公开(公告)号：US11605021B1

公开(公告)日：2023-03-14

申请号：US16588245

申请日：2019-09-30

Applicant: Amazon Technologies, Inc.

Inventor： Vineet Khare ,  Saurabh Gupta ,  Yijie Zhuang ,  Bharathan Balaji ,  Runfei Luo ,  Siddhartha Agarwal

IPC: G06N5/04 ,  G06N20/00 ,  G06N20/20 ,  G06N5/046

Abstract: Techniques for iterative model training and deployment for automated learning systems are described. A method of iterative model training and deployment for automated learning systems comprises generating training data based on inference data, provided by a first version of a model hosted at an endpoint of a machine learning service, and feedback data, received from a client application, using an identifier associated with the inference data and the feedback data, generating a second version of the model using the training data, and deploying the model to the endpoint of the machine learning service.

公开(公告)号：US11480968B1

公开(公告)日：2022-10-25

申请号：US16116285

申请日：2018-08-29

Applicant: AMAZON TECHNOLOGIES, INC.

Inventor： Amin Hani Atrash ,  Saurabh Gupta ,  Sandeep Samdaria ,  Raumi Sidki ,  Xiaowen Mao ,  Morelle Arian

IPC: G05D1/00 ,  G05D1/02

Abstract: A robot moves about an environment and may interact with a user. A waypoint specifies where the robot is to move to with respect to the user while a proxemic cost map is used to plan the path to the waypoint. User input or preferences may be used to modify the waypoint or the proxemic cost map. The waypoint may specify a particular distance and bearing with respect to the user. The proxemic cost map may be oriented with respect to the user and specifies costs for particular areas. For example, an area immediately behind the user may have a very high cost while an area in front of the user may have a low cost. Based on the waypoint and the proxemic cost map, a path is selected and the robot moves along that path, avoiding the high cost areas in favor of the low cost areas if possible.

公开(公告)号：US11256261B1

公开(公告)日：2022-02-22

申请号：US16162133

申请日：2018-10-16

Applicant: AMAZON TECHNOLOGIES, INC.

Inventor： Shi Bai ,  Saurabh Gupta

IPC: G05D1/02 ,  G01C21/34

Abstract: A system determines one or more constraint locations that are present in an environment. A constraint location is a location in the environment through which a user, pet, or moving device is deemed likely to pass due to one or more physical constraints such as walls, furniture, and so forth. For example, a constraint location may be located at a midpoint of a doorway, or where a corridor narrows. Movement of an autonomous mobile device in an environment takes these constraint locations into consideration. In one implementation the autonomous mobile device is prevented from stopping within a threshold distance of a constraint location to avoid blocking movement of others.

公开(公告)号：US20240272953A1

公开(公告)日：2024-08-15

申请号：US18642668

申请日：2024-04-22

Applicant: Amazon Technologies, Inc.

Inventor： Ramyanshu Datta ,  Ishaaq Chandy ,  Arvind Sowmyan ,  Wei You ,  Kunal Mehrotra ,  Kohen Berith Chia ,  Andrea Olgiati ,  Lakshmi Naarayanan Ramakrishnan ,  Saurabh Gupta

IPC: G06F9/50

CPC classification number: G06F9/5038 ,  G06F9/5022 ,  G06F9/5055

Abstract: A post-task-completion retention period for which a computing resource is to be retained, without de-activating the resource, on behalf of a set of requesters of machine learning tasks is determined at a machine learning service. A first task, identified at the service prior to expiration of the retention period at a first computing resource at which a second task has completed, is initiated at the first computing resource. In response to obtaining an indication of a third task and determining that a threshold criterion associated with the retention period satisfies a criterion, the third task is initiated at an additional computing resource. The additional computing resource is de-activated after the third task completes, without waiting for the retention period to expire.

公开(公告)号：US11837229B1

公开(公告)日：2023-12-05

申请号：US17363387

申请日：2021-06-30

Applicant: Amazon Technologies, Inc.

Inventor： Xing Fan ,  Saurabh Gupta ,  Chenlei Guo ,  Eunah Cho

IPC: G06F16/33 ,  G10L15/22 ,  G06N5/02 ,  G10L15/14 ,  G06F16/36

CPC classification number: G10L15/22 ,  G06N5/02 ,  G10L15/144 ,  G06F16/3338 ,  G06F16/367 ,  G10L2015/223

Abstract: Techniques for determining and using interaction affinity data are described. Interaction affinity data may indicate a latent affinity between information corresponding to an interaction, such as, intents, entities, device type from which a user input is received, domain, etc. A system may use the interaction affinity data to determine an alternative input representation for a spoken input to cause output of a desired response to the spoken input. The system may also use the interaction affinity data to recommend an action to a user.

公开(公告)号：US11409295B1

公开(公告)日：2022-08-09

申请号：US16119407

申请日：2018-08-31

Applicant: AMAZON TECHNOLOGIES, INC.

Inventor： Sandeep Samdaria ,  Amin Hani Atrash ,  Saurabh Gupta ,  Sven Cremer ,  Raumi Sidki

IPC: G05D1/02 ,  G01C21/00 ,  B60W50/00 ,  G01C21/34

Abstract: A robot uses types of behavior such as approach, follow, avoid, and so forth to move about an environment and interact with a user. An occupancy map provides information about obstacles in the environment. A predicted trajectory of the user is determined that is indicative of expected locations and confidence of those expected locations. The predicted trajectory may be based on the user's movement and the occupancy map. Based on the predicted trajectory and the occupancy map, a target point and a path to the target point is determined. The path may also be based on a proxemic cost map that specifies how regions with respect to the user may be traversed.

公开(公告)号：US11260536B1

公开(公告)日：2022-03-01

申请号：US16552278

申请日：2019-08-27

Applicant: AMAZON TECHNOLOGIES, INC.

Inventor： Yelin Kim ,  Amin Hani Atrash ,  Raumi Nahid Sidki ,  Vikas Deshpande ,  Saurabh Gupta

IPC: G06F17/00 ,  B25J11/00

Abstract: A simulated emotional state of a device is generated and maintained based on previous emotional state and contextual information. A trigger is associated with an effect value. For example, successful completion of a task may have an effect value of +5 while task failure may have an effect value of −6. Upon occurrence of a trigger, an associated effect value of the trigger may be combined with a previously determined effect value to determine a function input value (FIV). The FIV is used as input to a function that provides an emotion value as output. The emotion value may then be used as input to determine a particular action. For example, if the emotion value corresponds to “happy”, the resulting action may be presenting an animation of a smile on a display device. As triggers occur, the function input value is updated, resulting in updates to the emotion value.