of the working environment. The signal output terminal of the vehicle-mounted heterogeneous multi-sensor sensing unit is connected to the signal input terminal of the main controller 4 via vehicle Ethernet. The main controller 4 is fixedly installed in the sealed electrical compartment of the chassis 1. The main controller 4 is equipped with a coverage calibration module, a dual-dimensional perception processing module, a coverage priority obstacle avoidance decision module, a bound path planning module and a synchronous linkage control module that are connected in sequence. The first signal output terminal of the synchronous linkage control module is connected to the execution unit of the chassis 1 through the vehicle bus, and the second signal output terminal is connected to the electric actuator through the vehicle bus. The safety redundancy unit is a safety controller that operates independently of the main controller 4. The safety redundancy unit is connected to the vehicle heterogeneous multi-sensor perception unit 3, the main controller 4, and the chassis 1 and the vehicle bus through dual channels. Among them, the coverage calibration module of the main controller 4 is used to pre-calibrate the target cleaning coverage threshold for this operation, divide the allowable coverage loss range bound to the target coverage threshold, pre-store the obstacle avoidance decision rules, path planning constraint parameters and cleaning operation linkage parameters corresponding to the allowable coverage loss range, and store them in the local storage unit of the main controller 4. The dual-dimensional perception processing module is used to perform time synchronization, spatial calibration and fusion processing on the data collected by the vehicle-mounted heterogeneous multi-sensor perception unit 3, perform dual-dimensional calibration on obstacles, and output the collision risk level of the obstacle and the cleaning coverage loss value corresponding to the obstacle to the coverage priority obstacle avoidance decision module. The coverage-first obstacle avoidance decision module takes the cleaning coverage loss value not exceeding the allowable range as the first priority constraint, matches the corresponding obstacle avoidance decision rules, and generates obstacle avoidance decision instructions. The bound path planning module is used to call the path planning constraint parameters bound to the obstacle avoidance decision rules, correct the cost function of the path planning algorithm in real time, and generate a collision-free obstacle avoidance driving path that meets the coverage requirements. The synchronous linkage control module is used to synchronously call the cleaning operation linkage parameters bound to the obstacle avoidance decision rules. While the chassis 1 executes the obstacle avoidance driving path, it synchronously controls the operation status of the cleaning operation unit 2 to complete the closed-loop obstacle avoidance cleaning operation.

[0026] The unmanned road sweeper with automatic obstacle avoidance disclosed in this application has at least the following beneficial effects: (1) Before the operation, the target cleaning coverage threshold is pre-calibrated by the coverage calibration module, and the allowable coverage loss range bound to the threshold is divided. The range is then bound to the obstacle avoidance decision rules, path planning constraint parameters, and cleaning operation linkage parameters. When making obstacle avoidance decisions, the obstacle avoidance strategy is matched with the first priority constraint that the cleaning coverage loss value does not exceed the allowable range, rather than using collision risk as the sole decision basis. Through the linkage design of the coverage calibration module and the coverage priority obstacle avoidance decision module built into the main controller 4, the cleaning coverage is used as a rigid constraint condition in the entire obstacle avoidance process, rather than a post-optimization target after obstacle avoidance is completed. In non-emergency risk scenarios, the obstacle avoidance strategy is matched with coverage compliance as a premise. Only in emergency risk scenarios is safety prioritized. This achieves a full-scenario balance between obstacle avoidance safety and cleaning coverage. Through the closed-loop design of the coverage pre-constraint, a stable cleaning coverage is achieved in the full-scenario operation, while reducing the obstacle collision risk.

[0027] (2) The coverage calibration module, dual-dimensional perception processing module, coverage priority obstacle avoidance decision module, binding path planning module and synchronous linkage control module are sequentially connected in the main controller 4. The obstacle avoidance decision rules, path planning constraint parameters, cleaning operation linkage parameters and coverage loss interval are bound one by one. The output of the previous module is directly used as the only input reference of the next module, forming a closed loop of coverage calibration-dual-dimensional perception-coverage priority decision-binding path planning-synchronous linkage control. Each link takes the previous coverage constraint as the core reference and cannot be used separately. A complete closed-loop control architecture is constructed, thereby realizing the deep linkage between vehicle obstacle avoidance driving and cleaning operation, which can simultaneously take into account collision prevention and missed sweeping.

[0028] (3) Set up cleaning operation linkage parameters in advance that are bound to the five-level obstacle avoidance decision rules. While the vehicle performs the corresponding level of obstacle avoidance action, the working status of the dual side brushes 21, roller brushes 22 and vacuum suction nozzles 23 are controlled synchronously through the vehicle bus. The driving control of the chassis 1 and the execution control of the cleaning operation unit 2 are deeply coupled through the synchronous linkage control module. This avoids the risk of equipment damage when the cleaning operation unit 2 collides with obstacles during the obstacle avoidance process, and also reduces the missed sweeping range during the obstacle avoidance process. The cleaning coverage rate is guaranteed while avoiding obstacles.

[0029] (4) Load data of garbage bin, clean water bin and sewage bin are collected in real time through the load sensor and adaptive dynamic control unit set in chassis 1. The PID (proportional integral derivative) control parameters of steering and braking are adjusted in real time according to the load changes, so that the parameters are adaptively matched, which solves the problem of reduced control accuracy caused by load fluctuation, realizes the tracking control of obstacle avoidance path under low speed conditions, and ensures the accurate execution of obstacle avoidance path.

[0030] (5) A safety redundancy unit that operates independently of the main controller 4 is set up, and a four-level safety protection mechanism of perception warning, deceleration avoidance, emergency braking and mechanical collision fallback is constructed. The mechanical collision switch is connected to the braking actuator of the chassis 1 and is not affected by the operating status of the main controller 4 and the perception system. In extreme scenarios, it can directly trigger the lock-up braking to ensure operational safety and eliminate safety hazards in extreme scenarios such as perception failure and main controller 4 shutdown.

[0031] In the embodiments of this application, such as Figure 1 As shown, an unmanned road sweeper with automatic obstacle avoidance includes a body, chassis 1, sweeping operation unit 2, on-board heterogeneous multi-sensor sensing unit 3, main controller 4, and safety redundancy unit. The body is mounted on chassis 1. Chassis 1 is a pure electric chassis specifically developed for low-speed sanitation operation scenarios. Chassis 1 is a drive-by-wire chassis, with built-in drive-by-wire steering unit, drive-by-wire braking unit, and drive-by-wire drive unit, all using low-speed vehicle drive-by-wire actuators and communicating via on-board bus. A sealed garbage bin is located at the rear of chassis 1, and two cantilever beam-type weight load sensors are installed at the bottom of the garbage bin for real-time collection of empty and full load data of the garbage bin, clean water tank, and sewage tank. A sealed electrical compartment with IP67 protection rating is located in the middle of chassis 1, and mounting rails are installed inside the electrical compartment for fixing and installing electrical equipment such as the main controller 4, multi-sensor hardware synchronization box, and low-voltage power distribution unit. The chassis 1 is also equipped with an adaptive dynamics control unit, which is an executable program loaded in the VCU (vehicle controller) of the chassis 1. It communicates in real time with the synchronous linkage control module of the main controller 4 through the vehicle bus. It is used to adjust the PID control parameters of steering and braking in real time according to the real-time load data of the garbage bin, clean water tank, and sewage tank and the vehicle speed, so as to achieve high-precision tracking of the planned path.

[0032] The cleaning unit 2 is bolted to the bottom and left and right sides of the chassis 1. It includes dual side brushes 21, roller brushes 22, and suction nozzles 23. Each dual side brush 21 consists of two side brushes, one on the left and one on the right. Each actuator is equipped with an independent electric actuator with a vehicle bus communication interface, supporting remote independent control of the lifting and retraction, speed adjustment, and start / stop control of each actuator. Based on this hardware configuration, the synchronous linkage control module can adjust the extension length, speed, and lifting height of the corresponding side brushes differently for different obstacle avoidance scenarios. This allows for precise adjustment of the extension length, speed, and lifting height of the corresponding side brushes to reduce missed areas. Among them, the maximum lifting height of the dual side brushes 21 is 10cm, the maximum lateral extension length is 30cm, the rated speed is 120r / min, and the speed adjustment range is 0-180r / min. The extension length, lifting height and speed can be adjusted in real time according to the control command; the rated speed of the roller brush 22 is 180r / min, and the maximum lifting height is 15cm; the vacuum suction cup 23 is connected to the vehicle-mounted high-pressure fan, and the opening degree and start and stop of the air damper can be adjusted according to the control command.

[0033] The vehicle-mounted heterogeneous multi-sensor sensing unit 3 is installed at a preset mounting position on the chassis 1. The vehicle-mounted heterogeneous multi-sensor sensing unit 3 includes multiple sensors. The specific sensor configuration, installation parameters, and connection methods in this embodiment are as follows: (1) The main lidar can be a mechanical lidar, which is fixed to the middle of the roof of the vehicle body by a mounting bracket for medium and long distance environmental three-dimensional point cloud perception. For example, a 16-line mechanical lidar is used, with the installation height set to about 1.8m above the ground. The detection range covers the vehicle's perimeter of 5m-50m, and the data output interface is vehicle Ethernet.

[0034] (2) Set up multiple blind spot lidars and install them at the front, rear and left and right sides of the vehicle bumper. For example, use 4 solid-state blind spot lidars with an installation height of about 0.3m above the ground to cover the near blind spot within a range of 0.05m-5m of the vehicle body, and at the same time achieve accurate detection of low obstacles at the 10cm level on the road surface. The data output interface is vehicle Ethernet.

[0035] (3) Set up multiple millimeter-wave radars for dynamic obstacle detection in adverse weather conditions. For example, use four 77GHz millimeter-wave radars with a ranging range of 0.2m-100m for dynamic obstacle detection in adverse weather conditions such as rain, fog, and dust, to compensate for the performance degradation of lidar in adverse scenarios. The data output interface is the vehicle bus.

[0036] (4) Visible light and infrared binocular cameras 34 are four vehicle-mounted binocular cameras, which are fixedly installed on the mounting brackets on the front, rear and left and right sides of the vehicle. The visible light camera has a resolution of 1920×1080, and the infrared camera is adapted to low-light operation scenarios at night. It is used for semantic classification of obstacles and visual odometry positioning. The data output interface is vehicle Ethernet.

[0037] (5) Multiple ultrasonic radars are arranged along the circumference of the vehicle body for near-range obstacle detection. For example, 12 ultrasonic radars can be arranged with an adjacent radar spacing of about 30cm. The ranging range covers 0.05m-3m and is used for near-range obstacle detection within 3m of the vehicle body, covering the entire blind spot of the vehicle body. The data output interface is the vehicle bus.

[0038] (6) The multi-sensor hardware synchronization box is an on-board multi-sensor time synchronization box, which is fixedly installed in the electrical compartment. The trigger signal terminals of all sensors are connected to the trigger interface of the hardware synchronization box. Through the high-precision time synchronization protocol, the time synchronization of all sensor data is realized. The output terminal of the hardware synchronization box is connected to the main controller 4 via the on-board Ethernet.

[0039] (7) Sensor auxiliary devices: All lidar and camera lenses are equipped with an automatic high-pressure blowing device, a miniature wiper device and a heating diaphragm. The air inlet of the automatic high-pressure blowing device is connected to the high-pressure blower for cleaning, and the blowing frequency is linked to the blower speed to clean the dust and water mist on the lens surface in real time; the heating diaphragm prevents the lens from frosting or fogging; the miniature wiper device automatically starts when water stains are detected on the lens surface to ensure the lens is clean.

[0040] The main controller 4 adopts industrial-grade vehicle-mounted control equipment, possessing adaptability to harsh outdoor environments. Its housing has a high level of protection and is bolted to the sealed electrical compartment of the chassis. The main controller is equipped with hardware resources adapted for vehicle control, pre-installed with a vehicle-mounted embedded operating system and a modular operating environment. The coverage calibration module, multi-dimensional perception processing module, coverage-priority obstacle avoidance decision-making module, adaptive path planning module, and synchronous linkage control module within the device are all developed using a modular program architecture. Each functional unit achieves data transmission and logical linkage through distributed signal interaction.

[0041] Each functional module relies on a hierarchical communication mechanism to achieve time-series coordination and data interaction. Progressive signal associations are formed between functional units. The results of the pre-processing can serve as the basis for the subsequent control, enabling the cleaning coverage constraint to run through the entire process of environmental perception, obstacle avoidance decision-making, path planning and operation linkage. At the same time, it facilitates the later functional iteration and modular maintenance of the equipment.

[0042] Meanwhile, the main control unit integrates a multi-source fusion positioning module and an environmental collaborative communication unit, which can coordinate with external environmental information to complete obstacle avoidance. Among them, the multi-source fusion positioning module integrates multiple types of positioning sensor units, fuses multi-source spatial perception data, and adopts a distributed fusion architecture to achieve stable positioning in complex scenarios; During the positioning process, the dynamic data output in real time by the inertial sensing unit is used as the basis for prediction. A dynamic correction model is constructed by combining multiple parameters such as equipment spatial position, running attitude, and sensing deviation. In addition, multiple types of real-time data such as satellite positioning, space environment perception, and visual positioning are used as correction references. Real-time positioning correction and error compensation are completed by relying on multi-source data fusion algorithms. During operation, the equipment can detect the environmental quality of satellite positioning signals in real time and adaptively switch positioning modes based on conditions such as signal stability and environmental interference. When the satellite signal environment is good, it relies on satellite positioning data as the main positioning reference. In weak signal scenarios such as complex obstruction and environmental interference, it automatically switches to the environmental perception and visual positioning fusion mode to ensure continuous and stable positioning in all scenarios.

[0043] Through the loosely coupled fusion algorithm described above, real-time positioning can be achieved in weak GNSS signal scenarios such as building obstruction and tree shade, meeting the requirements of path planning and tracking control, and providing a precise position reference for path planning and tracking control.

[0044] In addition, the main controller 4 also includes a coverage closed-loop compensation module, which is connected to the coverage calibration module, the dual-dimensional perception processing module, the binding path planning module and the synchronous linkage control module respectively. The execution steps of the coverage closed-loop compensation module are as follows: First, the cumulative actual coverage loss value of the entire operation process is calculated in real time. The cumulative actual coverage loss value = (sum of missed areas caused by all avoidance actions in the entire operation process ÷ total planned cleaning area of ​​this operation) × 100%; Secondly, when the cumulative actual coverage loss reaches the preset cumulative loss warning threshold, the coverage constraint priority of subsequent obstacle avoidance decisions is automatically increased, and obstacle avoidance rules that estimate a single coverage loss exceeding the remaining total allowable loss limit (such as level 3 and above rules) are prohibited from being triggered. For obstacles that are prohibited from being triggered, the system will process them according to a higher-level rule that conforms to the coverage constraint (such as level 2 edge-keeping rules). If no rules are available, the system will stop and wait and report the issue.

[0045] Furthermore, when the cumulative actual coverage loss exceeds the total allowable loss limit corresponding to the target coverage threshold, the system automatically identifies the missed areas based on the environmental grid data output by the dual-dimensional perception processing module and generates a backtracking and re-sweeping path instruction. This instruction is then sent to the bound path planning module and the synchronous linkage control module to control the vehicle to complete the re-sweeping operation of the missed areas.

[0046] In addition to the obstacle avoidance control of the main controller, this application also sets up an independent safety redundancy system to further ensure operational safety. The safety redundancy unit is a safety controller, which operates independently of the main controller 4 to prevent the safety protection function from failing when the main controller 4 fails. The safety controller is connected to the vehicle-mounted heterogeneous multi-sensor sensing unit 3, the main controller 4, and the braking actuator of the chassis 1 through hard wiring to achieve dual-channel signal transmission.

[0047] The safety redundancy unit has four levels of safety protection sub-modules arranged sequentially: a perception and early warning sub-module, a deceleration and avoidance sub-module, an emergency braking sub-module, and a mechanical collision switch sub-module. The specific triggering logic and thresholds are as follows: (1) Perception and warning submodule: When an obstacle is detected within a preset distance range in front of the vehicle's driving path, a first-level warning is triggered, and the vehicle is controlled to decelerate to a preset safe speed. The warning distance can be set to 20m, and the vehicle decelerates to about 3km / h. At the same time, a warning prompt is issued through the vehicle voice module. (2) Deceleration and avoidance submodule: When an obstacle enters the preset deceleration distance range in front of the vehicle and the collision time is less than the preset deceleration threshold, the second-level deceleration and avoidance is triggered to control the vehicle to decelerate further; for example, the deceleration distance can be set to 10m, the deceleration threshold can be set to 3s, and the vehicle can be decelerated to about 1km / h. (3) Emergency braking submodule: When an obstacle enters within 3m in front of the vehicle and the collision time TTC≤1s, the three-level emergency braking is triggered, and the braking command is sent directly to the brake-by-wire unit through the hard wire to control the vehicle to stop immediately. (4) Mechanical collision switch sub-module: Mechanical collision switches are arranged at the bottom of the bumpers around the front, rear, left and right sides of the vehicle body. The triggering force of the mechanical collision switch can be set according to the protection requirements. In this embodiment, it can be set to no more than 50N. The collision switch is hard-wired to the brake actuator of the chassis 1. After being triggered, the power output of the whole vehicle is immediately cut off and the locking brake is executed. As the last line of defense, it avoids safety accidents when the main controller 4 and the sensing system fail.

[0048] The emergency braking trigger threshold of the safety redundancy unit is consistent with the level 5 emergency obstacle avoidance rule trigger threshold of the main controller 4. Emergency braking is only triggered in emergency scenarios where TTC≤1s to avoid unnecessary braking intervention. The consistency of the two thresholds can prevent the safety redundancy unit from erroneously triggering emergency braking when the main controller 4 performs obstacle avoidance actions, thus ensuring the smoothness of vehicle control.

[0049] The emergency braking submodule and mechanical collision switch submodule of the fully redundant unit have the highest priority and can interrupt all commands of the main controller 4, while the perception warning and deceleration avoidance submodule only intervenes when the main controller 4 fails or fails to take corresponding obstacle avoidance actions.

[0050] like Figure 2 As shown, this application discloses an unmanned road sweeper with automatic obstacle avoidance. Through five program modules connected by sequential signals within the main controller 4, it achieves a fully closed-loop automatic obstacle avoidance sweeping operation, encompassing coverage calibration, environmental perception, obstacle avoidance decision-making, path planning, and coordinated control. Based on the aforementioned hardware configuration, the core control flow of this application is as follows: S1: Coverage calibration before operation The coverage calibration module is a calibration node in the modular program unit system. It establishes the first topic and is executed before the job is started. The steps are as follows: S1.1: Receive the target cleaning coverage threshold for this operation input by the user through the vehicle touch screen or remote operation and maintenance platform. In this embodiment, the default setting is 99%, and the user can adjust it within the range of 95%-100% according to the operation requirements. S1.2: Based on the target cleaning coverage threshold, divide the 5 levels of allowable coverage loss intervals bound to the threshold. The allowable coverage loss intervals are divided into the allowable coverage loss interval for a single avoidance and the cumulative allowable coverage loss interval for the entire process. The general division rule is as follows: if the target cleaning coverage threshold is S (ranging from 95% to 100%), then the maximum allowable total coverage loss during the entire operation is L = 100% - S; based on the maximum total loss L, intervals are divided and bound one-to-one with the 5-level obstacle avoidance rules: Level 1 section: Single avoidance allows 0% coverage loss, total loss ≤ L; Level 2 section: Allowable coverage loss for a single avoidance ≤ 0.2×L, and cumulative loss over the entire route ≤ 0.5×L; Level 3 section: Allowable coverage loss for a single avoidance ≤ 0.5×L, and cumulative loss over the entire route ≤ 0.8×L; Level 4 section: Single avoidance coverage loss ≤ 1×L, total loss ≤ 1×L; Level 5 range: No coverage loss limit, only used in emergency obstacle avoidance scenarios; In this embodiment, the default target cleaning coverage threshold S=99%, corresponding to an upper limit of the total allowable coverage loss L=1%, therefore the 5-level intervals are divided as follows: Level 1 section: Single avoidance coverage loss is 0%, and the total cumulative loss over the entire process is ≤1%; Level 2 section: Single avoidance coverage loss ≤ 0.2%, total loss ≤ 0.5%; Level 3 section: Single avoidance coverage loss ≤ 0.5%, total loss ≤ 0.8%; Level 4 section: Single avoidance coverage loss ≤1%, total loss ≤1%; Level 5 zone: No coverage loss limit (only for emergency obstacle avoidance).

[0051] S1.3: Pre-store the obstacle avoidance decision rules, path planning constraint parameters, and cleaning operation linkage parameters that are bound one-to-one with each allowable coverage loss interval. Store the above-mentioned bound parameters in the local storage unit of the main controller 4 for subsequent modules to call. S1.4: Send a calibration completion signal to other modules of the main controller 4, and the vehicle enters the work-ready state.

[0052] In this embodiment, by pre-calibrating the coverage rate before the operation, the entire link of target coverage rate, allowable loss range, obstacle avoidance rules, and control parameters is bound together. This provides a unified pre-benchmark and constraint for the coverage rate priority control of the entire subsequent process, ensuring that subsequent obstacle avoidance decisions, path planning, and cleaning control are all executed around the preset target coverage rate, thus locking in the compliance of the cleaning coverage rate from the start of the operation. At the same time, it supports users to flexibly adjust the target coverage rate threshold according to the operation requirements, adapting to the differentiated operation requirements of different scenarios such as main roads, parks, and back streets and alleys.

[0053] S2: Environmental Perception and Two-Dimensional Obstacle Calibration The dual-dimensional perception processing module is a perception fusion node in the modular program unit system. It establishes a second topic, receives synchronized multi-source perception data output from the multi-sensor hardware synchronization box, and performs the following processing steps: S2.1: Multi-sensor data time synchronization and spatial calibration. First, the time synchronization of all sensor data is achieved through the trigger signal of the multi-sensor hardware synchronization box. Then, the spatial joint calibration of the multi-sensor is performed. Based on the hand-eye calibration principle, the multi-sensor spatial joint calibration method uses the main lidar 31 as the calibration reference. The external parameters of the blind spot lidar 32, millimeter-wave radar 33, and binocular camera 34 relative to the main lidar 31 are calibrated respectively, namely the rotation matrix R and the translation vector T. The data of all sensors are uniformly transformed into the vehicle coordinate system. The origin of the vehicle coordinate system is the rear axle center of the vehicle. The X-axis is forward along the vehicle driving direction, the Y-axis is laterally to the left along the vehicle, and the Z-axis is perpendicular to the ground and upward.

[0054] S2.2: Multi-sensor raw-level pre-fusion and adverse scene data enhancement. For adverse working conditions such as dust, rain, fog, and backlight in sanitation operations, multi-source data is pre-processed and fused to ensure the accuracy of perception data under all working conditions and provide reliable input for obstacle avoidance decision-making.

[0055] (1) For the point cloud denoising of LiDAR in dusty scenarios, a point cloud denoising network based on deep learning can be used to denoise the LiDAR point cloud.

[0056] In one implementation, a deep learning-based point cloud denoising network can be used to denoise LiDAR point clouds. This network combines multiple consecutive frames of point cloud data and identifies and removes dust noise by analyzing the spatiotemporal motion consistency of the point clouds, while retaining the true obstacle point clouds. For example, a spatiotemporal consistency constraint branch can be added after the feature extraction branch of a point cloud feature extraction network. This branch performs convolution processing on the local features of consecutive frames and outputs the spatiotemporal consistency weight for each point. By comparing this weight with a preset threshold, static obstacle point clouds and dynamic dust noise can be effectively distinguished.

[0057] (2) For complex operating scenarios such as rain, fog, and backlight, adaptive image enhancement technology is used to optimize the images captured by the binocular camera, which can effectively suppress the image quality degradation caused by environmental interference.

[0058] For images in complex scenes such as rain, fog, and backlighting, image enhancement algorithms can be applied. This algorithm can simultaneously analyze the local details and global illumination distribution of an image, and improve image blurring, overexposure, or underexposure through reconstruction, outputting a sharper image and thus improving the accuracy of subsequent target recognition.

[0059] In the overall architecture, the encoding end has the ability to extract features at multiple scales, which can mine image detail information; the decoding end combines a cross-layer fusion mechanism to integrate multi-level feature information and output an image with optimized image quality.

[0060] During image optimization processing, the image specifications acquired by the device can be uniformly adapted to complete image normalization. After multi-dimensional enhancement processing, the image is further combined with contrast adaptive optimization technology to improve the detail of the image, reduce interference from complex environments, and stably output the processed image to the subsequent multi-source data fusion stage, ensuring the accuracy of environmental recognition under complex working conditions.

[0061] The trained algorithm can significantly improve image clarity in adverse scenes, ensuring semantic segmentation accuracy in such conditions. It should be noted that the effectiveness of the enhancement algorithm is limited by the signal-to-noise ratio of the input image; the enhancement effect may diminish in extremely heavy fog or when the lens is severely dirty. In such cases, automatic sensor cleaning devices (such as automatic high-pressure blowing or windshield wipers) can be used to ensure input quality.

[0062] (3) The preprocessed lidar point cloud, millimeter-wave radar 33 echo and camera image data are uniformly synchronized in time and spatially calibrated, and multimodal feature fusion technology is used to generate the fused environmental feature map.

[0063] Feature processing branches corresponding to different sensor data are constructed separately, and each branch independently completes the extraction of basic features for a single modality. Through a cross-modal feature fusion module, the fusion weights of each modality feature are dynamically adjusted according to the working environment. In normal working environments, the basic weight ratio is configured according to the sanitation scene recognition requirements. In harsh working conditions such as rain, fog, and dust, the weight of sensor data with stronger environmental adaptability is automatically increased, and the weight ratio of other modal features is adjusted simultaneously to achieve adaptive feature fusion in complex environments.

[0064] By uniformly converting the weighted and fused multimodal features into global bird's-eye view spatial features in the vehicle coordinate system and outputting them to the subsequent 3D environment modeling and target recognition stages, the accuracy of obstacle detection in harsh scenarios can be effectively improved.

[0065] S2.3: 3D environment modeling. Based on the fused environmental feature map, a real-time updated 3D environment occupancy grid model is constructed, dividing the work area into three states: occupied, idle, and unknown, providing accurate environmental spatial information for subsequent obstacle avoidance path planning.

[0066] S2.4: Dedicated semantic segmentation and dual-dimensional obstacle calibration for sanitation scenarios. It features a dedicated target recognition module for sanitation scenarios, employing a lightweight target recognition architecture optimized for sanitation operation characteristics. By adding a dedicated sanitation classification unit, environmental targets are divided into two main categories: cleanable obstacles and avoidable obstacles. The target subcategories can be further refined according to actual operational needs. Lightweight optimization techniques reduce computational load while meeting the real-time processing requirements of the vehicle control unit.

[0067] The target recognition module outputs the category information, boundary information, region information, and confidence level of the target, and determines the target with a confidence level that meets the preset determination criteria as a valid target. Among them: Swept obstacles include targets such as fallen leaves, paper scraps, and light debris that can be removed through sweeping operations, and are determined not to require avoidance; Obstacles that need to be avoided include targets such as curbs, fixed facilities, parked vehicles, pedestrians, and hard debris that cannot be removed through sweeping operations, and are determined to require obstacle avoidance actions.

[0068] The training dataset is a large number of环卫场景图像 (sanitation scene images) collected in actual sanitation operations, with instance segmentation masks of various targets annotated; a conventional optimizer is used for training. After sufficient training, the network can complete target classification and instance segmentation with high accuracy.

[0069] For the detected obstacles that need to be avoided, two-dimensional calibration is performed: The first dimension is the calibration of the collision risk level. Calculate the time to collision TTC (Time-To-Collision) between the obstacle and the vehicle, and divide it into four levels according to the TTC value: low risk (TTC>5s), medium risk (2s<TTC≤5s), high risk (TTC≤2s), and emergency risk (TTC≤1s); The second dimension is the calibration of the cleaning coverage loss value. Calculate the cleaning coverage loss value that will be caused if this obstacle is avoided. The calculation formula is: Coverage loss value = (area not swept due to avoiding this obstacle ÷ total planned cleaning area of this operation) × 100%, Among them, the area not swept is calculated based on the size of the obstacle, the cleaning width of the vehicle, and the planned cleaning path. Based on the three-dimensional bounding box size of the obstacle and the predicted coverage range of the brush disk of the vehicle on the candidate obstacle avoidance trajectory, calculate the number of grid cells that are not covered by the brush disk and belong to the planned cleaning area on the grid map, so as to obtain the area not swept.

[0070] The calculation of the above cleaning coverage loss value is a pre-evaluation of the possible area not swept caused by future obstacle avoidance actions based on the current perception information. For example, according to the bounding box size of the obstacle, the currently planned cleaning path, and the preset maximum lateral deviation distance or safe following distance under the corresponding obstacle avoidance level, a candidate obstacle avoidance trajectory can be pre-simulated, and the difference between the cleaning coverage area of this simulated trajectory and the cleaning coverage area of the original planned path can be calculated, and this is used as the pre-estimated value of the cleaning coverage loss corresponding to this obstacle.

[0071] S2.5: Output the two-dimensional calibration results of the obstacle to the coverage priority obstacle avoidance decision module through the second topic of the modular program unit;

[0072] S3: Coverage-Prioritized Tiered Obstacle Avoidance Decision The coverage-priority obstacle avoidance decision-making module is a decision node in the modular program unit system. It establishes a third topic, receives the two-dimensional calibration results, and executes the following decision-making steps: S3.1: Receive the two-dimensional calibration results of obstacles, first determine the emergency risk level of the obstacles, if the collision time TTC of the obstacles is ≤1s, directly match the level 5 emergency obstacle avoidance rule, generate emergency obstacle avoidance decision instructions, without considering coverage loss; S3.2: If the obstacle is not an emergency risk, the first priority constraint for obstacle avoidance decision is that the sweep coverage loss value does not exceed the allowable range corresponding to the target coverage threshold. Calculate the cumulative coverage loss value after avoiding the obstacle, match the above 5-level obstacle avoidance decision rules, and generate the corresponding obstacle avoidance decision instruction.

[0073] The allowable coverage loss range is divided into a single obstacle avoidance allowable coverage loss range and a total allowable coverage loss range. The single obstacle avoidance allowable coverage loss range is the maximum coverage loss allowed to occur when avoiding a single obstacle. The total allowable coverage loss range is the upper limit of the total coverage loss allowed throughout the entire operation. Specifically, as an example configuration, it is applicable to scenarios where the target cleaning coverage threshold is 99%, and the rules can be set as follows: (1) Level 1 Sweepable Straight-Ahead Rule: The allowable coverage loss for a single avoidance is 0, and the total allowable coverage loss throughout the process does not exceed the upper limit of the total loss corresponding to the target coverage threshold. Match sweepable obstacles such as fallen leaves and light garbage, generate a decision instruction to maintain the original sweeping path and proceed straight-ahead, and do not execute avoidance actions; (2) Level 2 edge-following operation rules: The allowable coverage loss for a single avoidance is ≤0.2%, and the total allowable coverage loss is ≤50% of the target total loss limit. Matching static fixed obstacles such as curb stones and fixed trash cans, generating edge-following cleaning decision instructions that maintain a preset safe distance (e.g., 5-10cm) from the obstacles; (3) Level 3 small detour rule: The allowable coverage loss for a single avoidance is ≤0.5%, and the total allowable coverage loss is ≤80% of the target total loss limit. It matches static scattered obstacles such as loose stones and mineral water bottles on the road surface and generates a small detour decision instruction with a maximum lateral deviation of no more than the preset value (e.g., 30cm). (4) Level 4 Large Detour or Waiting Rules: The allowable coverage loss for a single avoidance is ≤1%, and the total allowable coverage loss is ≤100% of the target total loss limit. When encountering large obstacles such as vehicles occupying the road or dense pedestrians, and when small detours cannot avoid collisions, the OCC occupancy grid model is used to determine whether there is compliant detour space. If there is detour space, a large detour decision instruction with a maximum lateral deviation of no more than a preset value (e.g., 100cm) is generated. If there is no detour space, a stop and wait decision instruction is generated. Operations are resumed after the obstacle leaves. (5) Level 5 Emergency Obstacle Avoidance Rule: There is no coverage loss limit. When the collision time TTC is less than or equal to the emergency collision threshold, a sudden intrusion of a dynamic target is matched. When triggered, the safety of personnel and vehicles is prioritized, and an emergency obstacle avoidance decision command with coordinated steering and braking is generated. After the emergency scenario is resolved, the original operation path is automatically returned. The emergency collision threshold can be set according to the operation scenario and vehicle braking performance. For example, it can be set to 1 second.

[0074] It should be noted that the above values ​​(0.2%, 0.5%, 1%, 5-10cm, 30cm, 100cm, 1s) are only examples. In actual applications, they can be adjusted according to factors such as target coverage requirements, vehicle size, and operating scenario.

[0075] When multiple obstacles exist within the perception range, the obstacle with the highest collision risk level or the largest expected coverage loss value is used as the primary basis for the current decision.

[0076] S3.3: The generated obstacle avoidance decision command is simultaneously output to the bound path planning module and the synchronous linkage control module through the third topic of the modular program unit.

[0077] S4: Bound Obstacle Avoidance Path Planning The bound path planning module is a path planning node in the modular program unit system. It establishes a fourth topic, receives obstacle avoidance decision instructions, and executes the following path planning steps: S4.1: Receive obstacle avoidance decision instructions and retrieve the path planning constraint parameters bound to the decision rule from the local storage unit; S4.2: An improved DWA (Dynamic Window Approach) path planning algorithm is adopted. Based on the BEV (Bird's-eye View) + OCC (Occupied Grid) 3D environment occupancy grid model, the cost function of the algorithm is dynamically corrected in real time to generate a collision-free smooth obstacle avoidance driving path that meets the coverage constraint requirements.

[0078] In this embodiment, the cost function of the improved DWA algorithm is: Cost(v,ω)=α Speed_cost(v,ω)+β Obstacle_cost(v,ω)+γ Path_deviation_cost(v,ω)+δ Brush_cover_cost(v,ω); Where Cost(v,ω) is the total cost of the predicted trajectory corresponding to this velocity combination, v is the linear velocity of the vehicle, and ω is the angular velocity of the vehicle. Speed_cost(v,ω) is the speed cost term, used to constrain the vehicle's speed and ensure operational efficiency; The formula for calculating Speed_cost(v,ω) is: Speed_cost(v,ω) = 1 - v is the current linear velocity of the vehicle. In this embodiment, the rated maximum linear velocity for sanitation operations is... =5km / h; the smaller the value of the speed cost item, the closer the vehicle speed is to the rated operating speed, and the higher the operating efficiency. Obstacle_cost(v,ω) is the obstacle cost term, used to constrain the safe distance between the vehicle and the obstacle to ensure obstacle avoidance safety; The formula for calculating Obstacle_cost(v,ω) is: Obstacle_cost(v,ω) = 1 - ,in To predict the minimum distance between a vehicle and the nearest obstacle on its trajectory, The preset safe distance for the corresponding obstacle avoidance level is 1-2. =30cm, obstacle avoidance level 3-4 =50cm, Level 5 obstacle avoidance =100cm; The smaller the obstacle cost value, the farther the distance between the vehicle and the obstacle, and the higher the obstacle avoidance safety. Path_deviation_cost(v,ω) is the cleaning path deviation penalty term, which is used to constrain the lateral deviation of the vehicle from the original planned cleaning path to ensure cleaning coverage. The formula for calculating Path_deviation_cost(v,ω) is: Path_deviation_cost(v,ω) = Where Δd is the lateral deviation distance between the predicted trajectory endpoint and the original planned cleaning path. This represents the maximum lateral deviation under the corresponding obstacle avoidance level; the larger the value of the path deviation penalty, the greater the lateral deviation of the vehicle and the higher the loss of sweeping coverage. Brush_cover_cost(v,ω) is the brush operation coverage constraint term, which is used to constrain the coverage range of the cleaning brush and ensure the cleaning coverage rate of non-detour areas; The formula for calculating Brush_cover_cost(v,ω) is: Brush_cover_cost(v,ω) = 1 - , To predict the actual cleaning coverage area of ​​the dual-side brush 21 on the trajectory, This represents the rated cleaning coverage area of ​​the original planned route; the smaller the value of the brushing operation coverage constraint, the closer the cleaning coverage area is to the planned value, and the higher the cleaning coverage rate of the non-detour area. α, β, γ, and δ are the weight coefficients of the corresponding sub-cost terms, all of which are positive real numbers in the interval [0,1], satisfying α+β+γ+δ=1. The weight coefficients are bound one-to-one with the 5-level obstacle avoidance decision rules, and the specific binding parameters are shown in the table below:

[0079] As can be seen from the table above, the weight γ of the cleaning path deviation penalty item is negatively correlated with the obstacle avoidance level. As the obstacle avoidance level increases from level 1 to level 5, the value of γ gradually decreases from 0.4 to 0.1. That is, the higher the obstacle avoidance level, the more relaxed the constraint on the lateral deviation of the vehicle from the original planned cleaning path. Under the low level of obstacle avoidance, the path deviation is strictly limited to ensure the cleaning coverage rate, while under the high level of obstacle avoidance, the obstacle avoidance safety is given priority.

[0080] The weighting coefficients mentioned above are merely examples. In practical applications, they can be adaptively adjusted based on factors such as the work scenario, vehicle dynamics, and hardware computing power. For instance, in a narrow park setting, the speed weight α can be appropriately reduced to improve obstacle avoidance safety; in a wide main road setting, the path deviation weight γ can be increased to ensure cleaning coverage.

[0081] By dynamically adjusting the cost function weights bound to the obstacle avoidance decision rules, in low-level obstacle avoidance scenarios, the lateral deviation of the vehicle is strictly limited to ensure cleaning coverage by amplifying the weights of path deviation and brush coverage. In high-level obstacle avoidance scenarios, the obstacle weights are amplified to prioritize the safety of vehicles and personnel, achieving adaptive adaptation of the path planning algorithm to different scenarios. Simultaneously, the newly added brush coverage constraint further reduces the missed area in non-detour zones, improving overall cleaning coverage. When the brush state changes, the synchronous linkage control module feeds back the state (such as lift height and rotation speed) to the path planning module in real time to refresh the prediction model of the `Brush_cover_cost(v,ω)` term, achieving dynamic coordination.

[0082] S4.3: The generated smooth obstacle avoidance driving path is output to the synchronous linkage control module through the fourth topic of the modular program unit.

[0083] S5: Synchronous Control of Obstacle Avoidance and Cleaning The synchronous linkage control module is a linkage control node in the modular program unit system. It establishes the fifth topic, receives obstacle avoidance decision commands and planned obstacle avoidance driving paths, and executes the following synchronous control steps: S5.1: Receives the planned obstacle avoidance driving path and sends the path tracking command to the adaptive dynamics control unit of chassis 1 via the vehicle bus. The adaptive dynamics control unit outputs the real-time total mass m to the adaptive dynamics control unit based on the real-time load data of the garbage bin, clean water tank, and sewage tank, and the vehicle speed. The adaptive dynamics control unit communicates with the synchronous linkage control module to adjust the steering and braking PID (proportional-integral-derivative) control parameters in real time according to the real-time total mass m and driving speed, thereby achieving tracking control of the planned obstacle avoidance path. The adjustment formula for the PID control parameters is:

[0084] in , , These are incremental PID stability reference parameters that have been calibrated under vehicle unloaded preparation conditions, rated operating speeds of 3-5 km / h, and on smooth, hardened road surfaces. Let be the vehicle's unloaded curb weight, and m be the vehicle's real-time gross weight after first-order low-pass filtering. (clamp(x, ...)) , The parameter limiting function enables high-precision tracking and control of the planned path, ensuring the accurate execution of the planned obstacle avoidance path and further stabilizing the cleaning coverage rate.

[0085] The first-order low-pass filtering of the vehicle's real-time total mass m employs an adaptive filtering algorithm, with a cutoff frequency set to 1Hz and a sampling frequency set to 100Hz. The filter recursive formula is: =a +(1-a) ; This is the filtered total vehicle mass data at the current moment. This is the raw total mass data collected by the load sensor at the current moment. This is the filtered total vehicle mass data from the previous time step; 'a' is the filtering smoothing coefficient, with a value range of (0,1), and is calculated using the formula: a = ; The filter's -3dB cutoff frequency is set to 1Hz in this embodiment. The sampling period for the load sensor is 0.01s in this embodiment (corresponding to a sampling frequency of 100Hz). By using low-pass filtering, vibration and noise interference from load sensors during vehicle operation is eliminated, ensuring the smoothness of PID parameter adjustment and avoiding vehicle control jitter caused by frequent parameter jumps.

[0086] S5.2: Simultaneously, receive the obstacle avoidance decision command, retrieve the cleaning operation linkage parameters bound to the decision rule from the local storage unit, and while the chassis 1 executes the obstacle avoidance driving path, synchronously control the operation status of the cleaning operation unit 2 through the vehicle bus. The specific linkage control rules are as follows: (1) When executing the Level 1 sweepable straight rule, control the dual side brushes 21, roller brush 22, and vacuum suction cup 23 to maintain the rated operating state, with the dual side brushes 21 rotating at 120 r / min, the roller brush 22 rotating at 180 r / min, and the suction cup damper fully open; (2) When performing the Level 2 edge-fitting operation rules, control the edge-fitting side brush to automatically adjust the extension length and speed. For example, increase the extension length by 5cm and increase the speed to 150r / min to maintain a safe distance of 5cm from the curbstone. The non-edge-fitting side brush maintains the rated operation state. (3) When executing the level 3 small detour rule, the side brush on the detour side is automatically raised and the speed is reduced. In this embodiment, the side brush on the detour side is automatically raised by 5cm and the speed is reduced to 60r / min. The side brush on the non-detour side is kept in the rated working state. The roller brush 22 and the suction cup are kept in the rated working state to reduce the missed area. (4) When executing the Level 4 large detour or waiting rule, control the double side brushes 21 and roller brushes 22 to retract to the vehicle body range, the vacuum suction cup 23 to stop working, the air damper to close, and the vehicle to return to the original planned cleaning path and automatically resume the rated working state. (5) When executing the Level 5 emergency obstacle avoidance rule, control all cleaning operation units 2 to immediately retract to the vehicle body range and stop operation, and simultaneously trigger the emergency braking of chassis 1 through hard line to ensure the safety of vehicles and personnel.

[0087] S5.3: Real-time acquisition of the vehicle's actual driving path and the actual operating status of the sweeping unit 2, feeding back to the coverage calibration module, updating the cumulative coverage loss value in real time, and completing closed-loop control.

[0088] In one embodiment of this application, the example is a vehicle detecting a mineral water bottle-like obstacle in front of it while operating on an urban main road.

[0089] The dual-dimensional perception processing module outputs that the obstacle category is "mineral water bottle", the collision risk level is "low risk" (TTC>5s), and the expected coverage loss value is 0.3% (calculated based on the diameter of the mineral water bottle and the cleaning width).

[0090] The coverage priority obstacle avoidance decision module determines that the current cumulative coverage loss is 0.1% (no avoidance before), and the cumulative loss after this avoidance is 0.4%, which does not exceed the full allowable limit of 1%, and the single loss of 0.3% does not exceed the limit of 0.5% of the level 3 rule. Therefore, it matches the "level 3 minor detour rule".

[0091] The bound path planning module calls the cost function weights (α=0.15, β=0.25, γ=0.3, δ=0.3) corresponding to the level 3 rules to generate an obstacle avoidance path with a maximum lateral deviation of 25cm.

[0092] The synchronous linkage control module executes simultaneously, the chassis travels along the planned path, the side brush on the bypass side automatically rises 4cm and its speed drops to 60r / min; the side brush on the non-bypass side maintains its rated speed. After the vehicle completes the avoidance maneuver, it automatically returns to the original cleaning path, and the side brushes return to their rated state.

[0093] The coverage closed-loop compensation module records the missed area in this scan and updates the cumulative loss value to 0.4%.

[0094] S6: End-to-end security redundancy protection Throughout the operation, the independent safety controller of the safety redundancy unit operates in real time, performing obstacle detection and safety protection in parallel with the main controller 4. It is unaffected by the operating status of the main controller 4 and sequentially performs four levels of safety protection: perception warning, deceleration and avoidance, emergency braking, and mechanical collision. The mechanical collision switch is connected to the braking actuator of the chassis 1 and is unaffected by the operating status of the main controller 4 and the perception system. In extreme scenarios, it can directly trigger the lock-up braking to ensure operational safety and eliminate safety hazards in extreme scenarios such as perception failure and main controller 4 shutdown.

[0095] S7: Coverage Closed-Loop Compensation Control S7.1: Real-time statistics of cumulative coverage loss. It receives feedback from the synchronous linkage control module regarding the actual driving path, avoidance action records, and the operating status of cleaning unit 2. Combined with obstacle size data output by the dual-dimensional perception processing module, it calculates the missed area generated by each avoidance action in real time, accumulating the cumulative actual coverage loss value for the entire operation process. The calculation formula is: Cumulative actual coverage loss value = (Sum of missed areas generated by all avoidance actions throughout the operation process ÷ Total planned cleaning area for this operation) × 100%; S7.2: Coverage constraint priority is dynamically adjusted. When the cumulative actual coverage loss value reaches 80% of the total allowable loss limit corresponding to the target coverage threshold, a constraint upgrade instruction is automatically sent to the coverage priority obstacle avoidance decision module to increase the coverage constraint priority of subsequent obstacle avoidance decisions, lock the triggering permission of level 3 and level 4 high coverage loss obstacle avoidance rules, and only allow the triggering of level 1 and level 2 low loss obstacle avoidance rules to prevent the cumulative loss from exceeding the limit further. S7.3: Backtracking and Re-sweeping of Missed Areas. When the cumulative actual coverage loss exceeds the total allowable loss limit corresponding to the target coverage threshold, based on the BEV+OCC 3D environmental occupancy grid model, the system compares the actual cleaning coverage of the already worked path with the planned cleaning range. It automatically identifies missed areas within the already worked area, generates the shortest backtracking and re-sweeping path command, and sends it to the bound path planning module and synchronous linkage control module. This controls the vehicle to complete the re-sweeping operation along the re-sweeping path after the main task of this operation is completed. After the re-sweeping is completed, the cumulative coverage loss value is updated to ensure that the final operational coverage meets the target threshold requirement. The re-sweeping operation only targets the missed areas within the current operation range and does not affect the coverage statistics of subsequent work sections.

[0096] The technical solutions of this application have been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of this application is obviously not limited to these specific embodiments. Without departing from the principles of this application, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of this application.

Claims

1. An unmanned road sweeper with automatic obstacle avoidance, characterized in that, It includes a vehicle body, chassis, sweeping operation unit, vehicle-mounted heterogeneous multi-sensor perception unit, main controller and safety redundancy unit, with the vehicle body mounted on the chassis; The cleaning unit is installed on the bottom and side of the chassis. The cleaning unit includes dual side brushes, a roller brush and a vacuum suction cup. The cleaning unit is equipped with an electric actuator with vehicle bus communication. The vehicle-mounted heterogeneous multi-sensor sensing unit is fixed to the chassis by a mounting bracket and is used to collect full-dimensional data of the working environment. The signal output terminal of the vehicle-mounted heterogeneous multi-sensor sensing unit is connected to the signal input terminal of the main controller via vehicle Ethernet. The main controller is fixedly installed in the sealed electrical compartment of the chassis. The main controller is equipped with a coverage calibration module, a two-dimensional perception processing module, a coverage priority obstacle avoidance decision module, a bound path planning module, and a synchronous linkage control module that are connected in sequence by signals. The first signal output terminal of the synchronous linkage control module is communicatively connected to the execution unit of the chassis through the vehicle bus, and the second signal output terminal is communicatively connected to the electric actuator through the vehicle bus. The safety redundancy unit is a safety controller that operates independently of the main controller. The safety redundancy unit is connected to the vehicle heterogeneous multi-sensor perception unit, the main controller and the chassis through hard wiring and dual channels of the vehicle bus. The coverage calibration module of the main controller is used to pre-calibrate the target cleaning coverage threshold for this operation, divide the allowable coverage loss range bound to the target coverage threshold, pre-store the obstacle avoidance decision rules, path planning constraint parameters and cleaning operation linkage parameters corresponding to the allowable coverage loss range, and store them in the local storage unit of the main controller. The dual-dimensional perception processing module is used to perform time synchronization, spatial calibration and fusion processing on the data collected by the vehicle-mounted heterogeneous multi-sensor perception unit, perform dual-dimensional calibration on obstacles, and output the collision risk level of the obstacle and the cleaning coverage loss value corresponding to the obstacle to the coverage priority obstacle avoidance decision module. The coverage-priority obstacle avoidance decision module takes the sweeping coverage loss value not exceeding the allowable range as the first priority constraint, matches the corresponding obstacle avoidance decision rule, and generates an obstacle avoidance decision instruction. The bound path planning module is used to call the path planning constraint parameters bound to the obstacle avoidance decision rule, correct the cost function of the path planning algorithm in real time, and generate a collision-free obstacle avoidance driving path that meets the coverage requirements. The synchronous linkage control module is used to synchronously call the cleaning operation linkage parameters bound to the obstacle avoidance decision rules. While the chassis is executing the obstacle avoidance driving path, it synchronously controls the operation status of the cleaning operation unit to complete the closed-loop obstacle avoidance cleaning operation.

2. The unmanned road sweeper with automatic obstacle avoidance according to claim 1, characterized in that, The vehicle-mounted heterogeneous multi-sensor perception unit includes multiple sensors and a multi-sensor hardware synchronization box. The multiple sensors include a main lidar, a blind spot lidar, a millimeter-wave radar, a binocular camera, and an ultrasonic radar. The main lidar is fixedly installed on the top of the vehicle body, and the blind spot lidars are fixedly installed on the front, rear and left and right sides of the vehicle body respectively, for covering the near blind spots of the vehicle body and detecting low obstacles. The millimeter wave radar is fixedly installed on the left and right sides of the front and rear of the vehicle body respectively. The binocular camera is fixedly installed on the front, rear and left and right sides of the vehicle body respectively. The ultrasonic radar is arranged along the circumference of the vehicle body. The trigger signal terminals of the multiple sensors are all connected to the multi-sensor hardware synchronization box, and the time synchronization of the multi-sensor data is achieved through a high-precision time synchronization protocol. The output terminal of the multi-sensor hardware synchronization box is communicatively connected to the dual-dimensional perception processing module of the main controller. The front end of the lens of each of the multiple sensors is equipped with an automatic high-pressure blowing device, a wiper device, and a heating device that are linked to the main controller.

3. The unmanned road sweeper with automatic obstacle avoidance according to claim 2, characterized in that, The dual-dimensional perception processing module includes a multi-sensor raw-level pre-fusion submodule, a three-dimensional environment modeling submodule, and a sanitation scene semantic segmentation submodule. The multi-sensor raw-level pre-fusion submodule is used to perform time synchronization and spatial calibration of lidar point cloud, camera image, and millimeter-wave radar echo data, and to filter dust noise and complete image enhancement in rain, fog, and backlight conditions. The three-dimensional environment modeling submodule is used to construct a real-time three-dimensional environment occupancy grid model based on the fused sensor data. The semantic segmentation submodule for sanitation scenarios is equipped with a dedicated dataset segmentation network for sanitation scenarios. This network is used to distinguish between cleanable obstacles and obstacles that need to be avoided, and to perform a two-dimensional calibration of the collision risk level and coverage loss value of obstacles.

4. The driverless road sweeper with automatic obstacle avoidance according to claim 1, characterized in that, The coverage-priority obstacle avoidance decision module sets up 5 levels of obstacle avoidance decision rules, each corresponding to a one-to-one allowable coverage loss interval. The allowable coverage loss interval is divided into a single-avoidance allowable coverage loss interval and a total cumulative allowable coverage loss interval. The 5 levels of obstacle avoidance decision rules, from lowest to highest, are as follows: Level 1 Sweepable Straight-Ahead Rule: The allowable coverage loss for a single avoidance is 0, and the total allowable coverage loss throughout the process does not exceed the total loss limit corresponding to the target coverage threshold. When matching sweepable obstacles such as fallen leaves and light garbage, a decision instruction is generated to maintain the original sweeping path and continue straight-ahead operation without executing avoidance actions. Level 2 edge-keeping operation rules: The allowable coverage loss for a single avoidance is the first preset value, and the cumulative allowable coverage loss throughout the process is less than or equal to the first proportion of the target total loss limit. Matching static fixed obstacles such as curb stones and fixed trash cans, an edge-keeping cleaning decision instruction is generated to maintain a preset safe distance from the obstacles. Level 3 minor detour rule: The allowable coverage loss for a single avoidance is the second preset value, and the cumulative allowable coverage loss throughout the entire process is less than or equal to the second proportion of the target total loss limit. It matches static scattered obstacles such as loose stones and mineral water bottles on the road surface and generates a minor detour decision instruction with a maximum lateral deviation of no more than the first value. Level 4 Large Detour or Waiting Rule: The allowable coverage loss for a single avoidance is the third preset value, and the total allowable coverage loss is ≤ the third proportion of the target total loss limit. When encountering large obstacles such as vehicles occupying the road or dense pedestrians, and when a small detour cannot avoid the collision, the OCC occupancy grid model is used to determine whether there is compliant detour space. If there is detour space, a large detour decision instruction is generated with the maximum lateral deviation not exceeding the second value. If there is no detour space, a stop and wait decision instruction is generated, and the operation is resumed after the obstacle is removed. Level 5 Emergency Obstacle Avoidance Rules: No coverage loss limit. When the collision time is less than or equal to the emergency collision threshold, it matches a suddenly intruding dynamic target. When triggered, it prioritizes the safety of personnel and vehicles, generates emergency obstacle avoidance decision instructions that coordinate steering and braking, and automatically returns to the original operation path after the emergency scenario is resolved.

5. The driverless road sweeper with automatic obstacle avoidance according to claim 4, characterized in that, The bound path planning module incorporates an improved DWA path planning algorithm, the cost function of which is: Cost(v,ω)=α Speed_cost(v,ω)+β Obstacle_cost(v,ω)+γ Path_deviation_cost(v,ω)+δ Brush_cover_cost(v,ω); Where Cost(v,ω) is the total cost of the predicted trajectory corresponding to the speed combination, Speed_cost(v,ω) is the speed cost term, Obstacle_cost(v,ω) is the obstacle cost term, Path_deviation_cost(v,ω) is the sweeping path deviation penalty term, and Brush_cover_cost(v,ω) is the brushing operation coverage constraint term. v is the vehicle linear velocity, ω is the vehicle angular velocity, α, β, γ, and δ are the weight coefficients of the corresponding sub-cost items, all of which are positive real numbers in the interval [0,1]. Each weight coefficient is a preset value, and the weight coefficients are bound one by one to the 5-level obstacle avoidance decision rules. The weight of the cleaning path deviation penalty item is negatively correlated with the obstacle avoidance level.

6. The driverless road sweeper with automatic obstacle avoidance according to claim 4, characterized in that, The synchronous linkage control module has built-in cleaning operation linkage parameters that are bound one-to-one with the five-level obstacle avoidance decision rules, specifically: When executing the Level 1 cleanable straight-line rule, the dual side brushes, the roller brush, and the suction cup are controlled to maintain the rated operating state. When executing the Level 2 edge-applying operation rules, the double-sided brushes on the edge-applying side are controlled to automatically adjust their extension length and rotation speed to maintain a preset safe distance from the curb. When executing the Level 3 small detour rule, the double-sided brushes on the detour side are controlled to automatically raise to a preset height, while the double-sided brushes on the non-detour side remain in their rated operating state. When executing the Level 4 large detour or waiting rule, the dual side brushes and the roller brush are controlled to retract to the vehicle body area, the suction cups pause operation, and automatically resume rated operation after returning to the cleaning path; When executing the Level 5 emergency obstacle avoidance rule, control all the cleaning operation units to immediately retract and stop operation, and simultaneously trigger the emergency braking of the chassis.

7. The unmanned road sweeper with automatic obstacle avoidance according to claim 1, characterized in that, The chassis is equipped with a load sensor and an adaptive dynamics control unit. The load sensor is used to collect load data of the garbage bin, clean water tank, and sewage tank in real time, and output the real-time total mass m to the adaptive dynamics control unit. The adaptive dynamics control unit is communicatively connected to the synchronous linkage control module and is used to adjust the PID control parameters of steering and braking in real time according to the real-time total mass m and the vehicle speed, so as to realize the tracking control of the planned obstacle avoidance path. The adjustment formula for the PID control parameters is as follows: in , , These are the incremental PID stability reference parameters that have been calibrated under vehicle unloaded preparation state, preset rated operating speed, and standard operating road conditions. Let be the vehicle's unloaded curb weight, and m be the vehicle's real-time gross weight after first-order low-pass filtering. (clamp(x, ...)) , ) is the parameter limiting function.

8. The unmanned road sweeper with automatic obstacle avoidance according to claim 1, characterized in that, The main controller is also equipped with a multi-source fusion positioning module, which includes a Beidou dual-mode positioning unit, an inertial measurement unit, a laser SLAM unit, and a visual odometry unit, used to achieve real-time positioning in complex scenarios; the main controller is also equipped with a vehicle-road cooperative communication unit, used to acquire obstacle information in visual blind spots and achieve over-the-horizon obstacle avoidance.

9. The driverless road sweeper with automatic obstacle avoidance according to claim 1, characterized in that, The safety redundancy unit includes a perception and warning submodule, a deceleration and avoidance submodule, an emergency braking submodule, and a mechanical collision switch submodule arranged in sequence. The mechanical collision switch is arranged along the circumference of the vehicle body and connected to the braking actuator of the chassis. After being triggered, it immediately cuts off the power output and performs lock-up braking.

10. The driverless road sweeper with automatic obstacle avoidance according to claim 4, characterized in that, The main controller also includes a coverage closed-loop compensation module, which is connected to the coverage calibration module, the dual-dimensional perception processing module, the binding path planning module and the synchronous linkage control module respectively. The execution steps of the coverage closed-loop compensation module are as follows: First, the cumulative actual coverage loss value of the entire operation process is calculated in real time. The cumulative actual coverage loss value = (sum of missed areas caused by all avoidance actions in the entire operation process ÷ total planned cleaning area of ​​this operation) × 100%; Secondly, when the cumulative actual coverage loss value reaches the preset cumulative loss warning threshold, the coverage constraint priority of subsequent obstacle avoidance decisions is automatically increased, and the triggering permission of obstacle avoidance rules with high coverage loss at level 3 and above is locked. Furthermore, when the cumulative actual coverage loss value exceeds the upper limit of the total allowable loss corresponding to the target coverage threshold, based on the environmental grid data output by the dual-dimensional perception processing module, the missed areas are automatically identified and a backtracking and re-sweeping path instruction is generated and sent to the bound path planning module and the synchronous linkage control module to control the vehicle to complete the re-sweeping operation of the missed areas.

Images