Abstract: A machine-learning (ML) architecture for determining three or more outputs, such as a two and/or three-dimensional region of interest, semantic segmentation, direction logits, depth data, and/or instance segmentation associated with an object in an image. The ML architecture may output these outputs at a rate of 30 or more frames per second on consumer grade hardware.

Abstract: Techniques for utilizing microphone or audio data to detect and responding to low velocity impacts to a system such as an autonomous vehicle. In some cases, the system may be equipped with a plurality of microphones that may be used to detect impacts that fail to register on the data captured by the vehicle's inertial measurement units and may go undetected by the vehicle's perception system and sensors. In one specific example, the perception system of the autonomous vehicle may identify a period of time in which a potential low velocity impact may occur. The autonomous vehicle may then utilize the microphone or audio data associated with the period of time to determine if an impact occurred.

Abstract: Techniques for maintaining and synchronizing data is a processing pipeline data between multiple processing units to improve a system latency are described herein. For example, the techniques may include determining, in response to an invocation of vision processing on first vision data stored in a first memory range in a first memory associated with a central processing unit (CPU), that second vision data stored in a second memory range in a second memory associated with a graphic processing unit (GPU) is a modified copy of the first vision data. The second vision data may be obtained using a non-blocking operation from the second memory range. The first vision data stored in the first memory range may be replaced with the second vision data obtained from the second memory range. The vision processing may then be performed using the second vision data stored in the first memory.

Abstract: A vehicle computing system may implement techniques to determine attributes (or intent) of an object detected by a vehicle operating in the environment. The techniques may include determining a set of features with respect to a detected object by a first model and determining, by a second model and based on the set of features, one or more attributes of the object. The first model and the second model may be configured to process at least one image frame to determine the one or more attributes of the object. A model may receive sensor data as an input, and output features and/or an attribute for the detected object. Based on the attribute(s) of the object, a vehicle computing system may control operation of the vehicle.

Abstract: Techniques are discussed for determining a velocity of an object in an environment from a sequence of images (e.g., two or more). A first image of the sequence is transformed to align the object with an image center. Additional images in the sequence are transformed by the same amount to form a sequence of transformed images. Such sequence is input into a machine learned model trained to output a scaled velocity of the object (a relative object velocity (ROV)) according to the transformed coordinate system. The ROV is then converted to the camera coordinate system by applying an inverse of the transformation. Using a depth associated with the object and the ROV of the object in the camera coordinate frame, an actual velocity of the object in the environment is determined relative to the camera.

Abstract: Energy consumption for a vehicle may be reduced based at least in part on an environment characteristic associated with the environment through which the vehicle travels or an operation characteristic associated with operation of the vehicle, thereby increasing an operational time of the vehicle. In some situations, reducing energy consumption may be associated with operation of one or more of a sensor (e.g., turning the sensor off, reducing a frequency or resolution of the sensor, etc.) and/or one or more processors associated with the vehicle (e.g., turning a processor off, reducing a rate of computation, etc.) based at least in part on one or more of the environment characteristic signal or the operation characteristic signal.

Abstract: Techniques for utilizing multiple scales of images as input to machine learning (ML) models are discussed herein. Operations can include providing an image associated with a first scale to a first ML model. An output of the first ML model can include a first bounding box indicative of a first region of the image representing a first object, with the first bounding box falling within a first range of sizes. Next, a scaled image can be generated by scaling the image. The scaled image can be provided to a second ML model, which can output a second bounding box indicative of a second region of the image representing a second object, the second bounding falling within a second range of sizes. Thus, inputting a scaled image to a same ML model (or to different ML models) can result in different detected features in the images.

Abstract: Techniques for detecting attributes and/or gestures associated with pedestrians in an environment are described herein. The techniques may include receiving sensor data associated with a pedestrian in an environment of a vehicle and inputting the sensor data into a machine-learned model that is configured to determine a gesture and/or an attribute of the pedestrian. Based on the input data, an output may be received from the machine-learned model that indicates the gesture and/or the attribute of the pedestrian and the vehicle may be controlled based at least in part on the gesture and/or the attribute of the pedestrian. The techniques may also include training the machine-learned model to detect the attribute and/or the gesture of the pedestrian.

Abstract: Techniques are disclosed for implementing a neural network that outputs embeddings. Furthermore, techniques are disclosed for using sensor data to train a neural network to learn such embeddings. In some examples, the neural network may be trained to learn embeddings for instance segmentation of an object based on an embedding for a bounding box associated with the object being trained to match pixel embeddings for pixels associated with the object. The embeddings may be used for object identification, object matching, object classification, and/or object tracking in various examples.

Abstract: Techniques for training a machine learned (ML) model to determine depth data based on image data are discussed herein. Training can use stereo image data and depth data (e.g., lidar data). A first (e.g., left) image can be input to a ML model, which can output predicted disparity and/or depth data. The predicted disparity data can be used with second image data (e.g., a right image) to reconstruct the first image. Differences between the first and reconstructed images can be used to determine a loss. Losses may include pixel, smoothing, structural similarity, and/or consistency losses. Further, differences between the depth data and the predicted depth data and/or differences between the predicted disparity data and the predicted depth data can be determined, and the ML model can be trained based on the various losses. Thus, the techniques can use self-supervised training and supervised training to train a ML model.

Abstract: Techniques relating to monitoring map consistency are described. In an example, a monitoring component associated with a vehicle can receive sensor data associated with an environment in which the vehicle is positioned. The monitoring component can generate, based at least in part on the sensor data, an estimated map of the environment, wherein the estimated map is encoded with policy information for driving within the environment. The monitoring component can then compare first information associated with a stored map of the environment with second information associated with the estimated map to determine whether the estimated map and the stored map are consistent. Component(s) associated with the vehicle can then control the object based at least in part on results of the comparing.

Abstract: Using detection of low variance regions for improving detection is described. In an example, sensor data can be received from a sensor associated with a vehicle. The sensor data can represent an environment. An indication of a low variance region associated with the sensor data can be determined and an indication of a high variance region associated with the sensor data can be determined based at least in part on the indication of the low variance region. The vehicle can be controlled based on at least one of the sensor data or the indication of the high variance region.

Abstract: Low variance detection training is described herein. In an example, annotated data can be determined based on sensor data received from a sensor associated with a vehicle. The annotated data can comprise an annotated low variance region and/or an annotated high variance region. The sensor data can be input into a model, and the model can determine an output comprising a high variance output and a low variance output. In an example, a difference between the annotated data and the output can be determined and one or more parameters associated with the model can be altered based at least in part on the difference. The model can be transmitted to a vehicle configured to be controlled by another output of the model.

Abstract: Techniques for utilizing multiple scales of images as input to machine learning (ML) models are discussed herein. Operations can include providing an image associated with a first scale to a first ML model. An output of the first ML model can include a first bounding box indicative of a first region of the image representing a first object, with the first bounding box falling within a first range of sizes. Next, a scaled image can be generated by scaling the image. The scaled image can be provided to a second ML model, which can output a second bounding box indicative of a second region of the image representing a second object, the second bounding falling within a second range of sizes. Thus, inputting a scaled image to a same ML model (or to different ML models) can result in different detected features in the images.

Abstract: A vehicle computing system may implement techniques to predict behavior of objects detected by a vehicle operating in the environment. The techniques may include determining a feature with respect to a detected objects (e.g., likelihood that the detected object will impact operation of the vehicle) and/or a location of the vehicle and determining based on the feature a model to use to predict behavior (e.g., estimated states) of proximate objects (e.g., the detected object). The model may be configured to use one or more algorithms, classifiers, and/or computational resources to predict the behavior. Different models may be used to predict behavior of different objects and/or regions in the environment. Each model may receive sensor data as an input, and output predicted behavior for the detected object. Based on the predicted behavior of the object, a vehicle computing system may control operation of the vehicle.

Abstract: Techniques for detecting and responding to an emergency vehicle are discussed. A vehicle computing system may determine that an emergency vehicle based on sensor data, such as audio and visual data. In some examples, the vehicle computing system may determine aggregate actions of objects (e.g., other vehicles yielding) proximate the vehicle based on the sensor data. In such examples, a determination that the emergency vehicle is operating may be based on the actions of the objects. The vehicle computing system may, in turn, identify a location to move out of a path of the emergency vehicle (e.g., yield) and may control the vehicle to the location. The vehicle computing system may determine that the emergency vehicle is no longer relevant to the vehicle and may control the vehicle along a route to a destination. Determining to yield and/or returning to a mission may be confirmed by a remote operator.

Abstract: Techniques for maintaining and synchronizing data is a processing pipeline data between multiple processing units to improve a system latency are described herein. For example, the techniques may include determining, in response to an invocation of vision processing on first vision data stored in a first memory range in a first memory associated with a central processing unit (CPU), that second vision data stored in a second memory range in a second memory associated with a graphic processing unit (GPU) is a modified copy of the first vision data. The second vision data may be obtained using a non-blocking operation from the second memory range. The first vision data stored in the first memory range may be replaced with the second vision data obtained from the second memory range. The vision processing may then be performed using the second vision data stored in the first memory.

Abstract: A vehicle computing system may implement techniques to determine attributes (or intent) of an object detected by a vehicle operating in the environment. The techniques may include determining a set of features with respect to a detected object by a first model and determining, by a second model and based on the set of features, one or more attributes of the object. The first model and the second model may be configured to process at least one image frame to determine the one or more attributes of the object. A model may receive sensor data as an input, and output features and/or an attribute for the detected object. Based on the attribute(s) of the object, a vehicle computing system may control operation of the vehicle.

Abstract: Techniques relating to monitoring map consistency are described. In an example, a monitoring component associated with a vehicle can receive sensor data associated with an environment in which the vehicle is positioned. The monitoring component can generate, based at least in part on the sensor data, an estimated map of the environment, wherein the estimated map is encoded with policy information for driving within the environment. The monitoring component can then compare first information associated with a stored map of the environment with second information associated with the estimated map to determine whether the estimated map and the stored map are consistent. Component(s) associated with the vehicle can then control the object based at least in part on results of the comparing.

Abstract: Techniques for verifying a reliability of map data are discussed herein. In some examples, map data can be used by a vehicle, such as an autonomous vehicle, to traverse an environment. Sensor data (e.g., image data, lidar data, etc.) can be received from a sensor associated with a vehicle and may be used to generate an estimated map and confidence values associated with the estimated map. When the sensor data is image data, images data from multiple perspectives or different time instances may be combined to generate the estimated map. The estimated map may be compared to a stored map or to a proposed vehicle trajectory or corridor to determine a reliability of the stored map data.