Robot indoor-outdoor switching avoidance method and system based on 79ghz millimeter wave radar

Abstract

This invention discloses a robot indoor / outdoor switching avoidance method and system based on 79GHz millimeter-wave radar. The method includes acquiring echo signals using a 79GHz millimeter-wave radar installed on the robot, extracting frequency domain statistical features and spatial distribution features from the echo signals as radar environment features; determining the robot's current environment category using a classifier based on the radar environment features, and triggering a transition control mode when the environment is determined to be a transitional environment; detecting transparent obstacles and their opening / closing states based on radar echoes during the robot's traversal of the transitional environment, and controlling the robot's movement behavior based on the detection results; and when the environment is determined to be outdoor and the global navigation satellite system signal is available, using the current pose estimated based on radar echoes as a reference anchor point and fusing it with the positioning results of the global navigation satellite system to smoothly switch to outdoor navigation mode. This significantly improves the robot's safety, smoothness, and robustness in indoor / outdoor switching scenarios.

Description

Technical Field

[0001] This invention relates to the field of robot autonomous navigation and environmental perception technology, specifically to a robot indoor / outdoor switching avoidance method and system based on 79GHz millimeter-wave radar. Background Technology

[0002] In recent years, service robots, lawnmower robots, and security robots have gradually evolved from single indoor or outdoor scenarios to cross-regional operations. For example, home cleaning robots need to pass through glass doors from the living room to enter the yard, and delivery robots need to drive from indoor warehouses to outdoor parks.

[0003] However, during the transition between indoor and outdoor environments, when the robot must pass through transitional areas that are neither completely enclosed indoors nor open outdoors (such as roller shutters, glass partitions, and the end of corridors), the robot is prone to collisions, positioning freezes, or "flying" accidents. This is due to the misjudgment of "transparent / semi-transparent obstacles" and "open and infinite space" by traditional navigation methods, and the simultaneous unreliability of various signals and sensors on which the robot relies for positioning and perception. The Global Navigation Satellite System (GNSS) signal fluctuates violently or is even lost due to doorway obstruction. Visual sensors are overexposed or lose features due to sudden changes in indoor and outdoor lighting. LiDAR cannot reliably detect transparent glass and has difficulty distinguishing between closed glass doors and open doorways. Wheeled odometers slip and drift due to wet or loose ground. As a result, the robot cannot accurately determine whether there are closed transparent obstacles in front of it in the transitional area, cannot maintain continuous pose estimation, and cannot smoothly complete the switch between indoor and outdoor navigation modes. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, the present invention provides a robot indoor-outdoor switching avoidance method and system based on 79GHz millimeter-wave radar to solve the problems in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A robot indoor / outdoor switching avoidance method based on 79GHz millimeter-wave radar includes the following steps: Echo signals are acquired by a 79GHz millimeter-wave radar installed on a robot, and frequency domain statistical features and spatial distribution features are extracted from the echo signals as radar environment features. Based on the radar environmental characteristics, a classifier determines the current environmental category of the robot, which includes at least indoor environment, transitional environment and outdoor environment. When a transitional environment is identified, a transitional control mode is triggered. In the transitional control mode, the contribution of the wheel odometry to the positioning is reduced or frozen, and radar echo-based relative pose estimation is enabled to maintain the robot's continuous positioning. During the robot's journey through the transitional environment, transparent obstacles and their open / closed states are detected based on radar echoes, and the robot's movement behavior is controlled according to the detection results. When the environment is determined to be outdoor and the GPS signal is available, the current pose estimated based on radar echo is used as the reference anchor point and fused with the GPS positioning result to smoothly switch to outdoor navigation mode.

[0006] In one embodiment, under the transition control mode, one or more of the following operations are performed: Freeze the integral accumulation of the wheel odometer to prevent drift caused by slippage or sliding; Disable the visual perception processing thread to avoid interference caused by sudden changes in lighting conditions; Ignore global navigation satellite system signals to avoid signal jumps caused by porch obstruction; Pose estimation is performed using a tightly coupled filter between the radar and the inertial measurement unit.

[0007] In one embodiment, detecting a transparent obstacle and its open / closed state includes: Search for point clusters with high radar cross section in the radar point cloud, extract the edge diffraction features of the point clusters, and connect adjacent high RCS point clusters to form an electromagnetic boundary topology map. When the distance between the robot and the electromagnetic boundary topology is less than a preset threshold, it is determined that there is a potential transparent obstacle boundary. Within the azimuth interval corresponding to the electromagnetic boundary topology map, search for the peak value of the radar echo along the range dimension; If there are two peaks, the distance difference between them corresponds to the thickness and dielectric constant of the transparent medium, and the power difference between the two peaks is within a preset range, then the transparent obstacle is determined to be in a closed state. Otherwise, it is determined to be in an open state or without obstacles.

[0008] In one embodiment, the transition control mode further includes: Extract the velocity vector field of the ground point cloud, and calculate the divergence and curl of the velocity vector field; The stability of the robot's motion is determined based on the divergence, and the degree of slippage of the wheeled odometer is detected based on the curl. When the curl exceeds a preset threshold, the weight of the wheel odometer in state estimation is reduced, and the direct displacement estimation method based on ground point cloud phase correlation is enabled.

[0009] In one embodiment, the transition control mode further includes: Extract obstacle boundary points on the left and right sides from the radar point cloud, and calculate the robot's heading deviation and lateral offset. Steering commands are generated using pure tracking or model predictive control methods, enabling the robot to cross the centerline of the boundary point.

[0010] In one embodiment, the transition control mode further includes: The safe crossing speed is calculated based on the difference between the detected width of the transition area and the robot's own width. The robot's motion speed is shaped by using an S-shaped velocity curve with limited jerk, thus achieving a smooth process of deceleration, crossing, and then accelerating.

[0011] In one embodiment, it further includes: The location information, radar environment feature template, and geometric dimensions of each successfully traversed transition area are saved as an electromagnetic fingerprint node. When the robot approaches the saved electromagnetic fingerprint node again, it identifies the known node by matching the radar point cloud with the node position information, directly loads the parameters of the node, and uses the absolute coordinates of the node to perform closed-loop correction of the current pose.

[0012] In one embodiment, it further includes: The health status of the radar is monitored in real time, and the health status is evaluated based on at least one or more of the following: radar point cloud quantity, point cloud static point ratio, or inter-frame matching residual. When the health status is at level one, limit the robot's maximum speed and activate auxiliary sensors for slow path exploration. When the health status is at level two, the robot is controlled to stop in place, issue an alarm, and upload the raw radar data. When the health status is at level three and no human intervention is received within a preset time after stopping, the robot is controlled to backtrack along the original path to the nearest safe area.

[0013] In one embodiment, it further includes: When the robot is operating outdoors, it continuously monitors the radar environmental characteristics; When the radar environmental characteristics indicate a change from an outdoor environment to a transitional environment, the robot's maximum speed is limited, and the robot is controlled to center and align itself to enter based on the width of the transitional area detected in the radar point cloud. At the same time, the positioning contribution of the Global Navigation Satellite System is turned off, and the indoor navigation mode is reactivated.

[0014] A robot indoor / outdoor switching avoidance system based on 79GHz millimeter-wave radar includes: At least one 79 GHz millimeter-wave radar, mounted on a robot, is used to transmit and receive 79 GHz millimeter-wave signals; One or more processors; Memory, which stores computer program instructions; When the processor executes the program instructions, it implements the robot indoor / outdoor switching avoidance method based on 79GHz millimeter-wave radar as described above.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention leverages the immunity to light and weather conditions of 79GHz millimeter-wave radar, fundamentally preventing visual failure during sudden changes in lighting and the inability of lasers to detect glass, thus avoiding perception loss when switching between indoor and outdoor environments. Environmental classification is achieved through the statistical characteristics of radar echoes (frequency domain entropy and spatial entropy), independent of external signals. This avoids the inability to accurately determine environmental boundaries when GPS is unreliable or visual features are missing, due to reliance on GPS signal strength or visual features. Furthermore, the odometry weight is proactively reduced in transition zones, replacing it with radar point cloud matching for positioning, ensuring positioning continuity and preventing significant drift caused by wheeled odometry slippage in transition zones. Safe passage control is achieved by using radar echoes to detect transparent obstacles and their opening / closing states, preventing situations where the robot cannot distinguish between closed glass doors and open doorways, thus avoiding collisions. Finally, radar-estimated pose is used as an anchor point for smooth fusion, avoiding direct fusion of GPS and odometry, which can lead to pose jumps. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the workflow of a robot indoor / outdoor switching avoidance method based on 79GHz millimeter-wave radar. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] like Figure 1 As shown, the present invention provides a robot indoor / outdoor switching avoidance method based on 79GHz millimeter-wave radar, comprising the following steps: S100. A 79GHz millimeter-wave radar mounted on a robot transmits and receives echo signals, extracting radar environment features from the echo signals, including at least frequency domain statistical features and spatial distribution features. The frequency domain statistical features reflect the clutter of the Doppler spectrum energy distribution of the radar echo, while the spatial distribution features reflect the density and uniformity of the radar point cloud in space. Specifically: After the robot starts, the forward-mounted 79GHz millimeter-wave radar (in this embodiment, it operates in the 76GHz~81GHz frequency band, with a horizontal field of view of ±60° and a vertical field of view of ±15°, outputting 4D point clouds and range-Doppler maps) begins to continuously acquire echo signals. The processor extracts two features from the echo signals in real time: Frequency domain statistical characteristics: For each frame of radar output range-Doppler image, the Shannon entropy of the echo energy distribution along the Doppler dimension is calculated, called the Doppler spectral entropy. Specifically: First, the energy of each Doppler cell is counted, and the energy of all Doppler cells is summed to obtain the total energy; the energy of each cell is divided by the total energy to obtain the normalized probability of that cell; then, the logarithm of the normalized probability of each cell is taken to base 2, and multiplied by the probability itself; finally, the negative products of all cells are summed. This entropy value is used to characterize the clutter of the radar echo Doppler spectrum. In an enclosed indoor environment, walls, furniture, and other structures generate a large number of multipath reflections, and the Doppler spectral energy is dispersed across multiple Doppler cells, resulting in a higher entropy value; in an open outdoor environment, the echo mainly comes from the stationary ground and sparse obstacles, and the Doppler spectral energy is concentrated near zero frequency, resulting in a lower entropy value.

[0019] To avoid dynamic targets interfering with entropy calculation, dynamic interference is filtered out from the range-Doppler image before extraction: For the Doppler spectrum corresponding to each range gate, five frames of data are continuously collected to calculate the fluctuation variance of the second moment of the Doppler spectrum. If the variance exceeds the preset threshold, it is determined that there are dynamic targets such as pedestrians and vehicles in the area corresponding to the range gate, and the data of the range gate is removed from the entropy calculation to ensure that the environmental features only reflect static environmental attributes.

[0020] Spatial Distribution Characteristics: The 4D point cloud output from each radar frame is projected onto the horizontal plane of the robot's coordinate system. The area 5m forward and 3m to the left and right of the robot is divided into 0.1m × 0.1m grids. The presence of at least one point cloud within each grid is counted (1 for presence, 0 for absence). The occupancy counts of all grids are summed to obtain the total number of occupied grids. The occupancy probability of each grid is obtained by dividing its occupancy count by the total number of occupied grids. Then, the base-2 logarithm of each grid's occupancy probability is taken and multiplied by the probability itself. Finally, the negative products of all grids are summed to obtain the spatial distribution entropy. In indoor environments, the point cloud is dense and relatively uniformly distributed, with many occupied grids, a flat probability distribution, and a high entropy value (typically approximately 0.85 bits). In outdoor environments, the point cloud is sparse, with fewer occupied grids and a low entropy value (typically approximately 0.25 bits).

[0021] In one embodiment, before calculating the Doppler spectrum entropy, dynamic interference filtering is performed on the range-Doppler map of the radar echo, including: calculating the second moment of the Doppler spectrum of each range-Doppler unit as a function of time; if the variance of the second moment fluctuation exceeds a preset threshold, the unit is identified as dynamic target interference and removed from the entropy calculation.

[0022] It should be noted that dynamic targets (such as pedestrians) generate non-zero Doppler frequencies, and their speed changes over time, causing the Doppler spectrum to fluctuate rapidly in the time dimension. In contrast, the Doppler spectrum of static environments (walls, ground) remains relatively stable near zero frequency. Therefore, before calculating the Doppler spectrum entropy, dynamic interference is filtered out from the radar echo range-Doppler map. Specifically: for each range gate, the average frequency and second moment (i.e., the weighted average of the frequency variance) of its Doppler spectrum are calculated. A sequence of five consecutive frames of the second moment is recorded, and the sample variance of this sequence is calculated. If the variance exceeds a preset threshold (typically 0.1, dimensionless after normalization), it is determined that a dynamic target (such as a pedestrian or vehicle) exists in the azimuth corresponding to that range gate. When calculating the global Doppler spectrum entropy, the power of all Doppler units of that range gate is set to zero (removed), and the entropy value is only calculated for the remaining range gates. This avoids environmental classification errors caused by dynamic interference.

[0023] In this embodiment, dynamic targets such as pedestrians and vehicles generate varying Doppler frequencies, causing drastic fluctuations in Doppler spectrum entropy, leading to misclassification of the environment and an inability to distinguish between dynamic interference and static environment. This invention solves the problem of incorrect classification of indoor and outdoor environments caused by dynamic interference by calculating the second moment of the Doppler spectrum of each distance-Doppler unit over time. If the variance of the second moment exceeds a threshold, the unit is identified as dynamic target interference and removed from the entropy calculation. This improves the robustness of the system in real and complex scenarios.

[0024] It should be noted that the millimeter-wave radar operates in the 76GHz~81GHz frequency band; the frequency domain statistical feature is Doppler spectral entropy, and the spatial distribution feature is point cloud spatial distribution entropy; the classifier is a fuzzy logic classifier, whose output is the confidence score of each environment category, and a hysteresis comparison rule is used to prevent frequent state jumps. Through dual-modal entropy and fuzzy logic, the accuracy and stability of environment classification are significantly improved, especially near the boundary of the transition zone, where frequent state switching is avoided.

[0025] The 76–81 GHz band (i.e., the 79 GHz radar) possesses centimeter-level range resolution and high sensitivity to micro-Doppler, enabling it to distinguish glass bimodalities and minute ground textures. Choosing this band forms the basis for all subsequent feature extraction. Doppler spectral entropy and spatial distribution entropy characterize environmental features in the frequency and spatial domains, respectively, and are complementary: indoor multipath propagation leads to frequency domain clutter, while dense point clouds result in spatial uniformity; the opposite is true outdoors. A single entropy value is susceptible to dynamic disturbances; the dual-modal approach improves robustness. A fuzzy logic classifier combined with hysteresis comparisons avoids boundary jitter caused by hard thresholds, resulting in smooth state transitions.

[0026] S200. Based on the radar environmental characteristics, the robot's current environmental category is determined by a classifier. The environmental category includes at least indoor environment, transitional environment, and outdoor environment. In this embodiment, the extracted Doppler spectral entropy and spatial distribution entropy are combined to form a dual-modal feature vector, which is input into a pre-trained fuzzy logic classifier. The membership function of the classifier adopts a trapezoidal function, the rule base is 3×3, and the confidence scores of three categories are output: indoor confidence score, transition confidence score, and outdoor confidence score. The confidence score ranges from 0 to 1.

[0027] To avoid frequent state fluctuations at environmental boundaries, the classifier employs a hysteresis comparison rule: the robot is only allowed to switch environmental states when the confidence score of a certain environmental category exceeds 0.8 for three consecutive frames. Specifically, the double entropy value of indoor environments is generally high, while that of outdoor environments is generally low. The double entropy value of transitional environments is in the transition range between indoors and outdoors, corresponding to semi-open indoor-outdoor switching areas such as porches, doorways, and roller shutters.

[0028] For example, when the robot is in the center of the living room, its Doppler spectral entropy is approximately 4.5 bits, and its spatial distribution entropy is approximately 0.85 bits. The classifier outputs an indoor confidence score of 0.92, classifying it as an indoor environment. As the robot moves towards the doorway, its Doppler spectral entropy decreases to 2.5 bits, and its spatial distribution entropy decreases to 0.45 bits. The classifier outputs a transition confidence score of 0.86, classifying it as a transitional environment. After the robot has completely passed through the doorway, its Doppler spectral entropy decreases to 1.2 bits, and its spatial distribution entropy decreases to 0.25 bits. The classifier outputs an outdoor confidence score of 0.91, classifying it as an outdoor environment.

[0029] S300. When the environment is determined to be transitional, the transition control mode is triggered. In the transition control mode, the contribution of the wheel odometry to the positioning is reduced or frozen to at least part, and the point cloud matching based on radar echo or micro Doppler features is enabled for relative pose estimation to maintain the robot's continuous positioning. In this embodiment, when a transitional environment is identified, a transition control mode is triggered (e.g., when the classifier outputs a transitional environment confidence score exceeding 0.8 for three consecutive frames, the robot immediately triggers the transition control mode). In this mode, the system performs one or more of the following operations: Freeze the integration accumulation of the wheel odometer (but do not discard the original data, which will be used for subsequent slip detection), and no longer use the wheel odometer data for pose update to prevent drift caused by slip or sliding; Close the visual perception processing thread to avoid overexposure interference caused by strong outdoor light, such as visual feature loss and mismatch caused by sudden changes in lighting when switching between indoor and outdoor environments. Ignore Global Navigation Satellite System signals (even if signals are present, do not trust them for the time being) to avoid signal jumps caused by porch obstruction; avoid GNSS multipath interference and signal jumps caused by porch obstruction by temporarily ignoring the positioning data output by the GNSS module; High-frequency pose estimation is performed by using a tightly coupled filter between a 79GHz millimeter-wave radar and an IMU. The relative pose transformation obtained by inter-frame matching of radar point clouds is used as the observation value, and the IMU pre-integration result is used as the prediction value. The continuous pose at the centimeter level is output to ensure that the robot's positioning in the transition area is not affected by slippage, lighting, or GNSS signal fluctuations.

[0030] In one embodiment, under transition control mode, it further includes: S310. Extract the velocity vector field of the ground point cloud and calculate the divergence and curl of the velocity vector field; S320: Determine the stability of the robot's motion based on divergence, and detect the degree of slippage of the wheeled odometer based on curl; S330. When the curl exceeds the preset threshold, reduce the weight of the wheel odometer in the state estimation and enable the direct displacement estimation method based on the phase correlation of the ground point cloud.

[0031] In this embodiment, to address the problem of significant errors in wheel odometry caused by slippage when the robot is driving on wet or loose surfaces (such as lawns after rain, sand, or snow), this invention extracts the velocity vector field of the ground point cloud and calculates divergence and curl. Motion stability is determined based on divergence, and the degree of slippage is detected based on curl. When the curl exceeds a threshold, the weight of the wheel odometry is reduced, and phase-correlated displacement estimation of the ground point cloud is enabled. This achieves real-time detection and compensation for slippage, ensuring that the positioning in the transition zone is unaffected by ground conditions and guaranteeing the pose accuracy after fusion.

[0032] Specifically: Based on the radar's installation pitch angle and altitude, point clouds with pitch angles between -5° and +5° are selected as ground points. For each ground point, a two-dimensional velocity vector (forward and lateral components) is calculated using its radial velocity and azimuth angle. The velocity vectors of all ground points are represented in the robot coordinate system, forming a velocity vector field. The ground point cloud is divided into 0.2m × 0.2m grids along the forward and lateral directions, and the average velocity vector is calculated within each grid. Then, the divergence and curl of the velocity vector field are calculated using the finite difference method (the difference between adjacent grids divided by the grid spacing). Divergence represents the intensity of the velocity field at a point, acting as a "source" or "sink." Under ideal pure rolling conditions, the divergence is close to 0, while the absolute value of the divergence increases during longitudinal slippage. Curl represents the rotational intensity of the velocity field; under normal rolling conditions, the curl is close to 0, while the absolute value of the curl increases during lateral slippage or rotational slippage. Both the divergence and curl thresholds are set to 0.05 (dimensionless). If the absolute value of divergence or curl exceeds the corresponding threshold, slippage is determined to have occurred. In this case, during error state Kalman filtering, the observation noise covariance of the wheel odometry is multiplied by 10 (i.e., its weight is reduced), and a backup displacement estimation method is activated: two consecutive frames of ground point cloud are projected onto a two-dimensional plane, the two-dimensional cross-correlation coefficient is calculated, the offset corresponding to the peak position is found, and then the rotation angle is fitted by least squares. This displacement estimation is not affected by wheel slippage and directly reflects the robot's true motion.

[0033] It should be noted that in the transition control mode, ground point clouds with pitch angles in the range of -5° to +5° are continuously extracted from the radar point cloud. Based on the radial velocity and azimuth of each ground point, a two-dimensional velocity vector field of the ground point is reconstructed. The divergence and curl of the velocity vector field are calculated, where the divergence is used to determine the stability of the robot's longitudinal motion and the curl is used to detect the degree of slippage of the wheel odometry. When the divergence or curl exceeds a preset threshold, it is determined that the robot has wheel slippage, further reducing the weight of the wheel odometry in state estimation. At the same time, a direct displacement estimation method based on the phase correlation of the ground point cloud is enabled, which directly calculates the robot's true displacement by matching the ground point clouds of two consecutive frames, compensating for the positioning drift caused by slippage.

[0034] In one embodiment, under transition control mode, it further includes: S340. Extract the obstacle boundary points on the left and right sides from the radar point cloud, and calculate the robot's heading deviation and lateral offset. The S350 uses pure tracking or model predictive control methods to generate steering commands, enabling the robot to cross the centerline of the boundary point.

[0035] In this embodiment, if the robot's heading is incorrect or it deviates laterally when passing through a doorway, it may collide with the door frame. Traditional methods require additional lateral sensors or visual markers. This invention extracts the left and right obstacle boundary points from the radar point cloud, calculates the heading deviation and lateral deviation, and uses pure tracking or MPC to generate steering commands, enabling the robot to pass through the centerline. The control quantity is directly calculated using the corner points of the door frame already detected by the radar, requiring no additional hardware and avoiding collisions.

[0036] Specifically: Obstacle boundary points (i.e., doorway corner points) on both sides are extracted from the radar point cloud, and the robot's heading deviation and lateral offset are calculated. The heading deviation is the azimuth angle of the doorway centerline minus the robot's current heading angle; the lateral offset is the vertical distance from the robot's center to the doorway's central axis, with a positive value indicating a rightward deviation and a negative value indicating a leftward deviation. A pure tracking controller is used: the look-ahead distance is half the width of the doorway, but does not exceed 0.5 meters. The controller outputs a steering angular velocity command, the magnitude of which is directly proportional to the linear combination of the heading deviation and lateral offset, and inversely proportional to the look-ahead distance and the current linear velocity. This control law gradually aligns the robot with the doorway centerline, ensuring that the lateral error during passage is controlled within 3 centimeters.

[0037] It should be noted that by extracting the obstacle boundary points (i.e., door frame corner points) on both sides of the transition area from the radar point cloud, the direction of the doorway's centerline is determined. The deviation between the robot's current heading and the doorway's centerline, as well as the lateral offset of the robot's center relative to the doorway's centerline, are calculated. Using a pure tracking algorithm commonly used in the field of mobile robots, the above-mentioned heading deviation and lateral offset are used as inputs to generate a turning angular velocity command in real time. This controls the robot to adjust its heading, always traversing along the doorway's centerline, ensuring that the distance between the robot and the left and right door frames is uniform and avoiding collisions.

[0038] In one embodiment, under transition control mode, it further includes: S360. Calculate the safe crossing speed based on the difference between the detected width of the transition area and the robot's own width. S370 uses an S-shaped velocity curve with limited acceleration to shape the robot's motion speed, achieving a smooth process of deceleration, crossing, and then accelerating.

[0039] In this embodiment, the robot decelerates directly from the indoor cruising speed to the crossing speed and then accelerates. If the speed changes abruptly, it will cause shock and motor current spikes. The present invention calculates the safe crossing speed based on the difference between the width of the transition area and the width of the robot, and uses an S-shaped speed curve with limited jerk to shape the speed. The S-shaped curve ensures continuous acceleration, limited jerk, smooth motion, no jerking during the crossing process, and stable motor load.

[0040] Specifically: Based on the difference between the detected doorway width and the robot's own width, a speed coefficient (ranging from 0.2 to 0.5, with a typical value of 0.2) is multiplied to calculate the safe crossing speed, with an upper limit of 0.5 m / s. Then, an S-shaped speed curve with limited jerk is used to shape the robot's motion speed. Specific constraints are: maximum acceleration not exceeding 0.5 m / s², and maximum jerk not exceeding 0.8 m / s³. When the robot is 1 meter away from the doorway, a deceleration phase is triggered. The time and displacement required for deceleration are calculated based on the current speed and the safe crossing speed, generating a curve showing a linear change in acceleration over time (with constant jerk). After the robot passes through the doorway (with both doorway corners located behind the robot), an acceleration phase is triggered, accelerating from the safe crossing speed to the outdoor cruising speed. The speed planner outputs the desired speed to the chassis controller in real time, achieving a smooth deceleration and acceleration process, avoiding mechanical shocks and motor current spikes.

[0041] It should be noted that the safe crossing speed is calculated based on the difference between the detected doorway width and the robot's own width: the larger the difference between the doorway width and the robot's width, the higher the allowed crossing speed, and the smaller the difference, the lower the crossing speed. The upper limit of the safe crossing speed is set to 0.5m / s. The robot's movement speed is shaped using an S-shaped speed curve with limited acceleration. When the robot is 1m away from the doorway, it begins to decelerate to the safe crossing speed. After completing the crossing of the doorway, it smoothly accelerates to the outdoor cruising speed to avoid mechanical shock, motor current spikes, and wheel slippage caused by sudden speed changes.

[0042] S400. During the robot's traversal of the transitional environment, the presence and opening / closing status of transparent obstacles are detected based on radar echoes, and the robot's movement behavior is controlled according to the detection results, including: when a transparent obstacle is detected to be in a closed state, the robot is prohibited from moving forward; when a transparent obstacle is detected to be in an open state or there is no transparent obstacle, the robot is allowed to pass through; wherein, detecting transparent obstacles and their opening / closing status includes: S410. Search for point clusters with high radar cross section in the radar point cloud, extract the edge diffraction features of the point clusters, and connect adjacent high RCS point clusters to form an electromagnetic boundary topology map. S420. When the distance between the robot and the electromagnetic boundary topology is less than a preset threshold, it is determined that there is a potential transparent obstacle boundary. It should be noted that glass doors themselves are transparent to radar waves and cannot be directly detected. However, metal door frames, handles, etc., have high RCS (radio frequency cross-section), and their edges produce diffraction effects, forming stable point cloud clusters. These characteristics can be used to indirectly locate the presence of a door. Traditional methods rely on visual markers or intensity information from lidar; this invention utilizes the sensitivity of millimeter-wave radar to metal edges to achieve marker-free glass door boundary detection.

[0043] In this embodiment, points with high radar cross-sections (RCS) (e.g., metal door frames, handles) are searched in the radar point cloud, and points with RCS greater than -5 dBsm are retained. Euclidean clustering is used (distance threshold 0.1 meters) to aggregate points with high RCS into clusters, and noisy clusters with fewer than 5 points are removed. The minimum bounding rectangle is calculated for each cluster, and the direction of the longer side is extracted. If two clusters exist whose geometric center distance is between 0.5 and 2.0 meters, and the line connecting them is approximately perpendicular to the robot's forward direction (angle less than 30°), these clusters are marked as door frame corner points. Connecting the left and right door frame corner points forms a virtual electromagnetic boundary topology. When the distance between the robot and this electromagnetic boundary is less than a preset threshold (e.g., 0.5 meters), a potential transparent obstacle boundary is identified.

[0044] This invention searches for high RCS point clusters, extracts edge diffraction features, and connects adjacent high RCS point clusters to form an electromagnetic boundary topology map. When the robot is less than a preset threshold away from this boundary, it is determined that there is a potential transparent obstacle boundary. Even when the glass door is completely closed and there are no visual features, the robot can perceive that "there is a potential door frame structure here", thereby slowing down in advance and preparing to pass through or stop.

[0045] S430. Within the azimuth interval corresponding to the electromagnetic boundary topology map, search for the peak value of the radar echo along the range dimension. S440. If there are two peaks, the distance difference between them corresponds to the thickness and dielectric constant of the transparent medium, and the power difference between the two peaks is within a preset range, then the transparent obstacle is determined to be in a closed state. S450, otherwise it is determined to be in an open state or without obstacles.

[0046] It should be noted that existing radar technology cannot distinguish whether a glass door is closed or open, causing robots to either hesitate to pass through (conservative) or blindly pass through (dangerous). This invention utilizes the bimodal characteristics generated by the reflection and transmission of electromagnetic waves on the front and back surfaces of a medium to determine the door's state from a physical perspective, without the need for additional sensors.

[0047] In this embodiment, the peak value of the radar echo is searched along the range dimension within the azimuth angle interval (e.g., from -20° to +20°) corresponding to the electromagnetic boundary topology map. Specifically, the original range-Doppler map is extracted, and local peak values ​​are searched along each range gate (using a three-point sliding window, the peak value must be at least 3dB higher than the two sides). The range and power of all peak values ​​are recorded. If there are two peak values, and their range difference is within the range of 1 cm to 5 cm (corresponding to common glass thicknesses of 4 to 12 mm and dielectric constants of 4 to 7), and the power ratio of the two peak values ​​(the larger power divided by the smaller power, expressed in decibels) is between 10dB and 15dB, then the transparent obstacle is determined to be in a closed state; otherwise, it is determined to be in an open state or there is no obstacle.

[0048] When the robot detects a closed transparent obstacle, it immediately stops and sends a voice or app notification saying "Glass door closed, please open," awaiting human intervention. When the robot detects an open transparent obstacle or no transparent obstacle, it is allowed to continue traversing the obstacle.

[0049] Throughout the entire process of the robot traversing the transitional environment (step S300), this invention continuously detects the opening and closing status of transparent obstacles (glass doors) based on radar echoes, and controls the robot's movement behavior in real time according to the detection results, as follows: High-reflectivity point clusters with RCS values ​​higher than -5dBsm are selected from the radar point cloud, corresponding to strong reflective structures such as metal door frames and door handles. Euclidean clustering algorithm is used to aggregate the high-reflectivity point clusters, and noise clusters with fewer than 5 points are removed. If there are two point clusters with a horizontal distance between 0.5m and 2.0m and an angle of less than 30° between the line connecting them and the robot's forward direction, they are marked as candidates for left and right door frames. The centroids of the two point clusters are connected to form an electromagnetic boundary topology map to locate the azimuth angle range of the doorway. When the distance between the robot and the electromagnetic boundary is less than 0.5m, the fine detection of the door opening and closing status is triggered.

[0050] Within the azimuth angle range corresponding to the aforementioned door frame, a one-dimensional range image of the radar echo is extracted along the range dimension. A sliding window maxima algorithm is used to search for local peaks in the echo power, requiring the peak power to be at least 3dB higher than the power on either side. All pairs of local peaks are iterated to determine if a double peak matching the characteristics of a closed glass door exists. (1) The distance difference between the two peaks matches the thickness of the conventional glass door and the relative permittivity of the glass, with the corresponding distance difference range being 1cm to 5cm; (2) The power difference between the two peaks is in the range of 10dB to 15dB, which corresponds to the normal penetration loss range of radar waves through glass; If both of the above conditions are met at the same time, the glass door is determined to be in a closed state; If no matching double peaks are detected, the glass door is determined to be open or without any glass obstruction.

[0051] When the processor detects that the glass door is closed, it immediately sends a stop command to the chassis, causing the robot to stop in place. At the same time, it triggers an audible and visual alarm and pushes a prompt message to the remote monitoring platform or user APP via wireless communication, waiting for manual intervention. When the processor detects that the glass door is open or there are no obstacles, the robot is allowed to continue to cross the transition area along the planned path.

[0052] S500: When the environment is determined to be outdoor and there is a usable global navigation satellite system signal, the current pose estimated based on radar echo is used as the reference anchor point and fused with the positioning result of the global navigation satellite system to complete the smooth switch to outdoor navigation mode.

[0053] In this embodiment, when the robot has completely traversed the transition area, and the classifier outputs an outdoor environment confidence score exceeding 0.8 for three consecutive frames, it determines that the robot has entered the outdoor environment. At this time, the processor monitors the positioning status of the GNSS module in real time. If the GNSS module obtains an RTK fixed solution and the signal is available, the following smooth switching operation is performed: The continuous pose estimated by the tight coupling of radar and IMU at the moment of switching is used as the reference anchor point. The latitude and longitude coordinates output by the GNSS module are converted into pose data in a local Cartesian coordinate system. Based on the reference anchor point, the radar pose and GNSS pose are smoothly weighted and fused through cubic spline interpolation to eliminate pose jumps. After the integration is completed, the robot officially switches to outdoor navigation mode and operates using a combined navigation scheme of GNSS, radar and IMU.

[0054] Specifically, when the environment is determined to be outdoor and the global navigation satellite system signal is available (e.g., a fixed solution is obtained), the current pose estimated based on radar echo (i.e., the pose output by the tightly coupled filter of the radar-inertial measurement unit) is used as the reference anchor point and fused with the positioning result of the global navigation satellite system.

[0055] Specifically, after converting the latitude and longitude output from the Global Navigation Satellite System (GNSS) to a local Cartesian coordinate system, cubic spline interpolation or linear weighted fusion methods are used to smoothly transition between the two pose sequences, eliminating abrupt changes. After fusion, the robot switches to outdoor navigation mode and begins using combined navigation with GNSS, radar, and inertial measurement unit (IMU).

[0056] Through the synergistic effect of the above steps, this invention ensures that the robot does not require vision, does not rely on GPS signal strength, and is unaffected by slippage during the entire transition from indoor to outdoor environments. The transition mode is triggered only when the environment classification is correct. The transition mode enables radar positioning to ensure the consistency of the coordinate system for subsequent obstacle detection. The obstacle detection results directly affect the crossing behavior. Outdoor fusion uses the reliable pose accumulated in the transition mode as a reference, eliminating jumps and preventing collisions or pauses. Changes in lighting do not affect perception, thus solving the problems of visual overexposure, laser inability to detect glass, and GPS switching jumps in traditional solutions.

[0057] This invention leverages the immunity to light and weather conditions of 79GHz millimeter-wave radar, fundamentally preventing visual failure due to sudden changes in lighting and the inability of lasers to detect glass, thus avoiding perception loss during indoor-outdoor transitions. Environmental classification is achieved through the statistical characteristics of radar echoes (frequency domain entropy and spatial entropy), independent of external signals. This avoids the inability to accurately determine environmental boundaries when GPS is unreliable or visual features are lacking, which is caused by reliance on GPS signal strength or visual features. Furthermore, the odometry weight is proactively reduced in transition zones, replacing it with radar point cloud matching for positioning, ensuring positioning continuity and preventing significant drift due to slippage caused by reliance on wheeled odometry in transition zones. Safe passage control is achieved by using radar echoes to detect transparent obstacles and their opening / closing states, preventing situations where the robot cannot distinguish between closed glass doors and open doorways, thus avoiding collisions. Finally, radar-estimated pose is used as an anchor point for smooth fusion, avoiding direct fusion of GPS and odometry, which can lead to pose jumps.

[0058] In one embodiment, the method further includes constructing an electromagnetic fingerprint map, specifically including: The location information, radar environment feature template, and geometric dimensions of each successfully traversed transition area are saved as an electromagnetic fingerprint node. When the robot approaches the saved electromagnetic fingerprint node again, it identifies the known node by matching the radar point cloud with the node position information, directly loads the node's parameters, and uses the node's absolute coordinates to perform closed-loop correction of the current pose.

[0059] In this embodiment, the present invention saves the information of the transition region successfully traversed each time as an electromagnetic fingerprint node (location, entropy template, geometric dimensions); when approaching again, it identifies known nodes by point cloud matching, loads parameters and corrects the pose in a closed loop, avoiding the need to re-perform high RCS clustering, bimodal detection and other calculations for each traversal, which is inefficient and the odometry cumulative error cannot be eliminated during long-term operation.

[0060] In this embodiment, the location information, radar environment feature template, and geometric dimensions of the transition area after each successful indoor-outdoor bidirectional crossing (complete indoor→transition→outdoor or outdoor→transition→indoor) are saved as electromagnetic fingerprint nodes. Each node includes: node ID and timestamp, coordinates of the door frame corner in the global coordinate system (given by indoor SLAM or outdoor global navigation satellite system), doorway width, a typical Doppler spectral entropy curve template (time series), and the robot's heading angle during crossing. The nodes are stored in the robot's local map database (such as SQLite) and optionally uploaded to the cloud.

[0061] When the robot approaches a saved electromagnetic fingerprint node again, it searches for candidate nodes within 2 meters based on the current estimated pose. For each candidate node, it iterative nearest-point registration is performed between the current radar point cloud and the global coordinates of the door frame corner points saved in the node. If the registration residual is less than 0.2 meters, the match is successful. At this point, the node's parameters (door width, entropy curve template, etc.) are directly loaded, skipping the dual-modal entropy calculation and bimodal detection steps. The global coordinates in the node are then used to perform graph optimization closed-loop correction on the current pose, eliminating odometry cumulative errors. This solves the computational redundancy problem of needing to re-extract a large number of features for each crossing and provides absolute constraints for long-term localization.

[0062] In one embodiment, it further includes: Real-time monitoring of radar health status, which is assessed based on at least one or more of the following: radar point cloud quantity, point cloud static point ratio, or inter-frame matching residual. When the health status is at level one, limit the robot's maximum speed and activate auxiliary sensors for slow path exploration. When the health status is at level two, the robot is controlled to stop in place, issue an alarm, and upload the raw radar data. When the health status is at level three and no human intervention is received within a preset time after stopping, the robot is controlled to backtrack along the original path to the nearest safe area.

[0063] In this embodiment, the health status of the 79GHz millimeter-wave radar is monitored in real time throughout the robot's operation, and a weighted comprehensive evaluation is performed based on one or more of the following indicators: the number of radar point clouds, the normalized residual of the nearest point matching between point cloud frames, and the proportion of static points (radial velocity absolute value less than 0.1 m / s) in the point cloud. The comprehensive health score is calculated by assigning weights of 0.5, 0.3, and 0.2 to the normalized value of the point cloud number (current point cloud number divided by reference point cloud number 500), the matching consistency index (1 minus normalized residual), and the proportion of static points (static point number divided by total point cloud number), respectively, and then summing them. This yields a radar health score in the range of 0 to 1, with a higher score indicating a more stable radar operating status. Three risk levels are defined based on the health score, corresponding to different safety control strategies: for example, a score greater than 0.6 is normal, 0.2 to 0.6 is slightly abnormal, and less than 0.2 is severely abnormal.

[0064] When the health status is at Level 1 (minor abnormality, health score 0.2-0.6), the robot's maximum speed is limited to 0.1 m / s, and auxiliary sensors (such as ultrasonic sensors and contact bumpers) are activated for slow pathfinding; the radar health status is continuously monitored; if the health score does not return to the normal range within 3 seconds, the robot is upgraded to Level 2.

[0065] When the health status is at Level 2 (severe abnormality, health score less than 0.2 and lasting for more than 0.5 seconds), the robot is immediately stopped in place, an audible and visual alarm is issued, and an alarm message "radar dirty / malfunctioning, manual cleaning required" is sent to the remote monitoring platform via wireless communication. At the same time, the raw radar intermediate frequency data of the last 10 seconds is uploaded for remote diagnosis; the robot waits for a manual intervention command, with a maximum wait time of 30 seconds.

[0066] When the health status is at level three (no human intervention received within the timeout period and battery power is greater than 20%), the robot is controlled to backtrack along the original path to the nearest safe area at a speed of 0.2 m / s. Every 0.5 m backtracks, the robot checks whether a valid radar point cloud has been obtained again. If there is no improvement after backtracking to the nearest safe area (such as the location of an indoor charging dock), the robot stops and enters low-power standby mode.

[0067] In one embodiment, the method further includes performing a reverse switch from outdoor to indoor, specifically including: When the robot is operating outdoors, it continuously monitors radar environmental characteristics; When radar environmental characteristics indicate a change from an outdoor environment to a transitional environment, the robot's maximum speed is limited, and the robot is controlled to center and align itself based on the width of the transitional area detected in the radar point cloud. At the same time, the positioning contribution of the Global Navigation Satellite System is turned off, and the indoor navigation mode is reactivated.

[0068] In this embodiment, when the robot is operating outdoors, it continuously monitors radar environmental characteristics (Doppler spectral entropy and spatial distribution entropy). When the radar environmental characteristics indicate a change from the outdoor environment to a transitional environment (i.e., the Doppler spectral entropy and spatial distribution entropy begin to rise, and the classifier outputs a transition confidence score exceeding 0.8 for three consecutive frames), it determines that the robot is about to enter the transitional area from the outdoors, triggering the reverse transition control mode and performing the following operations: That is, the robot's maximum speed is limited to 0.2 m / s, and the robot is controlled to center and align itself to enter based on the width of the transition area detected in the radar point cloud (performing the same center alignment control as S340 to S350). At the same time, the positioning contribution of the Global Navigation Satellite System is turned off to avoid positioning failure caused by a sudden drop in GNSS signal after entering indoors; Reactivate the indoor navigation mode (restart the indoor SLAM node or pure radar-inertial measurement unit positioning), and simultaneously detect the door frame boundary of the transition area based on radar point cloud. Perform center alignment control consistent with the forward crossing, and control the robot to smoothly enter the room along the center line of the doorway. Simultaneously perform open and closed status detection of transparent obstacles. If a glass door is detected to be closed, stop immediately and sound an alarm; if a door is detected to be open, continue crossing. After traversing the indoor environment, once the robot is fully indoors, its bimodal entropy returns to a high value, and the classifier outputs an indoor confidence score exceeding 0.8 for three consecutive frames. The system then switches back to indoor navigation mode. This completes a smooth reverse transition from outdoors to indoors, with no abrupt changes throughout the process. It achieves seamless, symmetrical switching between indoor and outdoor environments, avoiding positioning failures caused by sudden drops in GPS signal strength when transitioning from outdoors to indoors.

[0069] A robot indoor / outdoor switching avoidance system based on 79GHz millimeter-wave radar includes: At least one 79 GHz millimeter-wave radar, mounted on a robot, is used to transmit and receive 79 GHz millimeter-wave signals; One or more processors; Memory, which stores computer program instructions; When the processor executes program instructions, it implements any of the above-mentioned indoor / outdoor switching avoidance methods for robots based on 79GHz millimeter-wave radar.

[0070] A robot indoor / outdoor switching avoidance system based on 79GHz millimeter-wave radar includes: The extraction module is used to transmit and receive echo signals from a 79GHz millimeter-wave radar mounted on a robot, and extract radar environment features from the echo signals, including at least frequency domain statistical features and spatial distribution features. The frequency domain statistical features reflect the degree of disorder in the Doppler spectrum energy distribution of the radar echo, and the spatial distribution features reflect the density and uniformity of the radar point cloud in space. The determination module is used to determine the current environment category of the robot based on radar environmental characteristics and a classifier. The environment category includes at least indoor environment, transitional environment and outdoor environment. Trigger mode is used to trigger transition control mode when the environment is determined to be transitional. In transition control mode, the contribution of wheel odometry to localization is reduced or frozen to at least partially, and radar echo-based point cloud matching or micro-Doppler features are enabled for relative pose estimation to maintain the robot's continuous localization. The control module is used to detect the presence and open / closed status of transparent obstacles based on radar echoes during the robot's traversal of the transitional environment, and to control the robot's movement behavior according to the detection results, including: prohibiting the robot from moving forward when a transparent obstacle is detected to be in a closed state; and allowing the robot to traverse when a transparent obstacle is detected to be in an open state or when there is no transparent obstacle. The switching module is used to smoothly switch to outdoor navigation mode when the environment is determined to be outdoor and there is a usable global navigation satellite system signal. It uses the current pose estimated based on radar echo as the reference anchor point and fuses it with the positioning result of the global navigation satellite system.

[0071] The foregoing description of specific exemplary embodiments of the present invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it is obvious that many changes and variations can be made based on the above teachings. Although embodiments of the invention have been shown and described, these specific embodiments are merely explanations of the invention and are not intended to limit it. The specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. The purpose of selecting and describing exemplary embodiments is to explain the specific principles of the invention and its practical application, so that those skilled in the art, after reading this specification, can make modifications, substitutions, variations, and various choices and changes to the embodiments as needed without departing from the principles and spirit of the invention, provided that such modifications, substitutions, variations, and choices and changes are within the scope of the claims of the invention and are protected by patent law.

Claims

1. A robot indoor / outdoor switching avoidance method based on 79GHz millimeter-wave radar, characterized in that, Includes the following steps: Echo signals are acquired by a 79GHz millimeter-wave radar installed on a robot, and frequency domain statistical features and spatial distribution features are extracted from the echo signals as radar environment features. Based on the radar environmental characteristics, a classifier determines the current environmental category of the robot, which includes at least indoor environment, transitional environment and outdoor environment. When a transitional environment is identified, a transitional control mode is triggered. In the transitional control mode, the contribution of the wheel odometry to the positioning is reduced or frozen, and radar echo-based relative pose estimation is enabled to maintain the robot's continuous positioning. During the robot's journey through the transitional environment, transparent obstacles and their open / closed states are detected based on radar echoes, and the robot's movement behavior is controlled according to the detection results. When the environment is determined to be outdoor and the GPS signal is available, the current pose estimated based on radar echo is used as the reference anchor point and fused with the GPS positioning result to smoothly switch to outdoor navigation mode.

2. The indoor / outdoor switching avoidance method for robots based on 79GHz millimeter-wave radar according to claim 1, characterized in that, Detecting transparent obstacles and their open / closed states, including: Search for point clusters with high radar cross section in the radar point cloud, extract the edge diffraction features of the point clusters, and connect adjacent high RCS point clusters to form an electromagnetic boundary topology map. When the distance between the robot and the electromagnetic boundary topology is less than a preset threshold, it is determined that there is a potential transparent obstacle boundary. Within the azimuth interval corresponding to the electromagnetic boundary topology map, search for the peak value of the radar echo along the range dimension; If there are two peaks, the distance difference between them corresponds to the thickness and dielectric constant of the transparent medium, and the power difference between the two peaks is within a preset range, then the transparent obstacle is determined to be in a closed state. Otherwise, it is determined to be in an open state or without obstacles.

3. The robot indoor / outdoor switching avoidance method based on 79GHz millimeter-wave radar according to claim 1, characterized in that, In the transition control mode, perform one or more of the following operations: Freeze the integral accumulation of the wheel odometer to prevent drift caused by slippage or sliding; Disable the visual perception processing thread to avoid interference caused by sudden changes in lighting conditions; Ignore global navigation satellite system signals to avoid signal jumps caused by porch obstruction; Pose estimation is performed using a tightly coupled filter between the radar and the inertial measurement unit.

4. The robot indoor / outdoor switching avoidance method based on 79GHz millimeter-wave radar according to claim 3, characterized in that, The transition control mode also includes: Extract the velocity vector field of the ground point cloud, and calculate the divergence and curl of the velocity vector field; The stability of the robot's motion is determined based on the divergence, and the degree of slippage of the wheeled odometer is detected based on the curl. When the curl exceeds a preset threshold, the weight of the wheel odometer in state estimation is reduced, and the direct displacement estimation method based on ground point cloud phase correlation is enabled.

5. The robot indoor / outdoor switching avoidance method based on 79GHz millimeter-wave radar according to claim 1, characterized in that, The transition control mode also includes: Extract obstacle boundary points on the left and right sides from the radar point cloud, and calculate the robot's heading deviation and lateral offset. Steering commands are generated using pure tracking or model predictive control methods, enabling the robot to cross the centerline of the boundary point.

6. The robot indoor / outdoor switching avoidance method based on 79GHz millimeter-wave radar according to claim 5, characterized in that, Also includes: The safe crossing speed is calculated based on the difference between the detected width of the transition area and the robot's own width. The robot's motion speed is shaped by using an S-shaped velocity curve with limited jerk, achieving a smooth process of deceleration, crossing, and then accelerating.

7. The robot indoor / outdoor switching avoidance method based on 79GHz millimeter-wave radar according to claim 1, characterized in that, Also includes: The location information, radar environment feature template, and geometric dimensions of each successfully traversed transition area are saved as an electromagnetic fingerprint node. When the robot approaches the saved electromagnetic fingerprint node again, it identifies the known node by matching the radar point cloud with the node position information, directly loads the parameters of the node, and uses the absolute coordinates of the node to perform closed-loop correction of the current pose.

8. The robot indoor / outdoor switching avoidance method based on 79GHz millimeter-wave radar according to claim 1, characterized in that, Also includes: The health status of the radar is monitored in real time, and the health status is evaluated based on at least one or more of the following: radar point cloud quantity, point cloud static point ratio, or inter-frame matching residual. When the health status is at level one, limit the robot's maximum speed and activate auxiliary sensors for slow path exploration. When the health status is at level two, the robot is controlled to stop in place, issue an alarm, and upload the raw radar data. When the health status is at level three and no human intervention is received within a preset time after stopping, the robot is controlled to backtrack along the original path to the nearest safe area.

9. The robot indoor / outdoor switching avoidance method based on 79GHz millimeter-wave radar according to claim 1, characterized in that, Also includes: When the robot is operating outdoors, it continuously monitors the radar environmental characteristics; When the radar environmental characteristics indicate a change from an outdoor environment to a transitional environment, the robot's maximum speed is limited, and the robot is controlled to center and align itself to enter based on the width of the transitional area detected in the radar point cloud. At the same time, the positioning contribution of the Global Navigation Satellite System is turned off, and the indoor navigation mode is reactivated.

10. A robot indoor / outdoor switching avoidance system based on 79GHz millimeter-wave radar, characterized in that, include: At least one 79 GHz millimeter-wave radar, mounted on a robot, is used to transmit and receive 79 GHz millimeter-wave signals; One or more processors; Memory, which stores computer program instructions; When the processor executes the program instructions, it implements the robot indoor / outdoor switching avoidance method based on 79GHz millimeter-wave radar as described in any one of claims 1-9.

Images