DETERMINING AND UTILIZING CORRECTIONS TO ROBOT ACTIONS
Methods, apparatus, and computer-readable media for determining and utilizing human corrections to robot actions. In some implementations, in response to determining a human correction of a robot action, a correction instance is generated that includes sensor data, captured by one or more sensors of the robot, that is relevant to the corrected action. The correction instance can further include determined incorrect parameter(s) utilized in performing the robot action and/or correction information that is based on the human correction. The correction instance can be utilized to generate training example(s) for training one or model(s), such as neural network model(s), corresponding to those used in determining the incorrect parameter(s).
1 . A method implemented by one or more processors, the method comprising:
receiving an instance of natural language user interface input provided by a user;
processing the instance of natural language user interface input to determine that the instance of natural language user interface input corresponds to a sequence of robotic actions;
in response to determining that the natural language user interface input corresponds to the sequence of robotic actions:
providing control commands to one or more actuators of a robot to autonomously perform a subset of the robotic actions of the sequence, wherein each of the robotic actions comprises autonomously moving one or more corresponding components of the robot;
during performance of the subset of the robotic actions of the sequence:
receiving an additional instance of natural language user interface input provided by the user;
processing the additional instance of natural language user interface input to determine that the additional instance of natural language user interface input indicates a correction; and
in response to determining that the additional instance of natural language user interface input indicates the correction:
generating one or more alternative robotic actions, and
providing further control commands to one or more of the actuators of the robot to autonomously perform the one or more alternative robotic actions.
2 . The method of claim 1 , wherein the additional instance of natural language user interface input includes one or more correction details and wherein generating the one or more alternative robotic actions comprises using the one or more correction details in generating the one or more alternative robotic actions.
3 . The method of claim 1 , further comprising:
based on determining that the additional instance of natural language user interface input indicates the correction:
providing user interface output that requests correction details; and
receiving, in response to providing the user interface output, further user interface input that includes the correction details;
wherein generating the one or more alternative robotic actions comprises using the one or more correction details in generating the one or more alternative robotic actions.
4 . The method of claim 1 , further comprising:
during performance of a robotic action of the subset of the robotic actions of the sequence:
rendering, via at least one user interface output device of the robot, natural language user interface output that describes the robotic action being performed.
5 . The method of claim 4 , wherein the rendering comprises visually rendering the natural language user interface output via a display of the at least one user interface output device of the robot.
6 . The method of claim 5 , wherein the rendering further comprises speaking the natural language user interface output via a speaker of the at least one user interface output device of the robot.
7 . The method of claim 4 , wherein the rendering comprises speaking the natural language user interface output via a speaker of the at least one user interface output device of the robot.
8 . The method of claim 1 , further comprising:
in response to determining that the additional instance of natural language user interface input indicates the correction:
generating a training example that is based on the correction.
9 . The method of claim 8 , further comprising:
training one or more neural network models based on the training example.
10 . The method of claim 1 , wherein the robotic actions of the sequence include a picking action directed to an object and a placing action directed to the object.
11 . The method of claim 1 , wherein the instance of natural language user interface input is spoken natural language user interface input received via a microphone of the robot.
12 . A system comprising:
memory storing instructions;
one or more processors configured to execute the instructions to:
receive an instance of natural language user interface input provided by a user;
process the instance of natural language user interface input to determine that the instance of natural language user interface input corresponds to a sequence of robotic actions;
in response to determining that the natural language user interface input corresponds to the sequence of robotic actions:
provide control commands to one or more actuators of a robot to autonomously perform a subset of the robotic actions of the sequence, wherein each of the robotic actions comprises autonomously moving one or more corresponding components of the robot;
during performance of the subset of the robotic actions of the sequence:
receive an additional instance of natural language user interface input provided by the user;
process the additional instance of natural language user interface input to determine that the additional instance of natural language user interface input indicates a correction; and
in response to determining that the additional instance of natural language user interface input indicates the correction:
generate one or more alternative robotic actions, and
provide further control commands to one or more of the actuators of the robot to autonomously perform the one or more alternative robotic actions.
13 . The system of claim 12 , wherein the additional instance of natural language user interface input includes one or more correction details and wherein in generating the one or more alternative robotic actions one or more of the processors are to use the one or more correction details in generating the one or more alternative robotic actions.
14 . The system of claim 12 , wherein one or more of the processors are further configured to execute the instructions to:
based on determining that the additional instance of natural language user interface input indicates the correction:
provide user interface output that requests correction details; and
receive, in response to providing the user interface output, further user interface input that includes the correction details;
wherein in generating the one or more alternative robotic actions one or more of the processors are to use the one or more correction details in generating the one or more alternative robotic actions.
15 . The system of claim 12 , wherein one or more of the processors are further configured to execute the instructions to:
during performance of a robotic action of the subset of the robotic actions of the sequence:
render, via at least one user interface output device of the robot, natural language user interface output that describes the robotic action being performed.
16 . The system of claim 15 , wherein in rendering the natural language user interface output one or more of the processors are to visually render the natural language user interface output via a display of the at least one user interface output device of the robot.
17 . The system of claim 15 , wherein in rendering the natural language user interface output one or more of the processors are to speak the corresponding natural language user interface output via a speaker of the at least one user interface output device of the robot.
18 . The system of claim 12 , wherein one or more of the processors are further configured to execute the instructions to:
in response to determining that the additional instance of natural language user interface input indicates the correction:
generate a training example that is based on the correction.
19 . The system of claim 18 , wherein one or more of the processors are further configured to execute the instructions to:
train one or more neural network models based on the training example.
20 . The system of claim 12 , wherein the robotic actions of the sequence include a picking action directed to an object and a placing action directed to the object.