Waymo LLC

Beijing Voyager Technology Co., Ltd.

Gm Cruise Holdings LLC

Yun Jiang

Ury Zhilinsky

Zhaoyin Jia

Alok Priyadarshi

Yuzhu Shen

Yimeng Zhang

Congcong Li

Zafar Takhirov

Brian Pierce

Anton Mario Bongio Karrman

Shaogang Wang

Kotung Lin

The present disclosure is directed to combining the strengths of different methods of analyzing collected sensor data to updating a driving pattern of an automated vehicle (AV). This may include combining data from sets of data that track movement of objects over time with instantaneously received sensor data based on a series of steps that include accessing data that tracks the motion of objects in the field of view of a sensing apparatus, receiving current sensor data that includes a component of current or instantaneous object motion, and identifying whether controls of AV should be maintained or changed. When controls of the AV are maintained, an AV may be controlled to stay in driving in a same lane of a roadway at a same velocity. When controls of an AV are changed, changes may include applying, increasing a velocity, or altering the course of the AV.

Unified radar perception architecture

Disclosed herein are neural networks for generating target classifications for an object from a set of input sequences. Each input sequence includes a respective input at each of multiple time steps, and each input sequence corresponds to a different sensing subsystem of multiple sensing subsystems. For each time step in the multiple time steps and for each input sequence in the set of input sequences, a respective feature representation is generated for the input sequence by processing the respective input from the input sequence at the time step using a respective encoder recurrent neural network (RNN) subsystem for the sensing subsystem that corresponds to the input sequence. For each time step in at least a subset of the multiple time steps, the respective feature representations are processed using a classification neural network subsystem to select a respective target classification for the object at the time step.

Recurrent neural network classifier

Systems and methods are provided for the accurate reproduction during simulation of distributed systems, such as vehicle-based processing systems. In a simulation, the execution times of individual tasks are aligned to those of a recorded process. Data consumed during the recorded process is also consumed during the corresponding portion of the simulation, and data generated during the simulation is associated with timing data from data generated during the recorded processes.

Architecture for distributed system simulation timing alignment

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating object detection predictions from a neural network. In some implementations, an input characterizing a first region of an environment is obtained. The input includes a projected laser image generated from a three-dimensional laser sensor reading of the first region, a camera image patch generated from a camera image of the first region, and a feature vector of features characterizing the first region. The input is processed using a high precision object detection neural network to generate a respective object score for each object category in a first set of one or more object categories. Each object score represents a respective likelihood that an object belonging to the object category is located in the first region of the environment.

Object detection neural networks

Systems and methods are provided for the realistic simulation of distributed systems, such as vehicle-based processing systems. A statistical property of message timestamps of a plurality of prior messages from one or more recorded processes is determined. During a simulation, messages that are generated are associated with timing data based on the statistical property from the recorded processes.

Architecture for distributed system simulation with realistic timing

Systems and methods are provided for the configurable simulation of distributed systems, such as vehicle-based processing systems. Simulations may be performed in various modes in which the timing of operations is based on timing observed in real-world instances or other recorded processes. In one simulation mode, the execution times of individual tasks are aligned to those of a recorded process. Data consumed during the recorded process is also consumed during the corresponding portion of the simulation, and data generated during the simulation is associated with timing data from data generated during the recorded processes. In another simulation mode, data generated during the simulation is associated with timing data based on a statistical analysis of timing data from one or more recorded processes. The various simulation modes, and a production mode for real-word instances, may be provided via a unified application programming interface.

Architecture for configurable distributed system simulation timing

Classifying objects using recurrent neural network and classifier neural network subsystems

Embodiments of the disclosure provide systems and methods for determining depth information in a two-dimensional (2D) image. An exemplary system may include a processor and a non-transitory memory storing instructions that, when executed by the processor, cause the system to perform the various operations. The operations may include receiving a first feature map based on the 2D image and applying an extraction network having a convolution operation and a pooling operation to the first feature map to obtain a second feature map. The operations may also include applying a reconstruction network having a deconvolution operation to the second feature map to obtain a depth map.

Growing community of inventors

Sunnyvale, CA, United States of America

Yun Jiang