Systems and methods for augmenting visual output from robotic devices

The method allows human operators to adjust robotic device paths and environment recognition by marking drivable regions, enhancing accuracy and efficiency without retraining neural networks.

JP7581320B2Active Publication Date: 2024-11-12TOYOTA JIDOSHA KK
View PDF 10 Cites 0 Cited by

Patent Information

Application Number
JP2022503979
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Priority Date
2019-09-13
Filing Date
2020-06-05
Publication Date
2024-11-12
Estimated Expiration
2040-06-05

AI Technical Summary

Technical Problem

Conventional systems for autonomous agents require retraining neural networks to correct visual output errors, which is inefficient and time-consuming.

Method used

A method and apparatus for displaying and editing visual output from robotic devices, allowing human operators to modify the agent's output by marking drivable and non-drivable regions without retraining the neural network, using pixel-level comparisons and keyframe matching.

Benefits of technology

Enables real-time adjustment of the robotic device's path and environment recognition, improving accuracy and efficiency by updating neural network decisions based on user input, reducing the need for extensive retraining.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007581320000001
    Figure 0007581320000001
  • Figure 0007581320000002
    Figure 0007581320000002
  • Figure 0007581320000003
    Figure 0007581320000003
Patent Text Reader

Abstract

A method for visualizing data generated by a robotic device is presented. The method includes displaying an intended path of the robotic device through an environment. The method also includes displaying a first area in the environment identified as drivable by the robotic device. The method further includes receiving input to identify a second area in the environment as drivable and to transmit the second area to the robotic device.
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] This application claims the benefit of U.S. Provisional Patent Application No. 62 / 877,792, filed June 23, 2019, entitled “Keyframe Matcher,” U.S. Provisional Patent Application No. 62 / 877,791, filed June 23, 2019, entitled “Visual Teaching and Iteration for Manipulation-Teaching VR,” and U.S. Provisional Patent Application No. 62 / 877,793, filed June 23, 2019, entitled “Visualization,” which are all incorporated herein by reference in their entireties.

[0002] Certain aspects of the present disclosure relate generally to robotic devices, and more particularly to systems and methods for providing a user interface for displaying and editing visual output from a robotic device. [Background technology]

[0003] Autonomous agents (e.g., vehicles, robots, drones, etc.) and semi-autonomous agents use machine vision to analyze regions of interest in their surrounding environments. While driving, an autonomous agent may rely on a trained neural network to identify objects present in a region of interest in an image of the surrounding environment. For example, the neural network may be trained to identify and follow objects captured by one or more sensors, such as a Light Detection and Ranging (LIDAR) sensor, a sonar sensor, an RGB camera, an RGB-D camera, etc. The sensors may be connected to or in communication with a device such as the autonomous agent. An object detection application for an autonomous agent may analyze the sensor image data to detect objects (e.g., pedestrians, bicyclists, other vehicles, etc.) in the scene surrounding the autonomous agent.

[0004] In conventional systems, a visual output of the agent may be provided to an operator. For example, the agent's intended path, such as an intended route, may be output to the operator. If the output is incorrect, the agent may be retrained to obtain a correct output. It would be desirable to improve the system to correct the agent's output without having to retrain the associated neural network. Summary of the Invention

[0005] In one aspect of the present disclosure, a method for visualizing data generated by a robotic device is disclosed. The method includes displaying an intended path of the robotic device in an environment. The method also includes displaying a first area in the environment identified as drivable for the robotic device. The method further includes receiving input to identify a second area in the environment as drivable. The method further includes transmitting the second area to the robotic device.

[0006] In another aspect of the present disclosure, a non-transitory computer readable medium is disclosed having stored thereon a non-transitory program code for visualizing data generated by a robotic device. The program code, when executed by a processor, includes program code for displaying an intended path of the robotic device through an environment. The program code also includes program code for displaying a first area in the environment identified as navigable by the robotic device. The program code further includes receiving an input for identifying a second area in the environment as navigable. The program code further includes program code for transmitting the second area to the robotic device.

[0007] Another aspect of the present disclosure relates to an apparatus for visualizing data generated by a robotic device. The apparatus includes a memory and one or more processors coupled to the memory. The processor displays an intended path of the robotic device through an environment. The processor also displays a first area in the environment identified as drivable for the robotic device. The processor further receives an input to identify a second area in the environment as drivable. The processor further transmits the second area to the robotic device.

[0008] The foregoing has outlined broadly the features and technical advantages of the present disclosure in order that the detailed description that follows may be better understood. Additional features and advantages of the present disclosure are set forth below. It should be appreciated by those skilled in the art that the present disclosure may be readily used as a basis for modifying or designing other structures for carrying out the same purposes of the present disclosure. It should also be recognized by those skilled in the art that such equivalent constructions do not depart from the teachings of the present disclosure as defined by the appended claims. The novel features believed to be characteristic of the present disclosure, both in terms of its organization and method of operation, together with further objects and advantages thereof, will be better understood from the following description when considered in conjunction with the accompanying drawings. It is to be expressly understood, however, that each of the drawings is provided for the purpose of illustration and description only and is not intended to define the limits of the present disclosure. [Brief description of the drawings]

[0009] The features, nature, and advantages of the present disclosure will become more apparent from the detailed description set forth below when considered in conjunction with the drawings in which like reference characters correspond throughout.

[0010] [Figure 1] 1 illustrates an example of an operator controlling a robotic device according to an aspect of the present disclosure. [Figure 2A] 1 illustrates an example of a robotic device operating in an environment according to an aspect of the present disclosure. [Figure 2B] 1 illustrates an example of a robotic device operating in an environment according to an aspect of the present disclosure. [Figure 2C]1 illustrates an example of a robotic device operating in an environment according to an aspect of the present disclosure. [Figure 3A] 1 illustrates an example of visual output generated based on information provided by a robotic device operating in an environment according to an aspect of the present disclosure. [Figure 3B] 1 illustrates an example of visual output generated based on information provided by a robotic device operating in an environment according to an aspect of the present disclosure. [Figure 4] 1 illustrates an example of a hardware implementation of a robotic device according to an aspect of the present disclosure. [Diagram 5] 1 illustrates a graphical sequence of operations taught according to aspects of the present disclosure. [Figure 6] 1 illustrates software modules for a robotic system according to an aspect of the present disclosure. [Figure 7] 1 illustrates a method for controlling a robotic device according to an aspect of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] The detailed description provided below in connection with the accompanying drawings is intended to describe various configurations and is not intended to present a single configuration for implementing the concepts described herein. The detailed description includes specific details for the purpose of providing a thorough understanding of the various concepts. However, it will be apparent to those skilled in the art that these concepts may be implemented without these specific details. In some instances, well-known structures and components are shown in block diagram form to avoid obscuring such concepts.

[0012] Based on the teachings, it should be understood by those skilled in the art that the scope of the present disclosure is intended to include any aspect of the present disclosure, whether implemented independently or combined with other aspects of the present disclosure. For example, an apparatus may be implemented or a method may be performed using any number of the aspects disclosed. In addition, the scope of the present disclosure is intended to include such an apparatus or method implemented using structures and functions in addition to or other than the various aspects disclosed in this disclosure. It should be understood that any aspect of the present disclosure may be embodied by one or more elements of a claim.

[0013] The word "exemplary" is used herein to mean "serving as an example, instance, or illustration." Any aspect herein described as "exemplary" is not necessarily to be construed as preferred or advantageous over other aspects.

[0014] Although specific embodiments are described herein, the scope of the disclosure includes numerous variations and permutations of these embodiments. Although some benefits and advantages of the preferred embodiments are described, the scope of the disclosure is not intended to be limited to any particular benefit, use, or purpose. Rather, the embodiments of the disclosure are intended to be broadly applicable to different technologies, system configurations, networks, and protocols, some of which are shown for illustrative purposes in the figures and description of the preferred embodiments. The detailed description and drawings are intended to be merely illustrative of the disclosure, rather than limiting, and the scope of the disclosure is defined by the appended claims and equivalents.

[0015] Autonomous and semi-autonomous agents perform tasks on detected objects. For example, an agent may navigate an environment, identify objects in the environment, and / or interact with objects in the environment. In this application, a robot or robotic device refers to an autonomous agent or semi-autonomous agent. For simplicity, a robotic device will be referred to as a robot.

[0016] Aspects of the present disclosure are not limited to a particular type of robot. Various types of robotic devices are contemplated that have one or more vision systems that provide visual / visualization output. The visual output may be provided to an operator as a feed video.

[0017] According to aspects of the present disclosure, the robot is locomotion-enabled, i.e., it has the ability to change the position of the end effector as a function of joint configuration. In one configuration, the robot has automatic full-body control and planning capabilities, allowing a human operator to demonstrate seamless end effector motion in a task space in virtual reality (VR) with little or no concern for kinematic constraints or robot pose.

[0018] In addition to training the robot in a VR environment, it would be ideal to update the training during testing time (e.g., in a real-world environment). Aspects of the present disclosure relate to providing a user interface for displaying and editing a path predicted by a neural network for the robot. Aspects of the present disclosure also relate to visualization of keyframe matching and outputting one or more charts visualizing the processing performed by the robot.

[0019] In one configuration, the robot includes one or more sensors, such as a red-green-blue-and-distance (RGB-D) field of view sensor on a pan-tilt head. The RGB-D image is a combination of an RGB image and a corresponding depth image. The depth image is an image channel in which each pixel is associated with the distance between the image plane and the corresponding object on the RGB image.

[0020] The robot may navigate through an environment based on images obtained from a source video (e.g., source images) of the environment. The robot may also perform tasks based on images from the source video. In traditional vision systems, the robot identifies images (e.g., images of objects) using a feature-based approach.

[0021] That is, the task is performed based on features extracted from an image in the robot's current vision. In feature point based approaches, a test image (e.g., the current image) is identified based on the similarity between the test image and a reference image. In particular, traditional vision systems compare the features of a test image with those of a reference image. A test image is identified when its features match those of the reference image.

[0022] A feature is a distinct (i.e. unique) feature of an image (e.g. corners, edges, high contrast regions, low contrast regions, etc.). An image may contain multiple features. A descriptor encodes the features in an image. A feature vector may be an example of a descriptor. In traditional vision systems, a descriptor is limited to encoding unique features of an image (e.g. objects in the image).

[0023] In most cases, the descriptors are robust to image transformations (e.g., localization, scale, brightness, etc.), i.e., conventional vision systems can identify objects in a variety of conditions, such as different lighting conditions. Nevertheless, robustness is limited because the descriptors are limited to encoding features unique to the image. To improve the ability of a robot to identify images and perform tasks, it is ideal to improve the robustness of the image vision system.

[0024] Aspects of the present disclosure relate to assigning descriptors to pixels of an image. That is, in contrast to conventional vision systems, in aspects of the present disclosure, the descriptors are not limited to unique features. Thus, the accuracy of the vision system may be improved. In particular, conventional vision systems are limited to comparing obvious features (e.g., features) of a test image with obvious features of a reference image. Images contain many more pixels than features. Thus, by comparing pixels instead of features, the accuracy of the comparison is increased.

[0025] That is, in one configuration, the vision system compares pixels of a test image to pixels of a reference image. The reference image may be referred to as a keyframe. The keyframes may be acquired during training. The use of keyframes provides invariance to pose and image transformations. The vision system determines that the test image matches the keyframe when the number of matching pixels is greater than a threshold.

[0026] In one configuration, the vision system compares pixel descriptors to identify matching pixels between the test image and the keyframe. The pixel descriptors include pixel-level information and depth information. The pixel-level information includes information such as the RGB value of the pixel and the context of the pixel within the image / surrounding pixels. The depth information indicates the distance from the surface to which the pixel corresponds to the sensor used to capture the image of that surface.

[0027] During training, as the robot performs a task, one or more sensors capture image data of the scene. Keyframes are generated from the captured image data. A keyframe may be conceptualized as a memory of the environment when the robot was performing a particular task. That is, a keyframe may be set as an anchor to a task or a location in the environment. After testing, when the robot matches a current image with a keyframe, the trained function corresponding to the keyframe may be executed.

[0028] Pixels in a keyframe may be assigned a descriptor. A descriptor may be a vector or array of values. For example, a descriptor may be a vector with 16 elements. The descriptor (e.g., a pixel descriptor) may be used to identify matching images.

[0029] After training, the robot may be initialized to perform one or more trained tasks. The robot may collect raw image data as it performs the commanded tasks. The raw image data is mapped at the pixel level to pixels in the keyframes as the robot passes through and / or manipulates objects in the environment. The correspondence between the current image and the keyframes is determined based on the similarity of pixel descriptors.

[0030] That is, the robot compares the pixel descriptor values ​​of the current image to the pixel descriptors of the key frame. The confidence of the comparison indicates the likelihood that a pixel of the current image matches a pixel of the key frame. A diagonal / parallel correspondence (e.g., a non-intersecting correlation) between multiple pixels of the current image data and the key frame may indicate a match; an intersecting correlation indicates a non-match between the current image data and the key frame.

[0031] In one configuration, a robot is positioned in an environment and image data of the scene is collected. The robot is then controlled (e.g., through a virtual reality interface) to perform a task. Restricting the human operator's view to the robot's view in virtual reality improves the robot's ability to perform tasks independently during training.

[0032] A robot may be taught to perform actions / tasks by parameterizing motions performed by an operator through a virtual reality interface. For example, the virtual reality interface may include the use of paddles, hand-held controllers, paintbrush tools, wiping tools, and / or placement tools that an operator manipulates while wearing a headset to paint the VR environment. Thus, rather than teaching motions in the task space directly, a human operator teaches a set of parameterized primitives (or motions). The parameterized primitives combine collision-free motion planning with hybrid (position and force) Cartesian control to reduce taught parameters and provide robustness during execution.

[0033] Parameterized motion means learning a task by breaking it down into a small number of discrete motion chunks. Each motion is defined by a set of parameters, such as joint angle changes, rotation angles, etc. The values ​​of these parameters may be configured and updated based on the robot's situation when performing the task.

[0034] Parameterized actions may be learned and extracted from one learned task and combined with other corresponding tasks to form a larger task. A parameterized action such as opening a door with a rotating handle may be implemented as performing the opening of any door handle (e.g., a door that requires a 30 degree turn, or a door that requires a 60 degree turn, or more turns). For example, the angle of rotation may be one parameter that defines the parameterized action for opening a door with a rotating door handle.

[0035] To perform a task, the robot may be located in the same or a similar environment (relative to an initial position during training). The robot may be located at a different starting point and may assume a different initial pose relative to the starting position (e.g., joint angles are adjusted to a different starting position). The robot may be tasked with performing the same task (e.g., a set of parameterized actions) (without control by a human operator), such as picking up a bottle, opening a cabinet, and placing a bottle in the cabinet. For example, the robot may perform the same task by updating parameters of the actions taught in a virtual reality controlled sequence. The parameters may be updated based on the current pose and / or position of the robot compared to the pose and / or position used during training.

[0036] As discussed, to update parameters, the robot takes an initial image of the scene and maps pixels and / or descriptors from the new image to the keyframe. The mapping defines a relative transformation between the initial image and the keyframe. The keyframe can be mapped to the new image using the relative transformation. The relative transformation may be defined by the changes in the robot's x-position, y-position, z-position, roll, pitch, and yaw. The relative transformation may be used to update the parameters of the parameterized motion from the taught parameters to the observed situation.

[0037] Relative transformations may be applied to parameterized movements. By applying relative transformations to parameterized movements, the robot may perform the same task as previously taught, even if the initial position and / or pose is changed. The robotic system may continually map pixel and / or dense neural network descriptors from the current scene to those from the keyframes, such that adjustments are made at intervals (e.g., successively) from the parameterized movements. For example, relative transformations may be applied to taught actions defined by a set of parameterized movements, such as pulling open a drawer, opening a door, picking up a cup or bottle, etc.

[0038] In some aspects, actions may be scene-wide and / or object-specific. For example, picking up a bottle may require the use of scene-wide keyframes to navigate to the bottle. Once the bottle is close, keyframes specific to the bottle may be analyzed independently of the environment.

[0039] Run behaviors may be used to move the robot from one point to another, which may allow the robot to locate objects that may be located anywhere in the environment during training of the "pick up" action and perform a task such as "pick up" a bottle regardless of its location. Manipulate behaviors may be used to move parts of the robot (e.g., the torso and / or arms) to make contact with a desired object.

[0040] FIG. 1 illustrates an example of an operator 100 controlling a robot 106 according to an embodiment of the present disclosure. As shown in FIG. 1, the operator 100 includes a vision system 102 and a motion controller 104 (e.g., a gesture following system) for controlling the robot 106. In this example, the operator 100 may control the robot 106 to perform one or more tasks. In the example of FIG. 1, the robot 106 is trained in a kitchen 108. Aspects of the present disclosure are not limited to training the robot 106 in a kitchen 108, as other environments are contemplated.

[0041] The vision system 102 may provide feed video as well as capture the vision of the operator 100. The operator 100 may be located remotely from the robot 106. In this example, the robot 106 is located in a kitchen 108 and the operator 100 is located in a different location from the kitchen 108, such as in a robot control center 114.

[0042] The vision system 102 may provide a feed video of the robot 106. For example, the vision system 102 may provide a view of the kitchen 108 based on a forward perspective of the robot 106. Other perspectives, such as a 360° landscape, may be provided. The perspectives are provided using one or more visual sensors, such as a video camera on the robot 106. The vision system 102 is not limited to a headset as shown in FIG. 1. The vision system 102 may be a monitor 110, an image projector, or other device capable of displaying the feed video from the robot 106.

[0043] One or more actions of the robot 106 may be controlled via the motion controller 104. For example, the motion controller 104 may capture gestures of the operator 100 and have the robot 106 mimic the captured gestures. The operator 100 may control the movement, limb movements, and other actions of the robot 106 via the motion controller 104. Aspects of the present disclosure are not limited to capturing the gestures of the operator 100 via the motion controller 104. Other types of gesture capture systems are contemplated. The operator 100 may control the robot 106 via a wireless connection 112. In addition, the robot 106 may provide feedback, such as a feed video, to the operator 100 via the wireless connection 112.

[0044] The robot 106 may be trained to navigate a particular environment (e.g., the kitchen 108) and / or similar environments. Additionally, the robot 106 may be trained to perform tasks on objects in the particular environment and / or similar objects in any environment. For example, the robot 106 may be trained to open and close drawers in the kitchen 108. Training may be implemented only for drawers in the kitchen 108 and / or for similar drawers in any environment, such as drawers in other kitchens.

[0045] As discussed, it is ideal to update the training during testing time (e.g., in a real-world environment). Aspects of the present disclosure relate to providing a user interface to display and edit the path determined by the neural network for the robot. In one configuration, the robot provides an intended path, and the user may update the intended path by marking pixels as drivable or undrivable.

[0046] During training, an operator may navigate the robot through the environment. Additionally or alternatively, the robot may navigate the environment automatically during training. The robot stores the paths traveled in memory. The stored paths traveled are used to train the network to determine pixels in the images that are associated with drivable paths and pixels in the images that correspond to non-drivable paths. The robot is trained to determine drivable paths in a new environment or in the same environment as that used for training.

[0047] A drivable path is a path that the robot may travel. The robot may not travel a non-drivable path. For example, an object on the path may make the path non-drivable. In one configuration, depth data is used to determine drivable paths. If the height of the object is above a threshold, the portion of the path with the object is marked as non-drivable.

[0048] FIG. 2A illustrates an example of a robot 200 in an environment 202 according to an embodiment of the present disclosure. For clarity, FIG. 2A is a top view of the environment 202. As shown in FIG. 2A, the environment 202 includes a dining table 204, a sink 206, and a counter 210. In this example, during training, the robot 200 traveled along a path 212 in the environment 202. The robot 200 stores the path 212 in a memory and uses the path 212 to train the neural network of the robot 200. The training is used to determine drivable paths in environments different from the current environment 202 in addition to the current environment 202.

[0049] 2B illustrates an example of an intended path 226 of the robot 200 determined based on a trained neural network according to an embodiment of the present disclosure. The neural network is trained based on one or more past paths, such as path 212 from the training environment 202. The current environment 250 of FIG. 2B is different from the training environment 202 of FIG. 2A. Nevertheless, the robot 200 may determine its intended path through the training environment 202 during or after training.

[0050] The example of Figure 2B is based on the perspective of the robot 200. The image of Figure 2B may be output to a remote operator and displayed on a remote display device. As shown in Figure 2B, the intended path 226 is to travel around the table 220. In addition to the intended path 226, the neural network determines a drivable region 228. The intended path 226 is within the boundaries of the drivable region 228. In one configuration, the drivable region 228 overlaps with the intended path 226. Pixels in the image are identified as belonging to the drivable region 228.

[0051] 2B, the neural network identifies an object 224 (e.g., a code on a refrigerator 222) on the floor. Based on training, the depth (e.g., height) of the object 224, and / or other factors, the neural network determines that the area of ​​the object 224 is not drivable. Thus, the drivable area 228 does not intersect with the area of ​​the object 224.

[0052] In one configuration, when the viewpoint of the robot 200 is output to an operator, the operator determines whether the neural network correctly identified the drivable region 228. The operator may update the drivable region 228 to modify the neural network's decision. Updates may be provided through user input, such as marking an area in the display as drivable or non-drivable.

[0053] 2C illustrates an example of updating the drivable area 228 of the robot 200 according to an embodiment of the present disclosure. As shown in FIG 2C, a human operator updates the image of the environment 250 by marking a new drivable area 230. Additionally or alternatively, the human operator may mark the drivable area 228 as not drivable (not shown in FIG 2C). After editing the drivable area 228 in the current environment, the neural network takes the edits by the human operator into account when determining a subsequent path.

[0054] The core weights of the neural network are not changed by the new user-edited information. In one configuration, a support vector machine (SVM) is added to the end of the network output. The weights of the SVM are updated based on the user-edited information. The update of the SVM weights may be done in real time. Thus, updating the SVM is faster than retraining the entire network. Furthermore, it is possible to adjust the importance of the newly taught drivable / undrivable regions using a user interface. For example, a slider may allow the user to adjust the weights of the new drivable / undrivable regions. Thus, the SVM can be made to make the appropriate classification in real time.

[0055] As discussed, after training, the robot may be initialized to perform one or more trained tasks. For example, the robot may be tasked with navigating an environment and / or manipulating objects in the environment. As the robot performs the commanded task, it collects raw image data. The raw image data is compared to the keyframes.

[0056] If one or more key frames match the current image data, the action associated with the matching key frames may be executed. For example, the robot may open or close a drawer. As another example, the robot may open or close a faucet. Additionally, or instead, an action may be added to a queue for execution upon completion of a previous action. As another example, the robot may determine a current location based on the key frames, which may be used to navigate a path from the current location to another location. As the robot moves through the environment, or as the positions of objects in the environment change, new key frames may be referenced.

[0057] In one configuration, one or more pixels are filtered before comparing the pixel descriptors. A function may be implemented to filter outliers. For example, a random sample consensus (RANSAC) function may be used to filter pixel outliers by randomly sampling the observed data. Filtering the pixels reduces noise and improves accuracy.

[0058] As discussed, the depth and / or angle of the current image may differ from the depth and / or angle of the matched keyframe. Thus, in one configuration, a relative transformation between the matched pixels of the current image data and the pixels of the keyframe may be determined. That is, a pose delta may be determined based on a comparison of the current image and the keyframe. The delta is the change in the pose of the robot from the keyframe to the current image.

[0059] The pose delta may be determined based on the depth information in the pixel descriptor. The depth information is the distance from the sensor to the object corresponding to the pixel. The pose delta is the change in (x,y,z) coordinates, roll, pitch, and yaw between the current image and the keyframe. That is, the pose delta describes how the image has changed from the keyframe to the current image based on the movement of the sensor.

[0060] There is an increased accuracy of the pose delta in relation to the increased number of matched pixels. In the present disclosure, the number of matched pixels may be greater than the number of matched features. Thus, aspects of the present disclosure improve the accuracy of pose delta determination compared to conventional systems that compare images based solely on features. A least squares function may be used to determine the pose delta.

[0061] A pose delta (e.g., relative translation) between one or more values ​​of the pixel descriptor of a pixel may be used to update the parameters of the action to be performed. For example, if the pose delta indicates that the robot is one foot away from an object compared to a position defined in a keyframe, the action may be updated to take into account the current position (e.g., pose) of the robot. The position defined in the keyframe is the position to use when training the robot to perform the task.

[0062] 3A illustrates another example of output 300 from a vision system according to an embodiment of the present disclosure. As shown in FIG. 3A, the output 300 includes a current image 302, a key frame 308, a histogram 306, and a pose confidence window 304. The output 300 may be displayed at a location remote from the current position of the robot. For example, the output 300 may be displayed at a control center (e.g., robot control center 114 of FIG. 1) that controls the operation of the robot.

[0063] Current image 302 is an image from the robot's current field of view. Keyframe 308 is one keyframe from the set of keyframes that is compared to current image 302. Histogram 306 shows the confidence score for a match from current image 302 to each keyframe in the set of keyframes.

[0064] During training, the robot captures keyframes at specific points along a path in the running task from the start to the finish. The keyframes captured along the path are used as a set of keyframes in the running task. During testing, the robot compares its current image to the set of keyframes to identify its current location.

[0065] Based on the number of matched pixels (e.g., pixel descriptors), the robot determines the confidence of a match from the current image to a particular keyframe. If the confidence of the match is higher than a threshold, the robot performs a task associated with the matched keyframe. The task may include location determination (e.g., determining the robot's current location). In the example of FIG. 3A, the robot determines that it is at a location on the path that corresponds to the location of the keyframe with the highest confidence of the match 310.

[0066] The histogram 306 may be different depending on the task. The histogram 306 in FIG. 3A is for a running task. The histogram 306 may be different if the robot is performing a manipulation task (e.g., object manipulation). The set of keyframes for the manipulation task may have fewer keyframes compared to the set of keyframes for the running task.

[0067] For example, when opening a cabinet, the robot may be trained to open the cabinet when it is closed or partially open. In this example, the histogram may have two keyframe matching bars in the bar graph: one bar indicating the confidence that the current image matches the keyframe with a closed cabinet, and another bar indicating the confidence that the current image matches the keyframe with a partially open cabinet.

[0068] The pose confidence window 304 indicates the confidence of the pose match obtained from the pose matcher. In one configuration, before determining the pose delta, the robot determines whether a number of criteria satisfy one or more thresholds. The criteria may be based on depth, pose, cliques, errors, and other factors.

[0069] In one configuration, the robot determines the number of pixels that have depth values. On surfaces such as glass or shiny surfaces, the robot may not be able to determine depth. Thus, pixels corresponding to these surfaces may not have depth values. The number of pixels that have depth values ​​may be determined before filtering the pixels. If the number of pixels that have depth values ​​is greater than a threshold, the depth criterion is met. The pose confidence window 304 may include a bar graph that compares a bar graph 320 of the number of pixels that have depth values ​​to a bar graph 322 that indicates a threshold value. The bar graph may be color coded.

[0070] Additionally or alternatively, the bar graph in the pose confidence window 304 may include a bar graph of the number of aligned features between the current image and the key frame compared to a bar graph of an alignment threshold. To determine the accuracy of the match between the key frame and the current image, the robot may apply a pose transformation to determine how many features of an object in the current image align with features of an object in the key frame, and vice versa. A feature may not align if it is associated with a non-static object. For example, a key frame may contain a cup that is no longer present in the current image. Thus, the cup feature may not align between the images. The number of aligned features may be referred to as clique inliers. A pose delta may be generated if the clique inliers are greater than a threshold.

[0071] Additionally or alternatively, the bar graph in the pose confidence window 304 may include a bar graph of the number of features used to determine the pose compared to a bar graph of a pose threshold. That is, a certain number of features should be used to determine the pose. If the number of features used to determine the pose is less than the pose threshold, then a pose delta may not be calculated. The number of features used to determine the pose may be referred to as pose inliers.

[0072] The bar graph in the pose confidence window 304 may also include a bar graph of the root mean square (RMS) error of the pose delta compared to a threshold. If the RMS error is less than the RMS error threshold, the pose delta may be satisfactory. If the RMS error is greater than the RMS error threshold, the user or robot may not use the pose delta to perform the task.

[0073] The pose confidence window 304 is not limited to a bar graph, other graphs or images may be used, and the criteria for the pose confidence window 304 are not limited to the criteria discussed, other criteria may be used.

[0074] Figure 3B illustrates yet another example of output 350 from a vision system according to an embodiment of the present disclosure. As shown in Figure 3B, output 350 includes a current image 352, a keyframe 358, a histogram 354, and a pose confidence window 356. Output 350 may be displayed at a location remote from the robot's current position.

[0075] The pose confidence window 356 visualizes the number of features and the RMS error over the match path. For each feature (e.g., baseline), the current value is compared to a threshold. For example, for depth, the current depth (389) is compared to a threshold (65). Each bar in the bar graph relates to a current value and a threshold value.

[0076] The histogram 354 visualizes the match score of each keyframe that is matched with the current image. The keyframe with the highest match score above a threshold may be selected as the matching keyframe. The matched keyframe may be displayed as keyframe 358 in the output 350.

[0077] As discussed, aspects of the present disclosure relate to a mobile manipulation hardware and software system capable of autonomously performing human-level tasks in a real-world environment after being taught the tasks by demonstration from a human in virtual reality. In one configuration, a mobile manipulation robot is used. The robot may include full-body task-space hybrid position / force control. Additionally, as discussed, the robot is taught parameterized primitives linked to a robustly learned dense visual embedding representation of the scene. Finally, a task graph of the taught behaviors may be generated.

[0078] Rather than programming and training a robot to recognize a fixed set of objects or perform predefined tasks, aspects of the present disclosure allow the robot to learn new objects and tasks from human demonstrations. The learned tasks may be performed autonomously by the robot under naturally changing conditions. The robot can be taught to associate a given set of actions to any scene and object from a single example, without using previous object models or maps. The vision system may be trained offline using existing supervised and unsupervised datasets, and the rest of the system may function without additional training data.

[0079] In contrast to conventional systems that directly teach task-space movements, aspects of the present disclosure teach a set of parameterized movements that combine collision-free motion planning and hybrid (position and force) Cartesian control of the end effector to minimize taught parameters and provide robustness during execution.

[0080] In one configuration, a task-specific dense embedding is computed over the trained vision. This pixel-wise embedding links the parameterized behavior to the scene, allowing the system to handle a variety of environments with greater robustness at the expense of generalizing to new situations.

[0081] Task actions may be taught independently using visual input conditions and success-based exit conditions. Actions may be linked together in a dynamic task graph. Because actions are linked, the robot may reuse actions to execute a task sequence.

[0082] The robot may have multiple degrees of freedom (DOF). For example, the robot may have 31 degrees of freedom (DOF) divided into five subsystems: chassis, torso, left arm, right arm, and head. In one configuration, the chassis includes four steerable wheels (e.g., 8 DOF total) that provide "quasi-holonomic" mobility. The drive / steering actuator package may include various motors and gear heads. The torso may have 5 DOF (yaw, pitch, pitch, pitch, yaw). Each arm may have 7 DOF. The head may have 2 DOF, pan / tilt. Each arm may include a 1 DOF gripper with underactuated fingers. Aspects of the present disclosure are not limited to the robots discussed above. Other configurations are contemplated. In one example, the robot may include a custom tool such as a sponge or mop.

[0083] In one configuration, the robot is integrated with force / torque sensors to measure interaction forces with the environment. For example, force / torque sensors may be located at the wrists of each arm. The head may be integrated with perception sensors to provide a wide field of view and provide a VR context for the human or robot to perform tasks.

[0084] Aspects of the present disclosure provide several levels of abstraction for robot control. In one configuration, the lowest control level provides real-time coordinated control of all degrees of freedom of the robot. Real-time control may include joint control and part control. The joint control implements low-level device communication and exposes device commands and states in a generic way. In addition, the joint control supports actuators, force sensors, and inertial measurement units. The joint control may be configured at run-time to support different robots.

[0085] Component control can handle higher level collaboration of the robot by breaking the robot into components (e.g. right arm, head, etc.) and providing a set of parameterized controllers for each component. Component control may provide controllers for joint position and velocity, joint admittance, camera vision, chassis position and velocity, as well as pose, velocity, and admittance control in a hybrid task space.

[0086] Task-space control of the end effector allows another level of abstraction in robot control. This level of abstraction solves the problem of posing the robot to achieve the desired motion. The inverse kinematics (IK) of the whole body for hybrid Cartesian control is formulated and solved as a second-order program. Parts may have linear constraints on joint positions, velocities, accelerations, and gravitational torques.

[0087] Whole-body IK may be used to plan motion to reach a goal pose in Cartesian coordinates. In one configuration, occupied voxels of the environment are fitted with spheres or capsule shapes. Voxel collision constraints are added to the quadratic IK program to avoid collisions between the robot and the world. Motion planning may be performed using a rapidly-exploring random tree (RRT) in the quadratic IK program, sampling in Cartesian space as the steering function between nodes.

[0088] Planning in Cartesian space results in natural and straightforward motion. Using a quadratic IK program as the steering function can improve the reliability of the plan, and the same controller may be used for planning and execution to reduce discrepancies between the two. Similarly, RRT can be used in combination with motion planning towards joint position goals and a part-controlled joint position controller that acts as the steering function.

[0089] The next level of abstraction defines parameterized behaviors. In one configuration, parameterized behaviors are primitive actions that can be parameterized and combined to accomplish a task. Behaviors may include, but are not limited to, manipulation actions such as grab, lift, place, pull, retract, wipe, and direct control; navigation actions such as moving a joint, driving by velocity command, driving by position command, and path following with active obstacle avoidance; and preparatory actions such as look and stop.

[0090] Each motion can have one or more different types of actions, such as motion at one or more joints of the robot's parts or in Cartesian coordinates. Each action can use a different control strategy, such as position, velocity, or admittance control, and can choose to use motion planning to avoid external obstacles. The motion of the robot, whether or not it uses motion planning, avoids self-collisions and satisfies motion control constraints.

[0091] Each operation is parameterized by different actions, which in turn may have their own parameters. For example, a gripping operation may consist of four parameters: gripper angle, 6D approach, grip, and (optionally) gripper lift pose. In this example, these parameters define a sequence of predefined actions: (1) open the gripper to the desired gripper angle, (2) plan and execute a collision-free path to the 6D approach pose, (3) move the gripper to the 6D gripping pose and stop on contact, (4) close the gripper, and (5) move the gripper to the 6D lift pose.

[0092] The final level of control abstraction is the task. In one construction, a task is defined as a sequence of actions that allows the robot to manipulate and navigate the human environment. A task graph (see Figure 5) is a valid, cyclic or acyclic graph with different tasks as nodes and different movement situations as edges, including anomaly detection and recovery. Edge situations include execution of each action to handle different objects and environments, inspection of an object in hand using force / torque sensors, voice commands, and matching keyframes.

[0093] According to an aspect of the present disclosure, a perception pipeline is designed for a robot to understand its surrounding environment. The perception pipeline also gives the robot the ability to recognize which action to take based on the taught task. In one configuration, a fused RGB-D image is created by projecting multiple depth images of a high-resolution color stereo pair onto one field of view image (e.g., a wide field of view left image). The system runs a set of deep neural networks to provide various pixel-level classifications and feature vectors (e.g., embeddings). Based on visual features recalled from the taught sequence, the pixel-level classifications and feature vectors are accumulated into a temporary 3D voxel representation. The pixel-level classification and feature vectors may be used to invoke actions to be performed.

[0094] In some configurations, no object categories are defined, nor are any models of objects or environments assumed. Rather than explicitly detecting and segmenting objects and explicitly estimating 6-DOF poses, dense pixel-level embeddings may be generated for a variety of tasks. Reference embeddings from taught sequences may be used to perform motion classification or pose estimation.

[0095] The trained model may be fully convolutional. In one configuration, pixels of the input image are each mapped to a point in an embedding space. The embedding space is given a loss function defined by the output of the model and a metric implicitly defined by the training procedure. The trained model may be used for a variety of tasks.

[0096] In one configuration, given an annotated example, the trained model detects all objects in a semantic class. Objects in a semantic class may be detected by comparing embeddings in the annotation with embeddings in other regions. The model may be trained with a discriminative loss function.

[0097] The model may be trained to determine object instances. The model may identify and / or count distinct objects. The model may be trained to predict a vector (a 2D embedding) for each pixel. The vector may point to the centroid of the object containing that pixel. At runtime, pixels that point to the same centroid may be grouped as a segment of the scene. Execution at runtime may be in 3D.

[0098] A model may be trained on 3D correspondences, providing per-pixel embeddings that are invariant to view and lighting, such that any view of a 3D point in the scene is mapped to the same embedding. The model may be trained using a loss function.

[0099] Pixel-wise embeddings (and depth data) for each RGB-D frame are fused into a dynamic 3D voxel map. Each voxel accumulates first and second order position, color and embedding statistics. Dynamic object lifetimes are based on the back projection of the voxel into the depth image. The voxel map is partitioned using standard graph segmentation based on semantic and instance labels, as well as geometric proximity. The voxel map is dimensionally reduced to a 2.5D map with classification statistics for elevation and drivability.

[0100] The 2.5D map is used for collision-free chassis motion, while the voxel map is used for collision-free whole-body motion planning. Voxels in the map may be grouped into capsules using a greedy algorithm for collision checking in 3D. Segmented objects may be used for motion to attach to the hand when the object is grasped.

[0101] The robot may be trained by a one-shot learning approach to recognize features in the scene (or of a particular manipulated object) that are highly related to features recorded in previously taught tasks. As the task is performed by the user, features are saved in the form of keyframes across the task. The keyframes may be RGB images with multidimensional embeddings and per-pixel depth (if available).

[0102] The embeddings act as feature descriptors that can establish pixel-by-pixel correspondences at run-time under the assumption that the current image is similar enough to the reference images present at training time. Since depth is present at (almost) every pixel, the correspondences can be used to solve for the pose delta between the current and reference images. Inliers may be detected using Euclidean constraints, and a Levenberg-Marquardt least-squares function is applied along with RANSAC to solve for the pose with 6 degrees of freedom.

[0103] The pose deltas serve as corrections that can be applied to adapt the taught sequence of movements to the current scene. Because embeddings may be defined for each pixel, keyframes may be broad enough to include all pixels in the image, or narrow enough to use only pixels within a user-defined mask. As discussed, the user may define the mask by selectively annotating regions in the image as being relevant to the task or on the object.

[0104] In addition to visual sensing, in one configuration the robot collects and processes voice input. Voice provides another set of embeddings as input for teaching the robot. As an example, the robot gets voice input by asking a question and understanding the spoken language of the response from the human. The voice response may be understood using a custom keyword detection module.

[0105] The robot may utilize a fully convolutional keyword spotting model to understand a custom wake word, a set of objects (e.g., "mug" or "jar"), and a set of locations (e.g., "cabinet" or "fridge"). In one configuration, the model listens for a wake word at some interval, e.g., 32 ms. Once the wake word is detected, the robot notes whether any object or location keywords are detected. During training, artificial noise is added to make the recognition more robust.

[0106] As discussed, to teach a task to the robot, the operator uses a set of VR modes. Each action may have a corresponding VR mode to set and command parameters specific to that action. Each action mode may include customized visualizations, depending on the type of parameter, to aid in setting each parameter. For example, when setting parameters for a door-pulling action, the hinge axis is labeled and visualized as a line, and possible poses for pulling the gripper are constrained to lie on a circular arc centered on the hinge. To aid in the teaching process, several utility VR modes are used, such as action reconstruction, annotating the environment with relevant objects, repositioning the virtual robot, camera images, and a menu of the VR world.

[0107] During execution, the robot's pose and parts in the environment may differ from those used during training. Feature matching may be used to find features in the environment similar to those taught. A pose delta may be established from the matched feature correspondences. The motion taught by the user may change according to the calculated pose delta. In one configuration, multiple keyframes are passed to the matching problem. Based on the number of correspondences, the best matching keyframe is selected.

[0108] 4 illustrates an example hardware implementation of a robotic system 400 according to an embodiment of the present disclosure. The robotic system 400 may be a component of an autonomous or semi-autonomous system, such as a vehicle, a robotic device 428, or other device. In the example of FIG. 4, the robotic system 400 is a component of a robotic device 428. The robotic system 400 may be used to control the actions of the robotic device 428 by inferring the intent of an operator of the robotic device 428.

[0109] The robotic system 400 may be implemented with a bus architecture generally represented as bus 430. The bus 430 may include any number of interconnected buses and bridges depending on the particular application and overall design constraints of the robotic system 400. The bus 430 connects various circuits such as one or more processors and / or hardware modules represented as processor 420, a communication module 422, a position module 418, a sensor module 402, a locomotion module 426, a memory 424, a visualization module 408, and a computer readable medium 414. The bus 430 may also connect various other circuits such as timing sources, peripherals, voltage regulators, and power management circuits that are known to those skilled in the art and therefore will not be described further.

[0110] The robotic system 400 includes a transceiver 416 connected to a processor 420, a sensor module 402, a visualization module 408, a communication module 422, a position module 418, a movement module 426, a memory 424, and a computer readable medium 414. The transceiver 416 is connected to an antenna 434. The transceiver 416 communicates with various devices via a transmission medium. For example, the transceiver 416 may receive instructions from an operator of the robotic device 428 via communication. As discussed herein, the operator may be remote from the robotic device 428. As another example, the transceiver 416 may transmit information regarding key frame matching from the visualization module 408 to the operator.

[0111] The robotic system 400 includes a processor 420 coupled to a computer readable medium 414. The processor 420 performs processes including executing software stored on the computer readable medium 414 that provides functionality according to the present disclosure. The software, when executed by the processor 420, causes the robotic system 400 to perform various functions described for particular devices, such as the robotic device 428 or modules 402, 408, 414, 416, 418, 420, 422, 424, 426. The computer readable medium 414 may also be used to store data that is manipulated by the processor 420 when executing the software.

[0112] The sensor module 402 may be used to obtain measurements via different sensors, such as a first sensor 406 and a second sensor 404. The first sensor 406 may be a visual sensor, such as a stereo camera or an RGB camera for capturing 2D images. The second sensor 404 may be a ranging sensor, such as a LiDAR sensor, a RADAR sensor, or an RGB-D sensor. Of course, aspects of the present disclosure are not limited to the above sensors, and other types of sensors, such as temperature, sonic, and / or laser, are also contemplated as either of the sensors 404, 406. The measurements from the first sensor 406 and the second sensor 404 may be processed by one or more of the processor 420, the sensor module 402, the communication module 422, the position module 418, the movement module 426, and the memory 424, in conjunction with the computer-readable medium 414, to implement the functionality described herein. In one configuration, the data captured by the first sensor 406 and the second sensor 404 may be transmitted to an operator via the transceiver 416 as a feed video. The first sensor 406 and the second sensor 404 may be connected to or in communication with the robotic device 428 .

[0113] The location module 418 may be used to determine the location of the robotic device 428. For example, the location module 418 may use a Global Positioning System (GPS) to determine the location of the robotic device 428. The communication module 422 may be used to facilitate communication via the transceiver 416. For example, the communication module 422 may provide communication capabilities via different wireless protocols, such as WiFi, long term evolution (LTE), 3G, etc. The communication module 422 may also be used to communicate with other components of the robotic device 428 that are not modules of the robotic system 400.

[0114] The locomotion module 426 may be used to facilitate locomotion of the robotic device 428 and / or components of the robotic device 428 (e.g., limbs, hands, etc.). For example, the locomotion module 426 may control the movement of the limbs 438 and / or the wheels 432. As another example, the locomotion module 426 may be in communication with a power source of the robotic device 428, such as an engine or a battery.

[0115] The robotic system 400 also includes a memory 424 for storing data related to the operation of the robotic device 428 and the visualization module 408. The modules may be software modules executing within the processor 420, resident / stored in the computer-readable medium 414 and / or memory 424, one or more hardware modules connected to the processor 420, or a combination thereof.

[0116] The visualization module 408 may be in communication with the sensor module 402, the transceiver 416, the processor 420, the communication module 422, the position module 418, the locomotion module 426, the memory 424, and the computer readable medium 414. In one configuration, the visualization module 408 generates an intended path for the robotic device through an environment. The intended path may be transmitted to a remote location, such as a robot operations center. A visualization corresponding to the intended path may be displayed at the remote location.

[0117] The visualization module 408 also identifies drivable areas in the environment. The drivable areas may be determined by a neural network based on previously driven routes. Based on training, depth information, and other information, areas in the environment may be identified as non-drivable. The intended route may be transmitted to a remote location, such as a robot operations center. The remote location may display a visualization corresponding to the intended route.

[0118] In one configuration, the visualization module 408 receives information identifying a second region in the environment as drivable. The information may be received via communication from a remote location. A user at a remote location marks pixels in an image of the environment as drivable. Similarly, the visualization module 408 may receive information identifying an area as not drivable. Information regarding drivable and / or not drivable areas may be received to correct areas previously identified by the visualization module 408. For example, the visualization module 408 may erroneously mark an area as not drivable. The received information may correct the erroneous marking.

[0119] The visualization module 408 may be trained to determine an intended route and a drivable area based on one or more previously driven routes. The drivable area is intended to be displayed overlaid on the intended route. After receiving updates from the remote location, the visualization module 408 identifies a third area as drivable based on the received updates.

[0120] In one configuration, the visualization module 408 provides information for displaying at the remote location a comparison of a current image and a keyframe image. The comparison includes a visualized mark identifying matched pixels in the current image and the keyframe image. Additionally, the visualization module 408 may provide information for displaying at the remote location a confidence of a match between the current image and each keyframe in the keyframe set. Further, the visualization module 408 may provide information for displaying at the remote location a graph comparing one or more pose delta terms to one or more thresholds. The information may include depth information, clique information, pose information, RMS error information, and / or other information.

[0121] Figure 5 illustrates an example graph sequence 500 of operations taught in accordance with an aspect of the present disclosure. As shown in Figure 5, the graph sequence 500 includes a start node 502 and an end node 504. The graph sequence 500 may branch or loop based on sensed visual input, audio input, or other conditions.

[0122] For example, as shown in FIG. 5, after start node 502, the robot may execute an action "listen_for_object." In this example, the robot determines whether it has sensed a visual or audio input corresponding to a cup or a bottle. In this example, a different sequence of actions is executed based on whether the sensed input corresponds to a cup or a bottle. Aspects of the present disclosure are not limited to the actions shown in FIG. 5.

[0123] 6 illustrates examples of software modules for a robotic system according to an embodiment of the present disclosure. The software modules of FIG. 6 may use one or more components of the hardware system of FIG. 4, such as the processor 420, the communication module 422, the position module 418, the sensor module 402, the movement module 426, the memory 424, the keyframe module 408, and the computer readable medium 414. Aspects of the present disclosure are not limited to the modules illustrated in FIG. 6.

[0124] As shown in Fig. 6, the robot may receive audio data 604 and / or image data 602. The image data 602 may be an RGB-D image. The audio network 606 may listen for a wake word at an interval. The audio network 606 receives the raw audio data 604 to detect wake words and extract keywords from the raw audio data 604.

[0125] A neural network, such as a dense embedding network 608, receives image data 602. The image data 602 may be received at intervals. The dense embedding network 608 processes the image data 602 and outputs an embedding 610 of the image data 602. The embedding 610 and the image data 602 may be combined to generate a voxel map 626. The embedding 610 may also be input to a keyframe matcher 612.

[0126] The keyframe matcher 612 compares the embedding 610 to multiple keyframes. A matching keyframe is identified if the embedding 610 corresponds to an embedding of a keyframe. The embedding 610 may include pixel descriptors, depth information, and other information.

[0127] The task module 614 may receive one or more task graphs 616. The task module 614 provides a response to a request from the keyframe matcher 612. The keyframe matcher 612 matches a task to a matching keyframe. The task may be determined from the task graph 616.

[0128] The task module 614 may also send action requests to a behavior module 618. The behavior module 618 provides action status to the task module 614. In addition, the behavior module 618 may request information about matched keyframes and corresponding tasks from the keyframe matcher 612. The keyframe matcher 612 provides information about matched keyframes and corresponding tasks to the behavior module 618. The behavior module 618 may receive voxels from a voxel map 626.

[0129] In one configuration, the motion module 618 receives a motion plan from the motion planner 620 in response to a motion plan request. The motion module 618 also receives part status from the part control module 622. The motion module 618 sends part commands to the part control module 622 in response to receiving the part status. Finally, the part control module 622 receives joint status from the joint control module 624. The part control module 622 sends joint commands to the joint control module 624 in response to receiving the joint status.

[0130] 7 illustrates a method 700 for visualizing data generated by a robotic device according to an embodiment of the present disclosure. In an optional configuration, at block 702, the robotic device is trained to determine an intended path based on one or more previously traversed paths, either autonomously and / or based on human input.

[0131] At block 704, the control center displays the intended path of the robotic device in an environment. The environment may be remote from the control center. In addition, the environment may be different or the same as the environment used to train the robotic device. The intended path is generated by the robotic device and transmitted to the control center.

[0132] At block 706, the control center displays a first region in the environment identified as drivable for the robotic device. The first region may be displayed superimposed on the intended path. Additionally, the intended path is within the first region. The first region is generated by the robotic device and transmitted to the control center.

[0133] At block 708, the control center receives an input to identify a second region in the environment as drivable. Additionally or alternatively, the input may mark the region as non-drivable, i.e., the first region determined by the robotic device may be adjusted by the input. The input may be provided by a human operator via an input device such as a touch screen display, a mouse, or a keyboard. The input may include marking pixels in an image of the environment as drivable or non-drivable.

[0134] At block 710, the control center transmits the second area to the robotic device. In an optional configuration, in response to receiving the second area, at block 712, the robotic device determines that it can navigate a third area. The third area may be in a different environment. The weights of the SVM are updated based on the second area (e.g., information edited by a user). The SVM may be referred to as a supervised learning model.

[0135] In an optional arrangement, the control center displays a comparison (see FIGS. 3A and 3B) between the current image and the keyframe images at block 714. The comparison may include visual markings identifying matching pixels in the current image and the keyframe images. In yet another optional arrangement, the control center displays (see FIGS. 3A and 3B) the confidence of the match between the current image and each keyframe image in the set of keyframes at block 716.

[0136] The various operations of the methods described above may be performed by any suitable means capable of performing the corresponding functions. The means may include various hardware and / or software components and / or modules, including, but not limited to, circuits, application specific integrated circuits (ASICs), or processors. Where there are operations illustrated in figures, these operations may have corresponding functional components generally similarly numbered.

[0137] As used herein, "determining" includes a wide variety of actions. For example, "determining" can include calculating, computing, processing, deriving, investigating, searching (e.g., searching in a table, database, or other structure), ascertaining, and the like. In addition, "determining" can include receiving (e.g., receiving information), accessing (e.g., accessing data in a memory), and the like. Further, "determining" can include resolving, selecting, choosing, establishing, and the like.

[0138] As used herein, the phrase "at least one of" refers to any combination of items from a list of items, including a single item. For example, "at least one of a, b, or c" is intended to include a, b, c, ab, ac, bc, abc.

[0139] The various example logic blocks, modules, and circuits described in connection with this disclosure may be implemented or performed by a processor configured in accordance with this disclosure, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a Field Programmable Gate Array (FPGA) signal or other programmable logic device (PLD), discrete gate or transistor logic, discrete hardware components, or any combination of the above designed to perform the functions described herein. A processor may be a microprocessor, controller, microcontroller, or state machine configured as described herein. A processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors in combination with a DSP core, or any other special configuration described herein.

[0140] The steps or algorithms of the methods described in connection with the present disclosure may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. The software modules may reside in a storage device or machine-readable medium, including a random access memory (RAM), a read-only memory (ROM), a flash memory, an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a register, a hard disk, a removable disk, a CD-ROM or other optical disk storage, a magnetic disk storage or other magnetic storage, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. A software module may comprise a single instruction, or many instructions, and may be distributed across several different code segments, among different programs, and across several storage media. A storage medium may be coupled to the processor such that the processor can write information to and read information from the storage medium. Alternatively, the storage medium may be integral to the processor.

[0141] The methods disclosed herein include one or more steps or actions for achieving the disclosed method. The method steps and / or actions may be interchanged with one another without departing from the scope of the claims. In other words, unless a specific order of steps or actions is specified, the order and / or application of specific steps and / or actions may be modified without departing from the scope of the claims.

[0142] The described functionality may be implemented by hardware, software, firmware, or any combination thereof. When implemented in hardware, an example hardware configuration may include a processing system in an apparatus. The processing system may be implemented using a bus architecture. The bus may include any number of interconnected buses and bridges depending on the particular application and overall design constraints of the processing system. The bus may connect various circuits including a processor, a machine-readable medium, and a bus interface. The bus interface may be used to connect, among other things, a network adapter to the processing system via the bus. The network adapter may be used to implement signal processing functions. In certain aspects, a user interface (e.g., a keypad, a display, a mouse, a joystick, etc.) may also be connected to the bus. The bus may also connect various other circuits such as timing sources, peripherals, voltage control, power management circuits, etc., known to those skilled in the art and therefore not described further.

[0143] The processor may be responsible for processing, including managing the bus and executing software stored on a machine-readable medium. Software shall be understood to mean instructions, data, or any combination thereof, whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise.

[0144] In a hardware implementation, the machine-readable medium may be part of a processing system separate from the processor. However, as one skilled in the art will readily appreciate, the machine-readable medium, or any portion thereof, may be external to the processing system. For example, the machine-readable medium may include a communication line, a carrier wave modulated by data, and / or a computer product separate from the device, all of which may be accessed by the processor via a bus interface. Alternatively, or in addition, the machine-readable medium, or any portion thereof, may be integrated into the processor, such as in the case of caches and / or special register files. Although the various components discussed have been described as having special locations, such as local components, they may be configured in various ways, such as specific components configured as part of a distributed computing system.

[0145] The processing system may be comprised of one or more microprocessors providing processor functionality, and external memory providing at least a portion of the machine-readable medium, all of which may be coupled through support circuitry by an external bus architecture. Alternatively, the processing system may comprise one or more neuromorphic processors to implement the neuron models and models of neural systems described herein. As another alternative, the processing system may be implemented as an application specific integrated circuit (ASIC) having a processor, bus interface, user interface, support circuitry, and at least a portion of the machine-readable medium integrated into a single chip, or as one or more Field Programmable Gate Arrays (FPGAs), programmable logic devices (PLDs), controllers, state machines, gate logic, discrete hardware components, or other suitable circuitry, or any combination of circuitry capable of performing the various functions described within this disclosure. Those skilled in the art will recognize how best to implement the described processing system functionality depending on the particular application and the overall design constraints imposed on the overall system.

[0146] The machine-readable medium may comprise a number of software modules. The software modules may include a transmitting module and a receiving module. Each software module may reside in a single storage device or may be distributed across multiple storage devices. For example, a software module may be loaded from a hard drive into RAM when a triggering event occurs. During execution of a software module, a processor may load some instructions into a cache to speed up access. One or more cache lines may then be loaded into a special purpose register file for execution by the processor. With reference to the following functions of a software module, it will be understood that the functions are performed by a processor upon execution of instructions by the software module. It should further be understood that aspects of the present disclosure improve the functionality of a processor, computer, machine, or other system implementing such aspects.

[0147] If implemented in software, the functions may be stored or transferred as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage devices and communication media, including any storage device that facilitates transfer of a computer program from one place to another. In addition, any connection is properly referred to as a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared (IR), radio, and microwave, the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of media. As used herein, disk and disc include compact discs (CDs), laser discs, optical discs, digital versatile discs (DVDs), floppy discs, and Blu-ray discs, where a disk typically reproduces data magnetically and a disc reproduces data optically using a laser. Thus, in some aspects computer readable medium may comprise non-transitory computer readable medium (e.g., tangible media). In addition, in other aspects computer readable medium may comprise transitory computer readable medium (e.g., a signal). Combinations of the above should be included within the scope of computer readable media.

[0148] Accordingly, certain aspects may comprise a computer program product for performing the operations presented herein. For example, such a computer program product may comprise a computer-readable medium having instructions stored thereon (and / or encoded therein), the instructions being executable by one or more processors to perform the operations described herein. In certain aspects, the computer program product may include packaging materials.

[0149] Further, it should be understood that modules and / or other suitable means for performing the methods and techniques described herein can be downloaded and / or acquired by the user terminal and / or base station as needed. For example, such devices can be connected to a server to facilitate the transfer of means for performing the methods described herein. Alternatively, the various methods described herein can be provided via a storage means, such that the user terminal and / or base station can acquire the various methods by connecting the storage means to the device or providing the storage means to the device. Moreover, any other technique for providing the methods and techniques described herein to the device can be used.

[0150] It is to be understood that the claims are not limited to the precise configuration and components illustrated above. Various modifications, changes and variations can be made in the arrangement, operation and details of the methods and apparatus described above without departing from the scope of the claims. The invention disclosed in this specification includes the following aspects. [Aspect 1] 1. A method for visualizing data generated by a robotic device, comprising: displaying an intended path of the robotic device in an environment; displaying a first area in the environment identified as drivable by the robotic device; receiving an input to identify a second area in the environment as drivable; transmitting the second region to the robotic device; A method comprising: [Aspect 2] 2. The method of embodiment 1, further comprising training the robotic device to determine the intended path based on previously traveled paths. [Aspect 3] 2. The method of embodiment 1, further comprising displaying the first region superimposed on the intended route. Aspect 4 2. The method of claim 1, wherein the input includes marking pixels corresponding to the second region. Aspect 5 2. The method of claim 1, wherein the robotic device determines a third area as drivable based on the received second area. Aspect 6 2. The method of embodiment 1, further comprising displaying a comparison between the current image and a keyframe image, the comparison having visible marks identifying matching pixels between the current image and the keyframe image. Aspect 7 2. The method of embodiment 1, further comprising displaying a confidence of a match between the current image and each keyframe in the set of keyframes. Aspect 8 2. The method of claim 1, wherein the intended path and the first region are displayed at a location remote from the robotic device. Aspect 9 Memory, at least one processor coupled to the memory, the at least one processor comprising: Displaying the intended path of the robot device in the environment; displaying a first area in the environment identified as drivable by the robotic device; receiving an input to identify a second region in the environment as drivable; transmitting the second region to the robotic device; An apparatus for visualizing data generated by a robotic device, the apparatus being configured to: Aspect 10 10. The apparatus of embodiment 9, wherein the at least one processor is further configured to train the robotic device to determine the intended path based on previously traveled paths. Aspect 11 10. The apparatus of claim 9, wherein the at least one processor is further configured to display the first region superimposed on the intended route. Aspect 12 10. The apparatus of claim 9, wherein the input includes marking pixels corresponding to the second region. Aspect 13 10. The apparatus of claim 9, wherein the robotic device determines a third area as drivable based on the received second area. Aspect 14 The apparatus of claim 9, wherein the at least one processor is further configured to display a comparison between the current image and the keyframe image, the comparison including a visualized mark identifying matching pixels between the current image and the keyframe image. Aspect 15 10. The apparatus of embodiment 9, wherein the at least one processor is further configured to display a matching confidence between the current image and each key frame in the set of key frames. Aspect 16 10. The apparatus of embodiment 9, further comprising: displaying the intended path and the first region at a location remote from the robotic device. Aspect 17 A non-transitory computer-readable medium having recorded thereon program code for visualizing data generated by a robotic device, the non-transitory computer-readable medium comprising: The program code is executed by a processor, program code for displaying an intended path of the robotic device in an environment; program code for displaying a first area in the environment identified as drivable by the robotic device; program code for receiving an input to identify a second region in the environment as operable; program code for transmitting the second region to the robotic device; A non-transitory computer readable medium comprising: Aspect 18 20. The non-transitory computer-readable medium of claim 17, wherein the program code further comprises program code for training the robotic device to determine the intended path based on previously traveled paths. Aspect 19 20. The non-transitory computer-readable medium of claim 17, wherein the program code further comprises program code for displaying the first region superimposed on the intended path. Aspect 20 20. The non-transitory computer-readable medium of embodiment 17, wherein the input includes marking pixels corresponding to the second region.

Claims

1. 1. A method for visualizing data generated by a robotic device, comprising: displaying, on a display of a user device, in a current environment at a first location, a first drivable area identified by the robotic device based on a neural network associated with learning a drivable area for the robotic device in a training environment at a second location different from the first location, the drivable area being learned based on sensor data and depth data associated with the robotic device navigating the drivable area in the training environment; displaying on the display an intended path of the robotic device within the first drivable region, the intended path of the robotic device being determined based on the neural network learning the drivable region; Receiving an input at the user device identifying a second drivable area in the current environment; transmitting a message to the robotic device indicating the second drivable area, and updating the neural network in response to the transmission of the message; A method comprising:

2. The method of claim 1 , further comprising training the robotic device to determine the intended path based on one or more previously traveled paths within a drivable area of ​​the training environment.

3. The method of claim 1 , further comprising displaying the first drivable area overlaid on the intended route.

4. The method of claim 1 , wherein the input includes marking pixels corresponding to the second drivable area.

5. The method of claim 1 , wherein the robotic device determines a third drivable area as drivable based on receiving the second drivable area.

6. The method of claim 1 , further comprising displaying a comparison of the current image and a keyframe image, the comparison having visible marks identifying matching pixels between the current image and the keyframe image.

7. The method of claim 1 , further comprising displaying a confidence of a match between the current image and each keyframe in the set of keyframes.

8. The method of claim 1 , further comprising displaying the intended path and the first drivable area at a location remote from the robotic device.

9. A device for visualizing data generated by a robot device, comprising: A processor; a memory coupled to the processor and storing instructions that, when executed by the processor, cause the device to: displaying, in a current environment at a first location, a first drivable area identified by the robotic device based on a neural network associated with learning a drivable area for the robotic device in a training environment at a second location different from the first location, the drivable area being learned based on sensor data and depth data associated with the robotic device navigating the drivable area in the training environment; displaying an intended path of the robotic device within the first drivable region, the intended path of the robotic device being determined based on the neural network learning the drivable region; receiving an input to identify a second drivable area in the current environment; transmitting a message indicating the second drivable area to the robotic device, and updating the neural network in response to the transmission of the message; a memory storing instructions operable to: An apparatus comprising:

10. 10. The apparatus of claim 9, wherein execution of the instructions further causes the device to train the robotic device to determine the intended path based on one or more paths previously traveled within a drivable area of ​​the training environment.

11. 10. The apparatus of claim 9, wherein execution of the instructions further causes the apparatus to display the first drivable area superimposed on the intended route.

12. The apparatus of claim 9 , wherein the input includes marking pixels corresponding to the second drivable area.

13. The apparatus of claim 9 , wherein the robotic device determines a third drivable area as drivable based on receiving the second drivable area.

14. 10. The apparatus of claim 9, wherein the processor is further configured to display a comparison of the current image and the keyframe image, the comparison including a visible mark identifying matching pixels between the current image and the keyframe image.

15. The apparatus of claim 9 , wherein the processor is further configured to display a matching confidence between the current image and each keyframe in the set of keyframes.

16. The device of claim 9 , further comprising: displaying the intended path and the first drivable area at a location remote from the robotic device.

17. A non-transitory computer-readable medium having recorded thereon program code for visualizing data generated by a robotic device, the non-transitory computer-readable medium comprising: The program code is executed by a processor, and program code for displaying, on a display of a user device, a first drivable area identified by the robotic device in a current environment at a first location based on a neural network associated with learning a drivable area for the robotic device in a training environment at a second location different from the first location, the drivable area being learned based on sensor data and depth data associated with the robotic device navigating the drivable area in the training environment; a program code for displaying, on the display, an intended path of the robotic device within the first drivable region, the intended path of the robotic device being determined based on the neural network learning the drivable region; program code for receiving, at the user device, an input for identifying a second drivable area in the current environment; program code for transmitting a message to the robotic device indicative of the second drivable area and causing the neural network to be updated in response to transmitting the message; A non-transitory computer readable medium comprising:

18. 20. The non-transitory computer readable medium of claim 17, wherein the program code further comprises program code for training the robotic device to determine the intended path based on one or more previously traveled paths within a drivable area of ​​the training environment.

19. 20. The non-transitory computer readable medium of claim 17, wherein the program code further comprises program code for displaying the first drivable area overlaid on the intended route.

20. 20. The non-transitory computer readable medium of claim 17, wherein the input includes marking pixels corresponding to the second drivable area.

Citation Information

Patent Citations

  • Unmanned mobile body system

    JP2010152835A

  • Device controller and self-propelled electronic device

    JP2014085829A

  • Moving body monitoring system

    JP2017033399A

  • Autonomous travel body

    JP2019106060A

  • Mobile cleaning robot artificial intelligence for situational awareness

    JP2019121364A