Anderson, Oliver Ben

Implementations enable scalable generation of high-quality, diverse training data for vision-language model(s) (VLM(s)). Processor(s) of a system can configure a dialog between at least a first VLM and a second VLM, cause the dialog to be conducted, and generate training instance(s) based on the dialog. In configuring the dialog, the first VLM is provided by a target image and a first set of instructions for the dialog, and the second VLM is provided with an ordered set of candidate images (e.g., the target image and additional image(s)) and a second set of instructions for the dialog. In causing the dialog to be conducted, the second VLM generates question(s) (e.g., using the second set of instructions) to ask the first VLM in furtherance of identifying the target image, in the ordered set of candidate images, and the first VLM generates response(s) (e.g., using the first set of instructions) to the question(s).

A method includes: generating a set of tasks; determining, by a machine learning model and based on multiple data types from multiple sources, that an overall risk score exceeds a first failure threshold due to a risk score of a task exceeding a second threshold; selecting a replacement task for the task, the selecting including: receiving, replacement candidates, each replacement candidate including a candidate offset potential and one or more candidate failure mechanisms; assigning, by the machine learning model and to each of the replacement candidates, a replacement score for the replacement candidate based on a failure correlation of the replacement candidate with respect to each other sets of the set of tasks; ranking the replacement candidates based on the replacement scores; and selecting, based on the ranking, the replacement task; and generating, an updated set of tasks including the replacement task.

In one example, a method is described. The method includes receiving, from an information presentation device, a sequence of pings. The sequence of pings includes a first ping at a first port number, a second ping at a second port number, and a third ping at a third port number at a streaming meter. The method further includes updating, a system log of the streaming meter, to show the first ping at the first port number, the second ping at the second port number, and the third ping at the third port number; comparing the sequence of pings to a key configuration stored in the streaming meter; and generating, in response to the comparing, a status event. The status event includes device information. The method further includes transmitting, over a network, the status event to a server associated with an audience measurement entity.

The disclosed systems and techniques are directed to identifying and responding to presence of vulnerable road users (VRUs) in driving environments that are at risk of loss of control of their driving trajectories. The techniques include collecting, using a sensing system of an autonomous vehicle, sensing data for an environment of the autonomous vehicle and processing the sensing data by one or more machine learning models to identify a plurality of reference points associated with a VRU in the environment. The techniques further include identifying one or more height differentials for the plurality of reference points, determining that the VRU is at risk of loss of control, based at least on a change of the one or more height differentials, and causing a control system of the autonomous vehicle to perform an avoidance action.

Devices, systems, and methods for isolating an insect pupa from an aqueous solution including a plurality of insect pupae. An isolation device may include a primary channel through which the aqueous solution and insect pupae travel, an outlet, and a secondary channel that intersects the primary channel. The secondary channel may transport a fluid to add it to the primary channel between an inlet and the outlet. When the sealable outlet is closed, an insect pupa located downstream of the intersection of the primary and secondary channels remains stationary while any insect pupae upstream of the intersection are pushed towards the inlet. The systems and methods describe herein may also provide for identifying and sorting insect pupae.

A method for on-device AI during a virtual meeting includes receiving, at a first client device of a first participant of one or more participants of a virtual meeting, a request to perform a task pertaining to the virtual meeting from a second client device of a second participant of the plurality of participants; obtaining, at the first client device, first text data based on first audio data associated with a first audio stream produced by the second client device; performing, using an AI model of the first client device and using the first text data as input to the AI model, the task pertaining to the virtual meeting; and providing, by the first client device, information associated with the performed task pertaining to the virtual meeting to the second client device.

The described aspects and implementations enable reduced use in computational resources by an autonomous vehicle (AV) by predicting third-party collisions for an AV. A method includes obtaining object indications for objects in a driving environment of an AV. An object indication may include a shape definition and one or more predicted future locations of a corresponding object. The method includes projecting each shape definition onto each predicted future location of the one or more predicted future locations of the corresponding object. The method includes determining, based on projected shape definitions of the objects, that the projected shape definitions of at least two objects overlap. The method includes, responsive to the overlap between at least one of the projected shape definitions meeting a collision criterion, modifying the operation of the AV to avoid an area of the overlapped projected shape definitions of the least two objects.

Techniques for identifying a close encounter between an aircraft and an obstacle are disclosed. The technique includes acquiring a video stream of a ground area below the aircraft with a stereovision camera system disposed onboard the aircraft. A depth perception map is generated with a stereovision processing pipeline indicating stereovision depth estimates of first image pixels from the video stream. An optical flow map is generated with an optical flow processing pipeline indicating optical flow depth estimates of second image pixels from the video stream. The depth perception and optical flow maps are compared. An encounter flag indicating that the close encounter between the aircraft and the obstacle occurred is asserted based at least on the comparing.

An example method includes decoding a portion of audio so as to determine, for respective symbol slots of a watermark, current strength values for respective symbols. The method also includes determining, for the respective symbol slots, expected symbols of an expected message. In addition, the method includes determining a current score for the portion of audio using current strength values for the expected symbols and averages of current strength values for unexpected symbols. The method also includes determining an accumulation score for the portion of audio using accumulated strength values for the expected symbols and averages of accumulated strength values for the unexpected symbols. And the method includes, based on a detected divergence between the current score and the accumulation score, reinitializing the accumulated strength values for the expected symbols and the accumulated strength values for the unexpected symbols.

Implementations for robot control are provided. A method involves, based on vision data depicting an environment of a robot and a natural language instruction for the robot, determining an affordance plan for performing a task. The affordance plan comprises a sequence of intermediate representations of the robot in visual space, such as end effector poses. An action input prompt is assembled with data indicative of the vision data, the natural language instruction, and the affordance plan. The action input prompt is processed using one or more generative models to generate action output indicative of one or more actions to be performed by the robot. Subsequently, a robot control signal is generated based on the one or more actions. This provides a spatially precise and dimensionally concise form of guidance for robot manipulation tasks, which can improve performance and generalization.