Target object interaction test system and method for indoor low-speed unmanned equipment

Abstract

The invention provides an indoor low-speed unmanned equipment target object interaction test system. A reproducible indoor test environment is constructed through cooperation of a track lifting system, a target object traction system, a spotlight illumination system and a multi-mode sensor. The track lifting system simulates complex terrains through a modular track (supporting arc / slope rapid replacement), the target object traction system carries a dynamic path planning module, and active avoidance of pedestrians or random behaviors of animals can be simulated in combination with GPS positioning and a PID control algorithm. Adaptive adjustment of brightness and color temperature and switching of a strong backlight / stroboscopic interference mode are realized; after data collected in the testing process is fused through a data analysis module, obstacle avoidance response time, a path optimization rate and a collision force peak value are evaluated in real time by adopting a space-time diagram convolutional network (ST-GCN), and a quantitative performance report is generated. According to the system, the authenticity, flexibility and data reliability of a test scene are remarkably improved, and an efficient verification scheme is provided for active obstacle avoidance and pedestrian protection capability of low-speed unmanned equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of low-speed driverless equipment testing, and particularly relates to an object interaction testing system and method for indoor low-speed driverless equipment. Background Technique

[0002] Existing driverless test systems mainly focus on closed-road testing and open-road testing of intelligent driving vehicles in outdoor high-speed scenarios. For the testing of autonomous driving capabilities in indoor low-speed scenarios, in the existing driverless test systems, the target trolley carrying the object has a large size, which inevitably interferes with the test accuracy of indoor low-speed driverless equipment during testing, and there is a risk of false triggering or collision, so it is not suitable for indoor low-speed scenarios; while through actual human testing, although the cost is low, it cannot guarantee the repeatability and consistency of the test. Therefore, an object interaction testing system for indoor low-speed driverless equipment is needed to replace manual testing. Summary of the Invention

[0003] An object interaction testing system, device and storage medium for indoor low-speed driverless equipment proposed by the present invention can at least solve one of the technical problems in the background technique.

[0004] To achieve the above object, the present invention adopts the following technical solutions:

[0005] An object interaction testing system for indoor low-speed driverless equipment includes a frame, a grating sensor, an orbital lifting system, an object traction system, a spotlight illumination system, a high-definition camera, and a GPS. The frame includes a bottom support frame, a side wall support frame, and a top support frame; the grating sensor is fixed on the side wall support frame; the orbital lifting system is fixed below the orbital platform; the object traction system and the spotlight illumination system are installed on the orbital lifting system; the high-definition camera is fixed at the four corners of the top support frame; the GPS is installed on the test robot.

[0006] The orbital lifting system includes a height adjustment system and an orbital platform; the height adjustment system includes four rotary motors, upper and lower limit devices, four lead screws, and a control handle; the height adjustment system is fixed at the four corners of the orbital platform, the four rotary motors are respectively horizontally installed and fixed on the side wall support frame, and are connected to the lead screws through bevel gears, and the lead screws are connected to the lifting frame through connectors. By using the orbital lifting system to lower the height of the orbital platform, the traction trajectory of the object traction system and the irradiation direction of the spotlight illumination system can be quickly adjusted to realize the switching of the test scenario.

[0007] The target traction system includes a traction system bracket, a traction system motor, a reducer, a conveyor belt driving wheel, a conveyor belt, conveyor belt wheels, a conveyor belt driven wheel, a conveyor belt wheel frame, a tensioning mechanism, and targets (adult targets, child targets, animal targets, etc.). The traction system bracket and the conveyor belt wheel frame are fixed under the track platform according to test requirements; the traction system motor is vertically fixed on the track platform through the traction system bracket, with the output shaft facing downward, and is connected to one end of the vertically fixed reducer through a coupling, and the other end of the reducer is connected to the conveyor belt driving wheel; the conveyor belt driving wheel, the conveyor belt wheels, and the conveyor belt driven wheel are connected by a conveyor belt; the tensioning mechanism is fixed on the conveyor belt driving wheel frame.

[0008] The track platform (12) adopts a modular splicing design, with standardized card slot interfaces provided at the platform edge, supporting the rapid replacement of arc tracks and ramp tracks (inclination angle 0° - 15°), and reserving an RS485 communication interface for expanding third-party sensors (including but not limited to LiDAR and millimeter-wave radar). The target can simulate the radar reflection characteristics of real targets and has the visual characteristics of real targets. The sensors of the target include an attitude sensor and a force sensor, which can simulate the normal walking postures of pedestrians or animals and test the force magnitude when a low-speed unmanned equipment impacts the target.

[0009] The spotlight lighting system includes 8 spotlight lamp groups and a spotlight lighting system control cabinet, which are evenly installed on the four sides of the track platform frame respectively. The spotlight lighting system includes spotlight lamp groups, a spotlight lighting system control cabinet, and an ambient light sensor; the spotlight lamp groups are evenly distributed on the four sides of the track platform, automatically adjust the brightness and color temperature through dimming technology, and support one-key switching between strong backlight, stroboscopic interference, and low illuminance at night modes.

[0010] The spotlight lighting system, including an integrated ambient light sensor and a mode switching controller, can monitor the light intensity of the test area in real time, automatically adjust the brightness and color temperature of the spotlight through PWM dimming technology, and support one-key switching between three extreme lighting modes: strong backlight, stroboscopic interference (frequency 120Hz), and low illuminance at night.

[0011] The data analysis module, by fusing the trigger signal of the grating sensor, the collision data of the force sensor, and the target attitude information captured by the high-definition camera, uses a spatio-temporal graph convolutional network (ST-GCN) to perform real-time obstacle avoidance risk assessment and generate a multi-dimensional performance report including obstacle avoidance response time, path deviation error, and peak collision force.

[0012] The data analysis module, by introducing the Spatio-Temporal Graph Convolutional Network (ST-GCN), conducts spatio-temporal joint modeling on multi-sensor time-series data (raster trigger signals, GPS trajectories, collision force waveforms), extracts high-order interaction patterns through node features (sensor types) and edge features (spatio-temporal correlations), and adds a gated attention mechanism to dynamically weight key sensor data, quantitatively evaluating the decision-making stability of the unmanned equipment under sudden disturbances and generating a spatio-temporal heat map to visualize the risk distribution.

[0013] The target object traction system is built-in with a dynamic path planning module. By receiving the GPS coordinate data of the tested robot in real time, it calculates the shortest avoidance path between the target object and the test equipment in combination with the hybrid path planning algorithm, and dynamically adjusts the speed and direction of the traction motor based on the PID controller to simulate the active avoidance or random escape behavior of the target object.

[0014] Furthermore, the dynamic path planning module integrates a hybrid path planning algorithm, including a local path optimization module based on Deep Reinforcement Learning (DRL) and a global path generation module based on the A* algorithm.

[0015] The Deep Reinforcement Learning (DRL) module constructs a state space, an action space, and a reward function, and uses the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm to online optimize the motion strategy of the target object, realizing multi-modal behavior simulation in a dynamic environment, including emergency braking, detouring, and group collaborative avoidance.

[0016] Among them, the state space includes the real-time GPS coordinates of the test equipment, the speed of the target object, the ambient light intensity, and the collision risk probability.

[0017] The action space includes the speed of the traction motor and the adjustment amount of the direction.

[0018] The reward function includes the obstacle avoidance success rate, path smoothness, and energy consumption efficiency.

[0019] After the A* algorithm generates the global basic path, based on Deep Reinforcement Learning (DRL), it dynamically adjusts the local path according to the real-time environmental data, and optimizes the network parameters through the priority experience replay mechanism, and cooperates with the PID controller to control the speed and direction of the traction motor, thereby reducing the response delay of the target object and reducing the path tracking error.

[0020] That is, the target object traction system adopts a hybrid path planning algorithm, which combines the global path planning ability of the A* algorithm with the dynamic decision-making ability of deep reinforcement learning (DRL). After the A* algorithm generates the initial path, the DRL module optimizes the motion strategy of the target object online according to real-time GPS data, ambient light intensity changes, and collision risk probability through the twin-delayed deep deterministic policy gradient (TD3) algorithm, simulating complex behaviors such as pedestrian emergency braking and animal random escape. At the same time, the data analysis module introduces a spatio-temporal graph convolutional network (ST-GCN), maps the time-series data of raster sensors, GPS, and force sensors into a spatio-temporal graph structure, and captures the dynamic correlation between sensors through a gated attention mechanism, significantly improving the collision warning accuracy and decision interpretability.

[0021] The present invention can reproduce a single indoor scene, and can also reproduce the interactive scene of composite indoor target objects by adjusting the irradiation direction of the spotlight illumination system, the traction trajectory and speed of the target object traction system, and the quantity and type of target objects.

[0022] On the other hand, the present invention also discloses a target object interaction test method for an indoor low-speed unmanned equipment, including the following steps:

[0023] S1. System initialization and track platform adjustment: Send instructions through the control cabinet of the height adjustment system of the track lifting system to start the rotating motor, drive the lead screw to lower the track platform to the installation height. The test engineer moves the lateral position of the track platform according to the test scene requirements, and adds a conveyor belt pulley on the left side of the track platform to deflect the conveyor belt 15° to the right, simulating the target object moving obliquely at 15°. Adjust the conveyor belt tension through the tensioning mechanism to ensure transmission stability; synchronously load the offline pre-trained deep reinforcement learning DRL policy network, load the global path map generated by the A* algorithm, and the ST-GCN model parameters.

[0024] S2. Target object deployment and path planning: Install the target object traction system under the track platform, and hoist a target pendulum leg dummy to the initial position coordinates (x0, y0) of the conveyor belt; then perform hybrid path planning: The A* algorithm calculates the global optimal path based on static map data and outputs the basic trajectory from the starting point to the end point.

[0025] The deep reinforcement learning DRL module receives real-time GPS data, ambient light intensity, and collision risk probability, and adjusts the target object behavior through the following steps to ensure that the target object and the test equipment reach the centerline collision point synchronously at the preset trigger point:

[0026] Definition of state space:

[0027] Definition of action space:

[0028] Among them, the speed difference between the left and right wheels (Δv = v 左轮 - v 右轮 ) controls the lateral displacement of the target object, and the emergency braking flag triggers the instantaneous stop of the conveyor belt;

[0029] Reward function design:

[0030] R t = ω1·ΙΙ 避障成功 + ω2·(1 - Δd 被测 / d 安全 ) + ω3·(1 - E 被测 / E 基准 ) + ω4·λ 难度

[0031] Among them, ω1, ω2, ω3, ω4 are weight values, and the obstacle avoidance success indication function Δd 被测 is the deviation error between the actual driving path of the measured equipment and the global planning path (such as generated by A*), d 安全 is the maximum allowable path deviation threshold. If it exceeds the limit, it is determined that the obstacle avoidance fails. E 被测 is the motor power consumption of the measured equipment, E 基准 is the reference energy consumption, and λ 难度 is a dynamically adjusted coefficient,

[0032] S3. Lighting environment configuration. The spotlight lighting system adjusts the brightness and color temperature of the spotlight group through the control cabinet, and dynamically matches the test requirements in combination with the feedback of the ambient light sensor; Through ST-GCN, real-time monitoring is carried out: the ambient light data is input into the spatio-temporal graph convolutional network, and the gated attention mechanism dynamically weights the weights of the lighting nodes to optimize the generation of the heat map;

[0033] S4. Test execution and dynamic control. The relevant data of the test scenario are observed and recorded through a high-definition camera, grating, GPS, and the force sensor of the target object; After the measured equipment activates the system from the preparation area, it travels uniformly at the test speed v0. The grating sensor triggers a signal, and the grating sensor sends a start signal to the control cabinet of the target object traction system;

[0034] The deep reinforcement learning DRL module outputs the target object speed command based on real-time GPS data: a t = π φ (s t ) + Ν t , where a t = v 目标物 , π φ is the policy network, and N t is the exploration noise; The PID controller adjusts the speed of the traction motor to drive the conveyor belt at v 目标物Run to make the target cross the center line at a preset angle; after the target reaches the position, if the tested equipment completely avoids collision with the target and provides an external warning, the test shall be considered passed.

[0035] Furthermore, it also includes

[0036] S5. After multiple rounds of testing, obtain the historical obstacle avoidance success rate, and based on the historical obstacle avoidance success rate, perform dynamic difficulty escalation and closed-loop optimization adjustment; first is the difficulty adaptive adjustment, including when the success rate ≥ 90%, increase the test difficulty; when the success rate ≥ 70% and < 90%, maintain the current parameters, but increase the random speed fluctuation; when the success rate < 70%: reduce the target speed, gradually reduce the environmental interference, and then screen high-value samples through Prioritized Experience Replay (PER) to update the parameters of the DRL policy network and optimize the action generation logic.

[0037] S6. Finally, based on the test results, perform Spatio-Temporal Graph Convolutional Network (ST-GCN) spatio-temporal analysis: including grating, GPS, and the force sensor of the target, and output a spatio-temporal heat map, and mark high-risk areas; output a test report, including performance indicators: obstacle avoidance success rate, average response time, energy consumption efficiency; behavior analysis: number of emergency brakes, path optimization suggestions, and difficulty suggestions.

[0038] As can be seen from the above technical solutions, the target interaction test system for the indoor low-speed driverless equipment of the present invention constructs a reproducible indoor test environment through the collaborative work of an orbital lifting system, a target traction system, a spotlight lighting system, and multi-modal sensors. During the test, place the driverless equipment in the preparation area, quickly adjust the height of the orbital platform through the orbital lifting system, and use the modular splicing structure to replace the arc or ramp track to simulate complex terrains; the target traction system is built-in with a dynamic path planning module, based on the real-time GPS data of the test equipment, dynamically adjusts the moving trajectory of the target through a hybrid path planning algorithm and PID control to simulate the active avoidance behavior of pedestrians or the random escape behavior of animals; the spotlight lighting system integrates an ambient light sensor, which can automatically adjust the brightness and color temperature, and can switch to extreme modes such as strong backlight and stroboscopic interference with one key. During the test process, the grating sensor, high-definition camera, force sensor, and GPS collect data synchronously, and the data analysis module fuses multi-source information, and through the Spatio-Temporal Graph Convolutional Network (ST-GCN), it evaluates the obstacle avoidance response time, path optimization rate, and peak collision force in real time, and generates a quantitative performance report. This system significantly improves the authenticity, flexibility, and data reliability of the test scenario, and can comprehensively verify the active obstacle avoidance ability and pedestrian protection ability of low-speed driverless equipment.

[0039] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0040] (1) Construction of comprehensive test environment: The system successfully constructs a reproducible indoor test environment by integrating multiple components such as a framework, a grating sensor, an orbital lifting system, an object traction system, a spotlight illumination system, a high-definition camera, and GPS. This comprehensive design enables the test environment to simulate various indoor lighting conditions and interaction scenarios of different objects, thus comprehensively evaluating the indoor autonomous driving performance of low-speed unmanned vehicles.

[0041] (2) High flexibility and adjustability: The height adjustment function of the orbital lifting system is achieved through components such as a rotary motor, a lead screw, and a control handle, enabling testers to easily adjust the height of the orbital platform according to test requirements. This flexibility not only facilitates the testing of unmanned vehicles of different sizes but also simulates height changes on different floors or terrains, enhancing the practicality of the test.

[0042] (3) Diversity of simulated dynamic behavior interactions of objects: The object traction system can tow various types of objects (such as adult objects, child objects, animal objects, etc.). By simulating the movement of the object within the test area through the movement of the conveyor belt, and through a hybrid path planning technology that combines deep reinforcement learning (DRL) and the A* algorithm, the object can autonomously generate non-fixed pattern behaviors (such as emergency braking, random escape, etc.), breaking through the limitation that the A* algorithm only supports preset paths. It reduces the object response delay and path tracking error, effectively simulating complex interaction scenarios such as pedestrians and animals.

[0043] (4) Controllability of the lighting environment: The spotlight illumination system provides the ability to simulate various lighting environments through spotlight lamp groups evenly installed on the four sides of the orbital platform frame. Testers can adjust the light brightness and color temperature according to test requirements to evaluate the performance of unmanned vehicles under different lighting conditions; the spotlight illumination system combines an ambient light sensor and a DRL dynamic dimming strategy, supporting one-key switching between strong backlight (>10,000 Lux), stroboscopic interference (120 Hz), and low night illumination (<5 Lux) modes, covering most indoor lighting scenarios. The system improves the test efficiency through modular orbital design (quick replacement of arc-shaped and ramp tracks) and algorithm collaborative control.

[0044] (5) Convenience of data recording and analysis: High-definition cameras are fixed at the four corners of the top support frame to record video data during the test. The GPS is installed on the test robot to track its movement trajectory. These data provide valuable basis for subsequent test analysis and performance evaluation, and help to discover and improve potential problems that may exist in the autonomous driving process of the unmanned equipment. The spatio-temporal graph convolutional network (ST-GCN) is introduced. Through spatio-temporal joint modeling (node feature: sensor type; edge feature: spatio-temporal correlation) and gated attention mechanism, key sensor data is dynamically weighted (for example, when the collision risk is high, the force sensor is preferentially processed). The spatio-temporal heat map and multi-dimensional performance report (obstacle avoidance response time, path deviation error, peak collision force) intuitively display the test results, support root cause analysis (such as decision vulnerability location), and improve the interpretability of the test compared with the traditional black box model (CNN). Description of the Drawings

[0045] Figure 1 is a schematic diagram of the overall structure of the target interaction test system for the indoor low-speed unmanned equipment of the present invention;

[0046] Figure 2 is a schematic diagram of the frame structure of the target interaction test system for the indoor low-speed unmanned equipment of the present invention;

[0047] Figure 3 is a schematic diagram of the track lifting system structure and target traction system of the target interaction test system for the indoor low-speed unmanned equipment of the present invention;;

[0048] Figure 4 is a flow chart of the hybrid path planning algorithm (DRL-A*);

[0049] Figure 5 is a schematic diagram of the spatio-temporal graph convolutional network (ST-GCN) architecture. Detailed Embodiments

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention.

[0051] As Figure 1 shown, the target interaction test system for the indoor low-speed unmanned equipment described in this embodiment includes a frame 1, a grating sensor 2, a track lifting system 3, a target traction system 4, a spotlight illumination system 5, a high-definition camera 6, a GPS 7, and a data analysis module.

[0052] As Figure 2As shown, the frame 1 is made of aluminum profiles or iron profiles, including a bottom support frame 8, a side wall support frame 9, and a top support frame 10. The bottom support frame 8 is fixed below the side wall support frame 9 by welding, and the top support frame 10 is fixedly connected above the side wall support frame 9 by welding, and the whole is covered with white gypsum board;

[0053] As Figure 3 shown, the track lifting system 3 includes a height adjustment system 11 and a track platform 12; the height adjustment system consists of four rotary motors 13, 14, 15, 16, an upper and lower limit device 17, four lead screws 18, 19, 20, 21, and a control cabinet 22; the four rotary motors 13, 14, 15, 16 are horizontally installed and fixed on the four columns of the side wall support frame 9 respectively, and are connected to the lead screws 18, 19, 20, 21 through bevel gears, and the lead screws are connected to the lifting frame through connectors. The track platform adopts a modular splicing design, with standardized slot interfaces at the edges, supporting the rapid replacement of arc tracks and ramp tracks, and reserved communication interfaces for expanding third-party sensors;

[0054] The target traction system 23 includes a traction system bracket 24, a traction system motor 25, a reducer 26, a conveyor belt driving wheel 27, a conveyor belt 28, conveyor belt wheels 29, a conveyor belt driven wheel 30, conveyor belt wheel frames 31, 32, 33, a tensioning mechanism 34, a target 35 (adult target, child target, animal target, etc.), and a control cabinet 36. The traction system bracket 24, conveyor belt wheel frames 31, 32, 33 are fixed below the track platform according to test requirements; the traction system motor 25 is vertically fixed on the track platform 12 through the traction system bracket 24, with the output shaft facing downwards, and is connected to one end of the vertically fixed reducer 26 through a coupling, and the other end of the reducer 26 is connected to the conveyor belt driving wheel 27; the conveyor belt driving wheel 27, conveyor belt wheels 29, conveyor belt driven wheel 30 are connected by a conveyor belt 28; the tensioning mechanism 34 is fixed on the conveyor belt driving wheel frame 31 for adaptively adjusting the conveyor belt tension;

[0055] The target traction system is built-in with a dynamic path planning module. Based on the real-time GPS data of the test robot, the avoidance path between the target and the test equipment is calculated by integrating the hybrid path planning algorithm through the dynamic path planning module, and the speed and direction of the traction motor are dynamically adjusted by using a PID controller to simulate the active avoidance or random escape behavior of the target;

[0056] The data analysis module generates a multi-dimensional performance report by fusing the trigger signals of grating sensors, the frame-level image data of high-definition cameras, the collision waveforms of force sensors of the target, and the motion trajectories of GPS, and evaluating the obstacle avoidance response time, path deviation error, and peak collision force in real time using a convolutional neural network;

[0057] The system constructs a reproducible test scenario that includes various indoor lighting environments, terrain conditions, and target interaction behaviors, and is used to verify the active obstacle avoidance ability and pedestrian protection performance of low-speed unmanned vehicles.

[0058] The spotlight lighting system 5 includes four groups of spotlight lamp groups 37, 38, 39, and 40 and a spotlight lighting system control cabinet 41, which are evenly installed on the four sides of the track platform frame. The spotlight lighting system includes a spotlight lamp group, a spotlight lighting system control cabinet, and an ambient light sensor; the spotlight lamp groups are evenly distributed on the four sides of the track platform, automatically adjust the brightness and color temperature through dimming technology, and support one-key switching of strong backlight, stroboscopic interference, and low illuminance at night modes.

[0059] The working principle and working process of the target interaction test system for the indoor low-speed unmanned vehicle of the present invention are as follows:

[0060] Since the working processes of various test scenarios are similar, taking the test scenario of an unobstructed adult pedestrian moving obliquely at 15° to the left indoors as an example, the specific steps are as follows:

[0061] 1. System initialization and track platform adjustment. Send instructions through the control cabinet 22 of the height adjustment system 11 of the track lifting system 3 to start the rotary motors 13-16, and drive the lead screws 18, 19, 20, and 21 to lower the track platform 12 to the installation height (default height 1.5 m, adjustment accuracy ±5 cm). The test engineer moves the lateral position of the track platform 12 according to the test scenario requirements, and adds a conveyor belt pulley on the left side of the track platform to deflect the conveyor belt 15° to the right to simulate the target moving obliquely at 15°. Adjust the conveyor belt tension through the tensioning mechanism 34 to ensure transmission stability; synchronously load the offline pre-trained DRL policy network:

[0062] Reward function R t = ω1·ΙΙ 避障成功 + ω2·(1 - Δd 被测 / d 安全 ) + ω3·(1 - E 被测 / E 基准 ) + ω4·λ 难度 );

[0063] And load the global path map generated by the A* algorithm and the ST-GCN model parameters;

[0064] 2. As Figure 4As shown in the figure, for the target deployment and path planning, the target traction system 23 is installed below the track platform 12, and the adult target swinging leg dummy 35 is hoisted to the initial position coordinates (x0, y0) of the conveyor belt 28. Then, for the hybrid path planning: The A* algorithm calculates the global optimal path based on the static map data (such as obstacle distribution, track length), and outputs the basic trajectory from the starting point to the ending point (the straight-line distance is 5 m). The DRL module receives the GPS data, ambient light intensity, and collision risk probability in real time, and adjusts the target behavior through the following steps to ensure that the target and the test equipment reach the centerline collision point synchronously at the preset trigger point:

[0065] State space definition:

[0066] Action space definition:

[0067] Among them, the speed difference between the left and right wheels (Δv = v 左轮 - v 右轮 ) controls the lateral displacement of the target, and the emergency braking flag triggers the instantaneous stop of the conveyor belt.

[0068] Reward function design:

[0069] R t = ω1·ΙΙ 避障成功 + ω2·(1 - Δd 被测 / d 安全 ) + ω3·(1 - E 被测 / E 基准 ) + ω4·λ 难度

[0070] Among them, ω1, ω2, ω3, and ω4 are weight values, Δd 被测 is the deviation error between the actual driving path of the equipment under test and the global planned path (such as generated by A*), d 安全 is the maximum allowable path deviation threshold, and if it exceeds the limit, it is determined that the obstacle avoidance fails. E 被测 is the motor power consumption of the equipment under test, E 基准 is the reference energy consumption, and λ 难度 is a dynamically adjusted coefficient,

[0071] 3. Lighting environment configuration: The spot - light illumination system 5 adjusts the brightness (range 0 - 10,000 Lux) and color temperature (2700K - 6500K) of the spot - light lamp groups 37 - 40 through the control cabinet 41, and combines the feedback of the ambient light sensor to dynamically match the test requirements (such as strong back - light mode: brightness > 8,000 Lux, color temperature 5,500K); through ST - GCN, it monitors in real - time: the ambient light data is input into the spatio - temporal graph convolutional network, and the gated attention mechanism dynamically weights the weights of the lighting nodes (w_light = 0.7) to optimize the generation of the heat map.

[0072] 4. Test execution and dynamic control: The relevant data of the test scene is observed and recorded through the high - definition camera 6, grating 2, GPS 7 and the force sensor of the target 35. After the tested equipment activates the system from the preparation area, it travels uniformly at the test speed v0. The grating sensor 2 triggers a signal, and the grating sensor 2 sends a start signal to the control cabinet 36 of the target traction system; the DRL module outputs the target speed command based on the real - time GPS data: a t =π φ (s t ) + Ν t , where a t =v 目标物 , π φ is the policy network, N t is the exploration noise; the PID controller adjusts the rotation speed of the traction motor 25 to drive the conveyor belt 28 to run at v 目标物 so that the target 35 crosses the center line along the preset angle. After the target 35 reaches the position, if the tested equipment completely avoids the collision with the target 35 and provides an external warning, the test should be regarded as passed.

[0073] 5. After multiple rounds of testing, the historical obstacle - avoidance success rate is obtained. Based on the historical obstacle - avoidance success rate, dynamic difficulty escalation and closed - loop optimization adjustment are carried out. First is the difficulty self - adaptive adjustment. For example, when the success rate ≥ 90%, the test difficulty is increased (the target speed is increased by 10% (to 6.05 km / h), the spot - light system activates strong back - light (> 10,000 Lux), etc.); when the success rate ≥ is 70% and < 90%, the current parameters are maintained, but random speed fluctuations (Δv ± 0.3 m / s Δv ± 0.3 m / s) are added; when the success rate < 70%, the target speed is reduced, and the environmental interference is gradually reduced. Then, through the prioritized experience replay (PER), high - value samples (such as decisions with a collision risk > 0.7) are screened, the parameters of the DRL policy network are updated, and the action generation logic is optimized.

[0074] 6. Such as Figure 5As shown, finally, based on the test results, ST-GCN spatio-temporal analysis is carried out: including grating, GPS, and the force sensor of the target 35, and a spatio-temporal heat map is output, and high-risk areas are marked; a test report is output, including performance indicators: obstacle avoidance success rate, average response time, energy consumption efficiency; behavior analysis: number of emergency brakes, path optimization suggestions (such as "reduce 5% sharp turns"); difficulty suggestion: recommend test parameters for the next stage according to historical data (such as "it is recommended to enable the multi-target mode").

[0075] Where the present invention is not described applies to the prior art.

[0076] In summary, for the target interaction test system of the indoor low-speed driverless equipment of the present invention, the system collaboratively constructs a reproducible indoor test environment through an orbital lifting system, a target traction system, a spotlight lighting system, and multi-modal sensors. During the test, the driverless equipment is placed in the preparation area, the height of the orbital platform is quickly adjusted through the orbital lifting system, and the arc or ramp track is replaced by using a modular splicing structure to simulate complex terrains; the target traction system is built-in with a dynamic path planning module, based on the real-time GPS data of the test equipment, and the moving trajectory of the target is dynamically adjusted through a hybrid path planning algorithm and PID control to simulate the active avoidance behavior of pedestrians or the random escape behavior of animals; the spotlight lighting system integrates an ambient light sensor, which can automatically adjust the brightness and color temperature, and can switch to extreme modes such as strong backlight and stroboscopic interference with one key. During the test process, the grating sensor, high-definition camera, force sensor, and GPS collect data synchronously, and the data analysis module fuses multi-source information, and the spatio-temporal graph convolutional network (ST-GCN) is used to evaluate the obstacle avoidance response time, path optimization rate, and peak collision force in real time, and generate a quantitative performance report. This system significantly improves the authenticity, flexibility, and data reliability of the test scenario, and can comprehensively verify the active obstacle avoidance ability and pedestrian protection ability of low-speed driverless equipment.

[0077] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by wire (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)).

[0078] It should be noted that in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise", or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device that includes a series of elements includes not only those elements but also other elements that are not expressly listed, or also includes elements that are inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article, or device that includes the element.

[0079] Each embodiment in this specification is described in a related manner. The same or similar parts among the embodiments can be referred to each other, and the differences between each embodiment and other embodiments are emphasized. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment.

[0080] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A target interaction test system for indoor low-speed driverless equipment, comprising a frame, a grating sensor, an orbit lifting system, a target traction system, a spotlight lighting system, a high-definition camera, GPS, and a data analysis module; characterized in that, The frame includes a bottom support frame, a side wall support frame, and a top support frame; the grating sensor is fixed on the side wall support frame; the orbit lifting system is fixed below the orbit platform; the target traction system and the spotlight lighting system are installed on the orbit lifting system; the high-definition camera is fixed at the four corners of the top support frame; the GPS is installed on the test robot; The target traction system is built-in with a dynamic path planning module. Based on the real-time GPS data of the test robot, the avoidance path between the target and the test equipment is calculated by integrating the hybrid path planning algorithm through the dynamic path planning module, and the speed and direction of the traction motor are dynamically adjusted by using a PID controller to simulate the active avoidance or random escape behavior of the target. The data analysis module fuses the trigger signal of the grating sensor, the frame-level image data of the high-definition camera, the collision waveform of the force sensor of the target, and the movement trajectory of the GPS, and uses a convolutional neural network to evaluate the obstacle avoidance response time, path deviation error, and peak collision force in real time, and generates a multi-dimensional performance report. The system constructs a reproducible test scenario including various indoor lighting environments, terrain conditions, and target interaction behaviors, which is used to verify the active obstacle avoidance ability and pedestrian protection performance of low-speed driverless equipment.

2. The target object interaction test system for the indoor low-speed driverless equipment according to claim 1, characterized in that: The orbit lifting system includes a height adjustment system and an orbit platform; The height adjustment system consists of four rotating motors, upper and lower limit devices, four lead screws, and a control handle, and is fixed at the four corners of the orbit platform; The rotating motors are horizontally installed on the side wall support frame, connected to the lead screws through bevel gears, and the lead screws are linked with the lifting frame through connectors; The orbit platform adopts a modular splicing design, with standardized slot interfaces at the edges, supporting the rapid replacement of arc-shaped orbits and ramp orbits, and reserved communication interfaces for expanding third-party sensors.

3. The target object interaction test system for the indoor low-speed driverless equipment according to claim 1, characterized in that: The target traction system includes a traction system support frame, a traction system motor, a reducer, a conveyor belt driving wheel, a conveyor belt, conveyor belt wheels, a conveyor belt driven wheel, a conveyor belt wheel frame, a tensioning mechanism, and a target; The traction system support frame and the conveyor belt wheel frame are fixed below the orbit platform; The traction system motor is vertically fixed on the orbit platform through the traction system support frame, and the output shaft is connected to the reducer downward, and the other end of the reducer drives the conveyor belt driving wheel; The tensioning mechanism adopts a spring-damping composite structure and is fixed on the conveyor belt driving wheel frame for adaptively adjusting the tension of the conveyor belt.

4. The target object interaction test system for the indoor low-speed driverless equipment according to claim 1, characterized in that: The spotlight lighting system includes a spotlight lamp group, a spotlight lighting system control cabinet, and an ambient light sensor; The spotlight lamp group is evenly distributed on the four sides of the orbit platform, automatically adjusts the brightness and color temperature through dimming technology, and supports one-key switching of strong backlight, stroboscopic interference, and low-illumination at night modes.

5. The target object interaction test system for indoor low-speed driverless equipment according to claim 3, wherein: The track platform adopts a modular splicing design. Standardized card slot interfaces are provided at the edge of the platform, supporting the rapid replacement of arc tracks, ramp tracks with an inclination angle of 0° - 15°, and an RS485 communication interface is reserved for expanding third-party sensors.

6. The target object interaction test system for the indoor low-speed driverless equipment according to claim 1, characterized in that: The dynamic path planning module integrates a hybrid path planning algorithm, including a local path optimization module based on deep reinforcement learning (DRL) and a global path generation module based on the A* algorithm. The DRL module constructs a state space, an action space, and a reward function, and uses the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm to optimize the motion strategy of the target object online, realizing multi-modal behavior simulation in a dynamic environment, including emergency braking, detouring, and group collaborative avoidance. Among them, the state space includes the real-time GPS coordinates of the test equipment, the speed of the target object, the ambient light intensity, and the probability of collision risk. The action space includes the rotational speed of the traction motor and the direction adjustment amount. The reward function includes the obstacle avoidance success rate, path smoothness, and energy consumption efficiency. After the A* algorithm generates the global basic path, the DRL dynamically adjusts the local path according to the real-time environmental data, and optimizes the network parameters through the priority experience replay mechanism, and cooperates with the PID controller to control the rotational speed and direction of the traction motor, thereby reducing the response delay of the target object and reducing the path tracking error.

7. The target object interaction test system for the indoor low-speed driverless equipment according to claim 4, characterized in that: The spotlight lighting system includes an integrated ambient light sensor and a mode switching controller, which can monitor the light intensity of the test area in real time, automatically adjust the brightness and color temperature of the spotlight through PWM dimming technology, and supports one-key switching of three extreme lighting modes: strong backlight, stroboscopic interference, and low illuminance at night.

8. The target object interaction test system for the indoor low-speed driverless equipment according to claim 1, characterized in that: The data analysis module introduces a spatio-temporal graph convolutional network (ST-GCN) to perform spatio-temporal joint modeling on multi-sensor time series data, namely grating trigger signals, GPS trajectories, and collision force waveforms. High-order interaction patterns are extracted through node features, i.e., sensor types, and edge features, i.e., spatio-temporal correlation, and a gated attention mechanism is added to dynamically weight key sensor data, quantitatively evaluating the decision-making stability of the unmanned equipment under sudden interference, and generating a spatio-temporal heat map to visualize the risk distribution.

9. A method for target interaction testing of an indoor low-speed driverless device, characterized in that: It includes the following steps. S1. System initialization and track platform adjustment: Send instructions to the control cabinet of the height adjustment system of the track lifting system to start the rotary motor, drive the lead screw to lower the track platform to the installation height. The test engineer moves the lateral position of the track platform according to the test scenario requirements, and adds a conveyor pulley on the left side of the track platform to deflect the conveyor belt 15° to the right, simulating the target object moving obliquely at 15°. Adjust the conveyor belt tension through the tensioning mechanism to ensure transmission stability; synchronously load the offline pre-trained DRL policy network, the global path map generated by the A* algorithm, and the ST-GCN model parameters. S2. Target object deployment and path planning: Install the target object traction system under the track platform, and hoist the target pendulum leg dummy to the initial position coordinates (x0, y0) of the conveyor belt; then perform hybrid path planning: The A* algorithm calculates the global optimal path based on the static map data and outputs the basic trajectory from the starting point to the end point. The deep reinforcement learning DRL module receives GPS data, ambient light intensity, and collision risk probability in real time, and adjusts the behavior of the target object through the following steps to ensure that the target object and the test equipment reach the centerline collision point synchronously at the preset trigger point: State space definition: s t = [x GPS , y GPS , v 目标物 , light intensity, collision risk probability]; Action space definition: a t = [v 左轮 , v 右轮 , emergency braking dynamic position] Among them, the speed difference between the left and right wheels (Δv = v 左轮 - v 右轮 ) controls the lateral displacement of the target object, and the emergency braking flag triggers the instantaneous stop of the conveyor belt; Reward function design: R t = ω1·ΙΙ 避障成功 + ω2·(1 - Δd 被测 / d 安全 ) + ω3·(1 - E 被测 / E 基准 ) + ω4·λ 难度 Among them, ω1, ω2, ω3, and ω4 are weight values, and the obstacle avoidance success indication function Δd 被测 is the deviation error between the actual driving path of the equipment under test and the global planned path, and d 安全 is the maximum allowable path deviation threshold. If it exceeds the limit, it is determined that the obstacle avoidance fails, and E 被测 is the motor power consumption of the equipment under test, and E 基准 is the reference energy consumption, and λ 难度 is a dynamically adjusted coefficient, S3. Lighting environment configuration: The spotlight lighting system adjusts the brightness and color temperature of the spotlight group through the control cabinet, and dynamically matches the test requirements in combination with the feedback of the ambient light sensor; Real-time monitoring through ST-GCN: Ambient light data is input into the spatio-temporal graph convolutional network, and the gated attention mechanism dynamically weights the weights of the lighting nodes to optimize the generation of the heat map; S4. Test execution and dynamic control: Observe and record relevant data of the test scenario through a high-definition camera, grating, GPS, and force sensor of the target object; After the equipment under test activates the system from the preparation area, it travels uniformly at the test speed v0, triggers a signal through the grating sensor, and the grating sensor sends a start signal to the control cabinet of the target object traction system; The deep reinforcement learning DRL module outputs the target object speed command based on real-time GPS data: a t = π φ (s t ) + Ν t , where a t = v 目标物 , π φ is the policy network, N t is the exploration noise; the PID controller adjusts the traction motor speed to drive the conveyor belt to run at v 目标物 so that the target object crosses the center line at a preset angle; after the target object reaches the position, if the tested equipment completely avoids collisions with the target object and provides an external warning, the test should be considered passed.

10. A method for target interaction testing of an indoor low-speed driverless device according to claim 9, characterized in that: It also includes S5. After multiple rounds of testing, obtain the historical obstacle avoidance success rate, and perform dynamic difficulty upgrade and closed-loop optimization adjustment based on the historical obstacle avoidance success rate; First is the difficulty adaptive adjustment, including when the success rate ≥ 90%, increase the test difficulty; when the success rate ≥ 70% and < 90%, maintain the current parameters, but increase the random speed fluctuation; when the success rate < 70%: reduce the speed of the target object, gradually reduce the environmental interference, and then screen high-value samples through prioritized experience replay PER, update the parameters of the DRL policy network, and optimize the action generation logic; S6. Finally, based on the test results, perform ST-GCN spatio-temporal analysis: including grating, GPS, and force sensor of the target object, and output a spatio-temporal heat map, and mark high-risk areas; Output a test report, including performance indicators: obstacle avoidance success rate, average response time, energy consumption efficiency; Behavior analysis: number of emergency brakes, path optimization suggestions, and difficulty suggestions.

Images