Google Inc.

Jie Tan

Thomas Nonn

Gang Pan

Jon Karafin

Sehoon Ha

Tingnan Zhang

Peng Xu

Sergey Levine

Yunfei Bai

Aleksandra Faust

Julio Cesar Hernandez Zaragoza

Tingfang Du

Erwin Coumans

Atil Iscen

Francois Bleibel

Xinlei Pan

Brian Andrew Ichter

Zhenyu Tan

Training and/or using a machine learning model for locomotion control of a robot, where the model is decoupled. In many implementations, the model is decoupled into an open loop component and a feedback component, where a user can provide a desired reference trajectory (e.g., a symmetric sine curve) as input for the open loop component. In additional and/or alternative implementations, the model is decoupled into a pattern generator component and a feedback component, where a user can provide controlled parameter(s) as input for the pattern generator component to generate pattern generator phase data (e.g., an asymmetric sine curve). The neural network model can be used to generate robot control parameters.

Systems and methods for learning agile locomotion for multiped robots

Techniques are disclosed that enable training a plurality of policy networks, each policy network corresponding to a disparate robotic training task, using a mobile robot in a real world workspace. Various implementations include selecting a training task based on comparing a pose of the mobile robot to at least one parameter of a real world training workspace. For example, the training task can be selected based on the position of a landmark, within the workspace, relative to the pose. For instance, the training task can be selected such that the selected training task moves the mobile robot towards the landmark.

System and methods for training robot policies in the real world

A computer-implemented method is disclosed for training one or more machine-learned models. The method can include inputting a first image frame and a second image frame into a feature disentanglement model and receiving, as an output of the machine-learned feature disentanglement model, a state feature and a perspective feature. The method can include inputting the state feature and the perspective feature into a machine-learned decoder model and receiving, as an output of the machine-learned decoder model, the reconstructed image frame. The method can include comparing the reconstructed image frame with a third image frame corresponding with the location and the perspective orientation. The method can include adjusting one or more parameters of the machine-learned feature disentanglement model based on the comparison of the reconstructed image frame and the third image frame.

Systems and methods for training a machine learned model for agent navigation

Depths of one or more objects in a scene may be measured with enhanced accuracy through the use of a light-field camera and a depth sensor. The light-field camera may capture a light-field image of the scene. The depth sensor may capture depth sensor data of the scene. Light-field depth data may be extracted from the light-field image and used, in combination with the sensor depth data, to generate a depth map indicative of distance between the light-field camera and one or more objects in the scene. The depth sensor may be an active depth sensor that transmits electromagnetic energy toward the scene; the electromagnetic energy may be reflected off of the scene and detected by the active depth sensor. The active depth sensor may have a 360&#xb0; field of view; accordingly, one or more mirrors may be used to direct the electromagnetic energy between the active depth sensor and the scene.

Combining light-field data with active depth data for depth map generation

A light-field video stream may be processed to modify the camera pathway from which the light-field video stream is projected. A plurality of target pixels may be selected, in a plurality of key frames of the light-field video stream. The target pixels may be used to generate a camera pathway indicative of motion of the camera during generation of the light-field video stream. The camera pathway may be adjusted to generate an adjusted camera pathway. This may be done, for example, to carry out image stabilization. The light-field video stream may be projected to a viewpoint defined by the adjusted camera pathway to generate a projected video stream with the image stabilization.

4D camera tracking and optical stabilization

A depth-based effect may be applied to a multi-view video stream to generate a modified multi-view video stream. User input may designate a boundary between a foreground region and a background region, at a different depth from the foreground region, of a reference image of the video stream. Based on the user input, a reference mask may be generated to indicate the foreground region and the background region. The reference mask may be used to generate one or more other masks that indicate the foreground and background regions for one or more different images, from different frames and/or different views from the reference image. The reference mask and other mask(s) may be used to apply the effect to the multi-view video stream to generate the modified multi-view video stream.

Multi-view scene segmentation and propagation

Growing community of inventors

Foster City, CA, United States of America

Jie Tan