Google Inc.

Waymo LLC

National Technology & Engineering Solutions of Sandia, LLC

Stc.unm

Aleksandra Faust

Dilek Zeynep Hakkani-Tur

Vadim Furman

Matthieu Devin

Yu-Hsin Chen

Franklin Morley

Pararth Shah

Marek Fiser

Carlos Alberto Fuertes Pascual

Ulrich Rueckert

Izzeddin Gur

Juliana Kew

Nathaniel Fairfield

Quoc V Le

Sergey Levine

Honglak Lee

Jie Tan

Anthony Matt Francis

Sehoon Ha

Timothy J Draelos

Tingnan Zhang

Matthew Gregor Peterson

Daiyi Peng

Lydia Tapia

Rafael Fierro

Xinlei Pan

Brian Andrew Ichter

Hunter A Knox

John Dalton Co-Reyes

Yingjie Miao

Arnold Peter Ruymgaart

Hao-tien Chiang

Methods, computer systems, and apparatus, including computer programs encoded on computer storage media, for generating and searching reinforcement learning algorithms. In some implementations, a computer-implemented system generates a sequence of candidate reinforcement learning algorithms. Each candidate reinforcement learning algorithm in the sequence is configured to receive an input environment state characterizing a state of an environment and to generate an output that specifies an action to be performed by an agent interacting with the environment. For each candidate reinforcement learning algorithm in the sequence, the system performs a performance evaluation for a set of a plurality of training environments. For each training environment, the system adjusts a set of environment-specific parameters of the candidate reinforcement learning algorithm by performing training of the candidate reinforcement learning algorithm to control a corresponding agent in the training environment. The system generates an environment-specific performance metric for the candidate reinforcement learning algorithm that measures a performance of the candidate reinforcement learning algorithm in controlling the corresponding agent in the training environment as a result of the training. After performing training in the set of training environments, the system generates a summary performance metric for the candidate reinforcement learning algorithm by combining the environment-specific performance metrics generated for the set of training environments. After evaluating each of the candidate reinforcement learning algorithms in the sequence, the system selects one or more output reinforcement learning algorithms from the sequence based on the summary performance metrics of the candidate reinforcement learning algorithms.

Reinforcement learning algorithm search

Implementations relate to using deep reinforcement learning to train a model that can be utilized, at each of a plurality of time steps, to determine a corresponding robotic action for completing a robotic task. Implementations additionally or alternatively relate to utilization of such a model in controlling a robot. The robotic action determined at a given time step utilizing such a model can be based on: current sensor data associated with the robot for the given time step, and free-form natural language input provided by a user. The free-form natural language input can direct the robot to accomplish a particular task, optionally with reference to one or more intermediary steps for accomplishing the particular task. For example, the free-form natural language input can direct the robot to navigate to a particular landmark, with reference to one or more intermediary landmarks to be encountered in navigating to the particular landmark.

Controlling a robot based on free-form natural language input

The present disclosure is generally directed to methods, apparatus, and computer-readable media (transitory and non-transitory) for learning to automatically navigate interactive web documents and/or websites. More particularly, various approaches are presented for training various deep Q network (DQN) agents to perform various tasks associated with reinforcement learning, including hierarchical reinforcement learning, in challenging web navigation environments with sparse rewards and large state and action spaces. These agents include a web navigation agent that can use learned value function(s) to automatically navigate through interactive web documents, as well as a training agent, referred to herein as a 'meta-trainer,' that can be trained to generate synthetic training examples. Some approaches described herein may be implemented when expert demonstrations are available. Other approaches described herein may be implemented when expert demonstrations are not available. In either case, dense, potential-based rewards may be used to augment the training.

Automatic navigation of interactive web documents

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

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for implementing an interactive autonomous vehicle agent. One of the methods includes receiving a request to generate an experience tuple for a vehicle in a particular driving context. A predicted environment observation representing a predicted environment of the autonomous vehicle after the candidate action is taken by the autonomous vehicle in an initial environment is generated, including providing an initial environment observation and the candidate action as input to a vehicle behavior model neural network trained to generate predicted environment observations. An immediate quality value is generated from a context-specific quality model that generates immediate quality values that are specific to the particular driving context. An experience tuple comprising the initial environment observation, the candidate action, and the immediate quality value is generated and used as input to a reinforcement learning system for the autonomous vehicle.

Interactive autonomous vehicle agent

Systems and methods are disclosed that include tools that utilize Dynamic Detector Tuning (DDT) software that identifies near-optimal parameter settings for each sensor using a neuro-dynamic programming (reinforcement learning) paradigm. DDT adapts parameter values to the current state of the environment by leveraging cooperation within a neighborhood of sensors. The key metric that guides the dynamic tuning is consistency of each sensor with its nearest neighbors: parameters are automatically adjusted on a per station basis to be more or less sensitive to produce consistent agreement of detections in its neighborhood. The DDT algorithm adapts in near real-time to changing conditions in an attempt to automatically self-tune a signal detector to identify (detect) only signals from events of interest. The disclosed systems and methods reduce the number of missed legitimate detections and the number of false detections, resulting in improved event detection.

Systems, methods and computer program products for self-tuning sensor data processing

Aspects of the disclosure relate to detecting vehicle collisions. In one example, one or more computing devices may receive acceleration data of a vehicle and the expected acceleration data of the vehicle over a period of time. The one or more computing devices may determine a change in the vehicle's acceleration over the period of time, where the change in the vehicle's acceleration over the period of time is the difference between the expected acceleration data and the acceleration data. The one or more computing devices may detect an occurrence when the change in the vehicle's acceleration is greater than a threshold value and assign the occurrence into a collision category. Based on the assigned collision category, the one or more computing devices may perform a responsive action.

Unexpected impulse change collision detector

The present invention provides a vehicle with redundant systems to increase the overall safety of the vehicle. In other aspects, the present invention provides a method for learning control of non-linear motion systems through combined learning of state value and action-value functions.

Redundant component and intelligent computerized control system for multi-rotor VTOL aircraft

Growing community of inventors

Palo Alto, CA, United States of America

Aleksandra Faust