Disney Enterprises Inc

Systems and methods are disclosed for automated video quality assessment using a hybrid neural network architecture. A sequence of video frames is received and partitioned into fragments, which are further subdivided into patches and encoded as tokens. The tokens are processed in parallel by a state space model, configured to extract temporal features, and by a convolutional neural network, configured to extract spatial features. The resulting feature representations are combined to form a unified embedding, which is input to a prediction head to generate local and overall quality scores indicative of the perceptual quality of the video. In some embodiments, frame-level supervision is employed during training by comparing predicted per-frame scores to reference scores, improving accuracy and granularity. The invention enables robust, efficient, and scalable video quality assessment suitable for use with streaming optimization, compression, and quality monitoring systems, and is adaptable to various neural network backbones.

A method for controlling an actuated system includes receiving one or more data points, where each data point included in the one or more data points associates animation data with a different point in time, computing a curve based on the one or more data points, and causing the actuated system to perform a motion based on the curve.

Methods and systems for animated figure keyframing or physical robot keyframing and learning locomotion with high-level objectives are discussed herein. For example, generating motion for an animated figure may include generating a control policy for the animated figure using a reinforcement learning model, wherein the control policy is configured to control a movement of the animated figure to achieve one or more keyframes. Generating motion for the animated figure may further include encoding the control policy onto a processor of the animated figure. In some cases, the control policy for the animated figure may be generated using a multi-input single-output transformer encoder. Generating motion for the animated figure further includes determining one or more target keyframes and generating, using the control policy, the motion for the animated figure based on the one or more target keyframes.

A system and method for controlling and testing at least one remote media player device using a corresponding remote testing device and at least one user device not collocated with the remote testing device includes receiving from a user a selection of an available remote testing device, displaying on the user device a remote-control image or a list of automated tests, receiving from the user a selection of one of a remote-control command or an automated test as the selected remote command, sending the selected remote command to the selected remote testing device, receiving from the selected remote testing device a video output signal indicative of a medial player video output signal from the media player device in response to the selected remote command, and displaying the received video output signal on the user device simultaneously with the displaying of the remote-control image, while simultaneously performing the receiving and the sending of the selected remote commands.

A system includes a hardware processor, a memory storing software code, and a machine learning (ML) model trained to detect an interruption to a conversation. The system detects, during a conversational turn by an artificial intelligence (AI) character in interaction with a human and/or another AI character, sound produced by the human and/or the other AI character, classifies, using the ML model, the sound as an interruption or irrelevant to the interaction. When the sound is irrelevant to the interaction, the conversational turn of the AI character continues. When the sound is an interruption, the system identifies a response strategy for continuing the interaction, and executes the response strategy including at least one of: (i) retention of the conversational turn, (ii) relinquishment of the conversational turn, or (iii) a negotiation, with the human and/or the other AI character, to determine the retention or the relinquishment of the conversational turn.

Techniques are disclosed for denoising videos. In some embodiments, video frames are denoised using a denoising model that includes an encoder-decoder architecture and attention modules. During training of the denoising model, the attention modules learn weightings to upweight certain dimensions of input features to help pixel registration, remove ghosting artifacts, and improve temporal consistency when the frames of a video are being denoised. The denoising model can also be used to train a student denoising model that has a same architecture as, but is smaller and faster than, the denoising model. After training, noisy video frames can be input into the denoising model and/or the student denoising model to generate corresponding denoised video frames.

The present invention sets forth techniques for training a metamodel to mimic the operation of a simulation that calculates engagement values based on a quantity of services, a product type that consumes those services, a service date on which the services are to be consumed, and a time horizon of days prior to and including the service date. For each date in the time horizon, the simulation calculates an engagement value that represents the cumulative percentage of the services that is simulated to have been allocated as of that date. The metamodel infers engagement values for each date in the time horizon based on the same inputs provided to the simulation. The metamodel is iteratively trained based on differences between the simulated and inferred engagement values associated with multiple sets of inputs. The trained metamodel may require much less time and far fewer computing resources compared to the simulation.

The present invention sets forth techniques for generating robot action commands directed to robot locomotion. The techniques include encoding map scans representing terrain and proprioception data representing a current state of a robot. The techniques also include generating point-wise map features based on the encoded map scan, and transmitting the map features to a multi-head attention (MHA) mechanism. The MHA mechanism assigns an attention weight to each of multiple terrain points included in the map features, conditioned on the proprioception data. The MHA mechanism generates a map encoding representing the terrain points and their corresponding attention weights, where higher attention weights are associated with terrain points that may provide suitable footholds for the robot. Based on a concatenation of the map encoding and the proprioception data, a multilayer perceptron (MLP) generates robot action commands, such as joint movement commands, causing the robot to traverse the terrain.

A suspension system may include a plate, a frame extending at least partially around the plate, and a resilient member extending between to connect the plate to the frame. The resilient member may be configured to resiliently deform to allow a movement of the plate relative to the frame. The frame may be associated with a chassis configured to couple a track vehicle to a track. A pair of plates may define a rotation axis for a wheel configured to traverse along the track. Resilient members may extend between the pair of plates and the frame. A suspension part formed as a single monolithic component may include a first plate, a second plate, and a resilient member extending between the first plate and the second plate, the resilient member configured to resiliently deform to allow a movement of the first plate relative to the second plate.

A method for determining a trajectory of an object includes: determining, via a processing element, the object of the moving device; generating, via the processing element, a physics model of a moving device including the object; receiving, via the processing element, a motion profile of the moving device; discretizing, via the processing element, the motion profile in time into a plurality of time steps; simulating, via the processing element, a release of the object at each of the plurality of time steps; determining, via the processing element, a trajectory of the object as a result of the release.