Techniques for deploying trained machine learning models for robot control
AssigneeNVIDIA Corporation
InventorsYashraj Shyam Narang, Ankur Handa, Karl Van Wyk, Dieter Fox, Michael Andres Lin, Fabio Tozeto Ramos
Life of this patent
Estimating the term…
Abstract
One embodiment of a method for controlling a robot includes receiving sensor data indicating a state of the robot, generating an action based on the sensor data and a trained machine learning model, computing a target state of the robot based on the action and a previous target state of the robot, and causing the robot to move based on the target state of the robot.
Claims17
A computer-implemented method for controlling a robot, the method comprising:
- receiving first sensor data indicating a current state of the robot;
- generating a first action based on the first sensor data and a trained machine learning model;
- computing a target state of the robot by adding the first action to a previous target state of the robot, wherein the previous target state of the robot is different from the current state of the robot;
- determining that a difference between the target state of the robot and the current state of the robot is greater than a threshold value;
- generating a clamped difference by clamping the difference to the threshold value;
- regenerating the target state of the robot by adding the clamped difference to the previous target state of the robot;
- and
- causing the robot to move based on the target state of the robot.
One or more non-transitory computer-readable media storing instructions that, when executed by at least one processor, cause the at least one processor to perform the steps of:
- receiving first sensor data indicating a current state of a robot;
- generating a first action based on the first sensor data and a trained machine learning model;
- computing a target state of the robot by adding the first action to a previous target state of the robot, wherein the previous target state of the robot is different from the current state of the robot;
- determining that a difference between the target state of the robot and the current state of the robot is greater than a threshold value;
- generating a clamped difference by clamping the difference to the threshold value;
- regenerating the target state of the robot by adding the clamped difference to the previous target state of the robot;
- and
- causing the robot to move based on the target state of the robot.
A system, comprising:
- one or more memories storing instructions;
- and
- one or more processors that are coupled to the one or more memories and, when executing the instructions, are configured to:
- receive sensor data indicating a current state of a robot,
- generate an action based on the sensor data and a trained machine learning model,
- compute a target state of the robot by adding the action to a previous target state of the robot, wherein the previous target state of the robot is different from the current state of the robot,
- determining that a difference between the target state of the robot and the current state of the robot is greater than a threshold value,
- generating a clamped difference by clamping the difference to the threshold value,
- regenerating the target state of the robot by adding the clamped difference to the previous target state of the robot, and
- cause the robot to move based on the target state of the robot.
Dependencies are inferred from the claim text.