Waymo LLC

Aleksandar Gabrovski

Yang-Hua Chu

Menghui Wang

Xiaoyi Liu

Colin Andrew Braley

James Stout

Wesley Heung

Andrew Hughes Chatham

Michael Buchanan

David Yonchar Margines

Hui Seong Son

Arthur Safira

Brian Choung Choi

Harrison Lee McKenzie Chapter

The disclosure relate to detecting differences indicative of non-determinism in software used to control autonomous vehicles. For instance, a simulation may be run using the software at least two times. The output of the simulations may be tracked. The output of the simulations includes messages may be generated by a decision making module of the software that affect vehicle behavior. A pair of messages generated from each of the at least two times the simulation is run may be compared. Based on the comparison, a difference indicative of non-determinism in the software may be detected.

Non-determinism testing for software for controlling autonomous vehicles

The disclosure relate to determining a realism metric for testing software for operating a vehicle in an autonomous driving mode. For instance, a plurality of versions of a simulation may be run using log data collected by a vehicle operating in an autonomous driving mode. The plurality of versions may be run using the software to control a simulated vehicle, and each of the plurality of versions may have a set of timing requirements different from the set of timing requirements of other of the plurality of versions. Results of the plurality of versions may be compared to the log data. A realism metric defining timing requirements for one or more future simulations may be generated based on the comparison.

Realism metric for testing software for controlling autonomous vehicles

The disclosure relates to running simulations in order to test software used to control a vehicle in an autonomous driving mode. For instance, logged data may be identified for a simulation. The logged data may have been collected by a first vehicle and may identifying an agent that is a road user. The logged data may be analyzed to identify one or more signals of intent of the agent including a logged path of the agent. One or more characteristics may be identified based on the one or more signals. The simulation may be run using the logged data by replacing the agent with an interactive agent having the one or more characteristics. The interactive agent may be capable of responding to actions performed by a simulated vehicle in the simulation using software for controlling a vehicle in an autonomous driving mode.

Extracting agent intent from log data for running log-based simulations for evaluating autonomous vehicle software

Methods, systems, and apparatus for generation and use of surfel maps to plan for occlusions. One of the methods includes receiving a previously-generated surfel map depicting an area in which a vehicle is located, the surfel map comprising a plurality of surfels, each surfel corresponding to a respective different location in the area in which a vehicle is located; receiving, from one or more sensors, sensor data representing the area in which the vehicle is located; determining, based on the sensor data, that the area in which a vehicle is located includes a dynamic object having a changed shape relative to its representation in the surfel map; and generating an updated path for the vehicle to travel that avoids an occlusion by the changed shape of the dynamic object of a line of sight of one or more sensors to an area of interest.

Planning for occlusions with surfel maps

Aspects and implementations of the present disclosure address shortcomings of existing technology by enabling autonomous vehicle simulations based on retro-reflection optical data. The subject matter of this specification can be implemented in, among other things, a method that involves initiating a simulation of an environment of an autonomous driving vehicle, the simulation including a plurality of simulated objects, each having an identification of a material type of the respective object. The method can further involve accessing simulated reflection data based on the plurality of simulated objects and retro-reflectivity data for the material types of the simulated objects, and determining, using an autonomous vehicle control system for the autonomous vehicle, a driving path relative to the simulated objects, the driving path based on the simulated reflection data.

High fidelity simulations for autonomous vehicles based on retro-reflection metrology

The disclosure relates to testing software for operating an autonomous vehicle. In one instance, a simulation may be run using log data collected by a vehicle operating in an autonomous driving mode. The simulation may be run using the software to control a simulated vehicle and by modifying a characteristic of an agent identified in the log data. During the running of the simulation, that a first type of interaction between the first simulated vehicle and the modified agent will occur may be determined. In response to determining that the particular type of interaction will occur, the modified agent may be replaced by a interactive agent that simulates a road user corresponding to the modified agent that is capable of responding to actions performed by simulated vehicles. That the particular type of interaction between the simulated vehicle and the interactive agent has occurred in the simulation may be determined.

Simulations with modified agents for testing autonomous vehicle software

The disclosure relate to detecting performance regressions in software used to control autonomous vehicles. For instance, a simulation may be run using a first version of the software. While the simulation is running, CPU and memory usage by one or more functions of the first version of the software may be sampled. The sampled CPU and memory usage may be compared to CPU or memory usage by each of the one or more functions in a plurality of simulations each running a corresponding second version of the software. Based on the comparisons, an anomaly corresponding to a performance regression in the first version of the software relating to one of the one or more functions may be identified. In response to detecting the anomaly, the first version of the software and the one of the one or more functions may be flagged for review.

Growing community of inventors

Mountain View, CA, United States of America

Aleksandar Gabrovski