Robot control method, chip and robot

Abstract

The invention discloses a robot control method, a chip and a robot, and the method comprises the steps: enabling the robot to obtain an environment image through an RGB camera, and enabling the robot to process the environment image obtained by the RGB camera according to a received control instruction; the robot processes the environment image through different networks according to different control instructions, the passable area in the environment image, the current position of the robot and the object type in the environment image can be obtained, the function is powerful, the multiple functions can be achieved only by installing one RGB camera in the robot, and the robot is convenient to use. The production cost of the robot is reduced, and the size and power consumption of the robot are effectively reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent robots, and particularly to a robot control method, a chip and a robot. Background Art

[0002] At present, most indoor intelligent mobile robots have the ability of autonomous positioning and navigation. The robot first needs to identify which areas in the environment are passable, and then during the process of performing tasks, it locates its own position and posture according to the existing map data to achieve autonomous navigation, and also needs to identify obstacles in the environment to avoid them during the autonomous navigation process. This requires the robot to use different sensors to achieve different functions, resulting in a relatively high production cost of the robot and a relatively large volume of the robot. Summary of the Invention

[0003] To solve the above problems, the present invention provides a robot control method, a chip and a robot. The specific technical solutions of the present invention are as follows: A robot control method, the robot includes an RGB camera, and the method includes the following steps: the robot obtains an environmental image through the RGB camera, and then the robot processes the environmental image obtained by the RGB camera according to the received control instruction; if the robot receives a region recognition command, the robot extracts the point cloud information of the environmental image from the environmental image through a depth prediction network, and then segments the passable region in the environmental image according to the point cloud information; if the robot receives a visual positioning command, the robot extracts feature points from the environmental image through a feature point recognition network, and then determines the current position of the robot according to the extracted feature points; if the robot receives an object recognition command, the robot recognizes an object from the environmental image through an object recognition network, and marks the recognized object on the walking map.

[0004] Further, the robot obtains an environmental image through the RGB camera, including the following steps: before the robot obtains an environmental image through the RGB camera, it first detects the environmental light intensity through the infrared sensing points outside the lens of the RGB camera; if the environmental light intensity is greater than or equal to the set light intensity, the robot controls the daytime filter of the IR-CUT dual filter of the RGB camera to work, and then obtains an environmental image through the RGB camera; if the environmental light intensity is less than the set light intensity, the robot controls the night filter and the fill light of the IR-CUT dual filter of the RGB camera to work, and then obtains an environmental image through the RGB camera.

[0005] Further, the robot extracts the point cloud information of the environmental image through a depth prediction network, including the following steps: The robot traverses the acquired environmental image to obtain the pixel value of each pixel point in the environmental image; The robot maps each pixel point in the environmental image through a fitting function; The robot converts the pixel value of each pixel point into a depth value during the mapping process to obtain the depth map of the environmental image; The robot constructs a coordinate conversion formula based on the internal parameters of the RGB camera; The robot converts the pixel coordinates of each pixel point in the depth map into camera coordinates through the coordinate conversion formula to obtain the point cloud information of the depth map.

[0006] Further, the robot segments the passable area in the environmental image based on the point cloud information, including the following steps: The robot preprocesses the calibrated point cloud information to remove the noise points in the point cloud information; The robot divides the preprocessed point cloud information through a voxel grid to obtain several voxels; The robot calculates the height variance of each voxel, and then compares the height variance of each voxel with the set variance; The robot sets the voxels with height variance less than the set variance as ground seeds; The robot takes the ground seeds as the starting point and uses the adjacent voxels that meet the requirements as ground seeds to gradually expand the ground area; The robot fits the acquired ground seeds through the ransac algorithm to obtain the passable area in the environmental image.

[0007] Further, the robot extracts feature points from the environmental image through a feature point recognition network, and the feature point recognition network is a SuperPoint network, including the following steps: The robot inputs the environmental image acquired by the RGB camera into a shared encoding network, and the shared encoding network performs convolutional processing and pooling processing on the environmental image to obtain the tensor of the environmental image; The robot inputs the tensor of the environmental image into a feature point decoding network, and the feature point decoding network performs convolutional processing on the tensor of the environmental image, and then performs data organization processing on the tensor of the environmental image to obtain the feature point probability of the pixel points of the environmental image; The robot inputs the tensor of the environmental image into a feature point decoding network, and the feature point decoding network first performs convolutional processing on the tensor of the environmental image, and then performs linear interpolation processing and normalization processing on the tensor of the environmental image to obtain the descriptor operator of the environmental image.

[0008] Further, after the robot extracts feature points from the environmental image, it matches the feature points, including the following steps: The robot extracts specific points and description operators from the previous frame of environmental image obtained by the RGB camera; The robot determines the corresponding description operator according to the feature point probability of the pixel points of the environmental image; The robot calculates the similarity of the corresponding description operators to obtain a similarity matrix; The robot performs augmentation processing on the similarity matrix, and then obtains the optimal assignment of the augmented lower similarity matrix; The robot performs a summation calculation on the optimal assignment of the obtained similarity matrix to remove the mismatched feature points and obtain the matched feature points.

[0009] Further, the robot determines its current position according to the positions of the extracted feature points in the environmental image and the positions of the corresponding feature points in the matched environmental image, including the following steps: The robot determines the pixel coordinates of the extracted feature points in the environmental image according to the positions of the extracted feature points in the environmental image; The robot converts the pixel coordinates of the feature points in the environmental image into robot coordinates through a coordinate transformation formula; The robot determines the pixel coordinates of the feature points in the matched environmental image according to the positions of the corresponding feature points in the matched environmental image; The robot converts the pixel coordinates of the feature points in the matched environmental image into robot coordinates through a coordinate transformation formula; The robot uses the feature points in the environmental image and the feature points in the matched environmental image as the coordinate origins respectively to construct a feature point coordinate system and determine the feature coordinates of the robot in this coordinate system; The robot determines its position in the current walking map according to the coordinate changes of the robot in the feature point coordinate system.

[0010] Further, the robot identifies objects from the environmental image through an object recognition network, including the following steps: The robot performs grid processing on the obtained environmental image and divides the environmental image into N*N grids; The robot detects the grids of the environmental image to obtain 2N prediction boxes and the confidence of each prediction box; The robot performs target classification on the 2N prediction boxes according to the confidence of the prediction boxes; The robot performs non-maximum suppression calculation on the classified prediction boxes to obtain the recognition result; where N is a natural number greater than 1.

[0011] A chip with a built-in control program configured to execute the above robot control method.

[0012] A robot, the robot includes a main control chip, an IMU sensor, a gyroscope, an RGB camera and a fill light, the main control chip is the above chip, and the RGB camera includes an infrared sensing point and an IR-CUT dual filter.

[0013] Compared with the existing technologies, the beneficial effects of the present invention are as follows: After the robot described in this application obtains the environmental image through the RGB camera, according to different control instructions, different networks are used to process the environmental image, and then the passable area in the environmental image, the current position of the robot, and the object types in the environmental image can be obtained. It has powerful functions, and only one RGB camera needs to be installed in the robot to achieve so many functions, which not only reduces the production cost of the robot, but also effectively reduces the volume and power consumption of the robot. Brief Description of the Drawings

[0014] Figure 1 It is a schematic flowchart of a robot control method in an embodiment of the present invention. Embodiment

[0015] The embodiments of the present invention will be described in detail below. The illustrated embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0016] With reference to the accompanying drawings of the specification, through further description of the specific embodiments of the present invention, the technical solutions and their beneficial effects of the present invention will be made clearer and more definite. The embodiments described below by referring to the drawings are exemplary and are intended to explain the present invention, and should not be construed as limiting the present invention.

[0017] Such as Figure 1As shown in the figure, a robot control method, the robot includes an RGB camera, and the method includes the following steps: The robot obtains an environmental image through the RGB camera, and then the robot processes the environmental image obtained by the RGB camera according to the received control instruction. If the robot receives a region recognition command, the robot extracts the point cloud information of the environmental image from the environmental image through a depth prediction network, and then segments the passable region in the environmental image according to the point cloud information. The depth prediction network is a monocular depth prediction network, extended from U-Net. U-Net: Convolutional Networks for Biomedical Image Segmentation, a convolutional neural network specifically used for medical image segmentation. U-Net has a U-shaped structure. The left half is the contracting path (Encoder), which is a traditional convolutional neural network structure with two 3x3 convolutional layers, each convolutional layer having a ReLU activation function and a 2x2 downsampling max pooling. The right half is the expansive path (Decoder). Each 2x2 convolution (up-convolution) halves the number of feature channels and connects to the corresponding cropped feature map in the contracting path. If the robot receives a visual positioning command, the robot extracts feature points from the environmental image through a feature point recognition network, and then determines the current position of the robot according to the extracted feature points. Feature points in the environmental image generally refer to corner points. A corner point is usually defined as the intersection of two edges. More strictly speaking, the local neighborhood of a corner point should have boundaries with different directions in two different regions. In practical applications, most so-called corner point detection methods detect image points with specific features, rather than just "corner points". These feature points have specific coordinates in the image and have certain mathematical features, such as local maximum or minimum gray level, gradient features, etc. If the robot receives an object recognition command, the robot recognizes objects from the environmental image through an object recognition network and marks the recognized objects on the walking map. The object detection algorithm mainly recognizes multiple objects in an image through an object detection network and can locate different objects (give bounding boxes). Object detection is useful in many scenarios, such as unmanned driving and security systems. After the robot in the present application obtains the environmental image through the RGB camera, according to different control instructions, processes the environmental image through different networks, and can obtain the passable region in the environmental image, the current position of the robot, and the object types in the environmental image. It has powerful functions, and only one RGB camera needs to be installed in the robot to achieve so many functions, which not only reduces the production cost of the robot, but also effectively reduces the volume and power consumption of the robot.

[0018] The robot obtains the environmental image through an RGB camera, including the following steps: Before the robot obtains the environmental image through the RGB camera, it first detects the environmental light intensity through the infrared sensing points outside the lens of the RGB camera. If the environmental light intensity is greater than or equal to the set light intensity, the robot controls the daytime filter of the IR-CUT dual filter of the RGB camera to work, and then obtains the environmental image through the RGB camera. If the environmental light intensity is less than the set light intensity, the robot controls the night filter and fill light of the IR-CUT dual filter of the RGB camera to work, and then obtains the environmental image through the RGB camera. The IR-CUT dual filter refers to a set of filters built into the RGB camera lens group. When the infrared sensing points outside the lens detect changes in the intensity of light, the built-in IR-CUT automatic switching filter can automatically switch according to the intensity of external light, so that the image reaches the best effect. That is to say, in the daytime or at night, the dual filter can automatically switch filters, so that the best imaging effect can be obtained whether in the daytime or at night. With the help of the IR-CUT dual filter and fill light, the robot can obtain a clear environmental image in the night environment, enabling the robot to avoid obstacles in the night environment, and it has high practicability.

[0019] As one of the embodiments, in step S1, the depth prediction network is a monocular depth prediction network. The robot extracts the depth map of the environmental image from the environmental image through the monocular depth prediction network, including the following steps: The robot traverses the obtained environmental image to obtain the pixel value of each pixel point in the environmental image. The robot maps each pixel point in the environmental image through a fitting function. During the mapping process, the robot converts the pixel value of each pixel point into a depth value to obtain the depth map of the environmental image. The robot extracts the depth map of the environmental image from the environmental image through the monocular depth prediction network without setting a 3D sensor on the robot, reducing the production cost of the robot.

[0020] As one of the embodiments, in step S2, the robot obtains the point cloud information in the depth map through the internal parameters of the RGB camera, including the following steps: The robot constructs a coordinate conversion formula according to the internal parameters of the RGB camera. The robot converts the pixel coordinates of each pixel point in the depth map into camera coordinates through the coordinate conversion formula to obtain the point cloud information of the depth map. The robot can obtain the point cloud information from the depth map through coordinate conversion, with high flexibility.

[0021] As one of the embodiments, in step S2, the robot calibrates the point cloud information through IMU data and gyroscope data, including the following steps: The robot records the IMU data and gyroscope data when the RGB camera acquires the environmental image during movement. The robot calculates the shooting angle of the RGB camera on the robot when acquiring the environmental image according to the recorded IMU data and gyroscope data. The robot calibrates the camera coordinates of the point cloud information according to the acquired shooting angle to obtain the calibrated point cloud information. When the robot encounters jitter, such as crossing a bump, the robot will have a pitch change, resulting in an offset or distortion of the point cloud information of the acquired environmental image. Therefore, the robot needs to calculate the elevation angle of the RGB camera on the robot when acquiring the environmental image through the moving distance and the number of wheel rotations in the IMU data and gyroscope data, and then translate the camera coordinates of the point cloud information according to the elevation angle of the RGB camera. For example, if the shooting angle of the robot's RGB camera is an elevation angle, the robot translates the camera coordinates of the point cloud information downward; if the shooting angle of the robot's RGB camera is a depression angle, the robot translates the camera coordinates of the point cloud information upward. The upward and downward translation distances are determined by the angle size and the internal parameters of the RGB camera. The robot calibrates the point cloud information through IMU data and gyroscope data to improve the accuracy of the calculation results.

[0022] As one of the embodiments, in step S3, the robot operates on the calibrated point cloud information through a segmentation algorithm, including the following steps: The robot preprocesses the calibrated point cloud information to remove the noise points in the point cloud information. The robot divides the preprocessed point cloud information through a voxel grid to obtain a number of voxels. The robot calculates the height variance of each voxel, and then compares the height variance of each voxel with the set variance. The robot sets the voxels with height variance less than the set variance as ground seeds. Starting from the ground seeds, the robot takes the adjacent voxels that meet the requirements as ground seeds to gradually expand the ground area. The robot fits the acquired ground seeds through the ransac algorithm to obtain the passable area in the environmental image. The robot can quickly pass through the front area by segmenting the passable area from the point cloud information through the segmentation algorithm.

[0023] As one of the embodiments, the robot fits the acquired ground seed points through the RANSAC algorithm, including the following steps: The robot randomly selects several ground seeds from the acquired ground seeds as inliers, and then calculates the parameter model using the inliers through the model calculation formula. The robot traverses the acquired ground seeds using the parameter model, sets the ground seeds with errors less than or equal to the set error as inliers, and sets the ground seeds with errors greater than the set error as outliers. The robot repeats the above two steps for a set number of times, and sets the parameter model with the most inliers as the best model. The robot calculates the final model using the inliers of the best model through the model calculation formula. The robot traverses the acquired ground seeds using the final model to obtain the inliers of the final model. The robot sets the area formed by the inliers of the final model as the passable area in the environmental image.

[0024] As one of the embodiments, the preprocessing performed by the robot on the calibrated point cloud information includes removing outliers, downsampling, and data normalization. Removing outliers is to perform a statistical analysis on the neighborhood of each point and calculate its average distance to all neighboring points. Assuming that the obtained result is a Gaussian distribution, whose shape is determined by the mean and standard deviation, then the points with average distances outside the standard range (defined by the global distance average and variance) can be defined as outliers and removed from the data. Downsampling is the process of reducing the sampling rate of a specific signal, usually used to reduce the data transmission rate or data size. Data normalization is a basic task in data mining. Different evaluation indicators often have different dimensions and dimension units, and such situations will affect the results of data analysis. In order to eliminate the dimensional influence between indicators, data normalization processing is required to solve the comparability between data indicators. After the original data undergoes data normalization processing, each indicator is at the same order of magnitude and is suitable for comprehensive comparison and evaluation.

[0025] As one of the embodiments, the robot extracts feature points from the environmental image through a feature point recognition network. The feature point recognition network is the SuperPoint network, including the following steps: The robot inputs the environmental image acquired by the RGB camera into the shared encoding network, and the shared encoding network performs convolutional processing and pooling processing on the environmental image to obtain the tensor of the environmental image. The robot inputs the tensor of the environmental image into the feature point decoding network, and the feature point decoding network performs convolutional processing on the tensor of the environmental image, and then performs data organization processing on the tensor of the environmental image to obtain the feature point probabilities of the pixel points of the environmental image. The robot inputs the tensor of the environmental image into the feature point decoding network, and the feature point decoding network first performs convolutional processing on the tensor of the environmental image, and then performs linear interpolation processing and normalization processing on the tensor of the environmental image to obtain the descriptor of the environmental image.

[0026] As one of the embodiments, after the robot extracts feature points from the environmental image, it matches the feature points, including the following steps: The robot extracts specific points and description operators from the previous frame of environmental image obtained by the RGB camera. The robot determines the corresponding description operator according to the feature point probability of the pixel points of the environmental image. The robot calculates the similarity of the corresponding description operators to obtain a similarity matrix. The robot performs augmentation processing on the similarity matrix, and then obtains the optimal assignment of the augmented lower similarity matrix. The robot performs a summation calculation on the optimal assignment of the obtained similarity matrix to remove the mismatched feature points and obtain the matched feature points.

[0027] As one of the embodiments, the robot determines its current position according to the positions of the extracted feature points in the environmental image and the positions of the corresponding feature points in the matched environmental image, including the following steps: The robot determines the pixel coordinates of the extracted feature points in the environmental image according to the positions of the extracted feature points in the environmental image. The robot converts the pixel coordinates of the feature points in the environmental image into robot coordinates through a coordinate transformation formula. The robot determines the pixel coordinates of the feature points in the matched environmental image according to the positions of the corresponding feature points in the matched environmental image. The robot converts the pixel coordinates of the feature points in the matched environmental image into robot coordinates through a coordinate transformation formula. The robot takes the feature points in the environmental image and the robot coordinates of the feature points in the matched environmental image as the coordinate origin respectively, constructs a feature point coordinate system, and determines the feature coordinates of the robot in this coordinate system. The robot determines its position in the current walking map according to the coordinate change of the robot in the feature point coordinate system.

[0028] As one of the embodiments, the robot identifies an object from the environmental image through an object recognition network, including the following steps: The robot performs grid processing on the obtained environmental image and divides the environmental image into N*N grids. The robot detects the grids of the environmental image to obtain 2N prediction boxes and the confidence of each prediction box. The robot performs target classification on the 2N prediction boxes according to the confidence of the prediction boxes. The robot performs non-maximum suppression calculation on the classified prediction boxes to obtain the recognition result. Wherein, N is a natural number greater than 1.

[0029] A chip with a built-in control program configured to execute the above-mentioned robot control method.

[0030] A robot, which includes a main control chip, an IMU sensor, a gyroscope, an RGB camera and a fill light. The main control chip is the above-mentioned chip, and the RGB camera includes an infrared sensing point and an IR-CUT dual filter.

[0031] In the description of the specification, the description referring to terms such as "in one embodiment", "preferably", "example", "specific example" or "some examples", etc. means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. The schematic expressions of the above terms in this specification do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in a suitable manner in any one or more embodiments or examples. The connection manners described in the description of the specification have obvious effects and practical effectiveness.

[0032] Through the description of the above structures and principles, those skilled in the art should understand that the present invention is not limited to the above specific embodiments. Improvements and substitutions using well-known technologies in the art based on the present invention fall within the protection scope of the present invention, which should be defined by each claim.

Claims

1. A robot control method, characterized in that, The robot includes an RGB camera, and the method includes the following steps: The robot obtains an environmental image through the RGB camera, and then processes the environmental image obtained by the RGB camera according to the received control instruction; If the robot receives a region recognition command, the robot extracts the point cloud information of the environmental image from the environmental image through a depth prediction network, and then segments the passable region in the environmental image according to the point cloud information; If the robot receives a visual positioning command, the robot extracts feature points from the environmental image through a feature point recognition network, and then determines the current position of the robot according to the extracted feature points; If the robot receives an object recognition command, the robot recognizes the object from the environmental image through an object recognition network, and marks the recognized object on the navigation map.

2. The robot control method according to claim 1, wherein The robot obtains an environmental image through the RGB camera, including the following steps: Before the robot obtains an environmental image through the RGB camera, it first detects the environmental light intensity through the infrared sensing points outside the lens of the RGB camera; If the environmental light intensity is greater than or equal to the set light intensity, the robot controls the daytime filter of the IR-CUT dual filter of the RGB camera to work, and then obtains the environmental image through the RGB camera; If the environmental light intensity is less than the set light intensity, the robot controls the night filter and fill light of the IR-CUT dual filter of the RGB camera to work, and then obtains the environmental image through the RGB camera.

3. The robot control method according to claim 1, characterized in that, The robot extracts the point cloud information of the environmental image from the environmental image through a depth prediction network, including the following steps: The robot traverses the obtained environmental image to obtain the pixel value of each pixel point in the environmental image; The robot maps each pixel point in the environmental image through a fitting function; The robot converts the pixel value of each pixel point into a depth value during the mapping process to obtain the depth map of the environmental image; The robot constructs a coordinate conversion formula according to the internal parameters of the RGB camera; The robot converts the pixel coordinates of each pixel point in the depth map into camera coordinates through the coordinate conversion formula to obtain the point cloud information of the depth map.

4. The robot control method according to claim 3, wherein The robot segments the passable region in the environmental image according to the point cloud information, including the following steps: The robot preprocesses the calibrated point cloud information to remove the noise points in the point cloud information; The robot divides the preprocessed point cloud information through a voxel grid to obtain a number of voxels; The robot calculates the height variance of each voxel, and then compares the height variance of each voxel with the set variance; The robot sets the voxels with height variance less than the set variance as ground seeds; The robot uses the ground seeds as the starting point and takes the adjacent voxels that meet the requirements as ground seeds to gradually expand the ground area; The robot fits the obtained ground seeds through the ransac algorithm to obtain the passable region in the environmental image.

5. The robot control method according to claim 1, wherein The robot extracts feature points from the environmental image through a feature point recognition network, and the feature point recognition network is the SuperPoint network, including the following steps: The robot will input the environmental image obtained by the RGB camera into the shared encoding network, and the shared encoding network will perform convolutional processing and pooling processing on the environmental image to obtain the tensor of the environmental image; The robot will input the tensor of the environmental image into the feature point decoding network. The feature point decoding network will perform convolutional processing on the tensor of the environmental image, and then perform data organization processing on the tensor of the environmental image to obtain the feature point probability of the pixel points of the environmental image; The robot will input the tensor of the environmental image into the feature point decoding network. The feature point decoding network will first perform convolutional processing on the tensor of the environmental image, and then perform linear interpolation processing and normalization processing on the tensor of the environmental image to obtain the descriptor operator of the environmental image.

6. The robot control method according to claim 5, wherein After the robot extracts feature points from the environmental image, it matches the feature points, including the following steps: The robot extracts specific points and descriptor operators from the previous frame of environmental image obtained by the RGB camera; The robot determines the corresponding descriptor operator according to the feature point probability of the pixel points of the environmental image; The robot performs similarity calculation on the corresponding descriptor operators to obtain a similarity matrix; The robot performs augmentation processing on the similarity matrix, and then obtains the optimal assignment of the augmented lower similarity matrix; The robot performs summation calculation on the optimal assignment of the obtained similarity matrix to remove the mismatched feature points and obtain the matched feature points.

7. The robot control method according to claim 6, wherein The robot determines its current position according to the positions of the extracted feature points in the environmental image and the positions of the corresponding feature points in the matched environmental image, including the following steps: The robot determines the pixel coordinates of the extracted feature points in the environmental image according to the positions of the extracted feature points in the environmental image; The robot converts the pixel coordinates of the feature points in the environmental image into robot coordinates through the coordinate transformation formula; The robot determines the pixel coordinates of the corresponding feature points in the matched environmental image according to the positions of the corresponding feature points in the matched environmental image; The robot converts the pixel coordinates of the feature points in the matched environmental image into robot coordinates through the coordinate transformation formula; The robot takes the robot coordinates of the feature points in the environmental image and the robot coordinates of the corresponding feature points in the matched environmental image as the coordinate origin respectively, constructs a feature point coordinate system, and determines the feature coordinates of the robot in this coordinate system; The robot determines its position in the current walking map according to the coordinate changes of the robot in the feature point coordinate system.

8. The robot control method according to claim 1, wherein The robot identifies objects from the environmental image through the object recognition network, including the following steps: The robot performs grid processing on the obtained environmental image and divides the environmental image into N*N grids; The robot detects the grids of the environmental image to obtain 2N prediction boxes and the confidence of each prediction box; The robot performs target classification on the 2N prediction boxes according to the confidence of the prediction boxes; The robot performs non-maximum suppression calculation on the classified prediction boxes to obtain the recognition result; Wherein, N is a natural number greater than 1.

9. A chip with a built-in control program, characterized in that, This program is configured to execute the robot control method described in any one of claims 1 to 8.

10. A robot, characterized in that, The robot includes a main control chip, an IMU sensor, a gyroscope, an RGB camera, and a fill light. The main control chip is the chip described in claim 9. The RGB camera includes infrared sensing points and an IR-CUT dual filter.

Images