Offline robot planning with online adaptation

A system comprising:

• one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:
• maintaining a library of pre-generated motion plans for performing a particular task in a workcell, wherein each pre-generated motion plan is associated with a different discrete label that represents a continuous interval in the workcell, wherein each pre-generated motion plan comprises a plurality of waypoints that collectively define a path to be taken by a robot to perform the particular task and one or more actions to be performed by the robot when the robot is at each of one or more of the plurality of waypoints;
• obtaining one or more present observations of a robot in the workcell, each present observation having a respective observation value;
• classifying one or more of the observation values, including assigning a first discrete label of a predetermined number of possible discrete labels to a continuous observation value in the observation values, each possible discrete label representing a respective interval within a possible range for the continuous observation value;
• selecting, from the library of pre-generated motion plans associated with the different discrete labels, a pre-generated candidate motion plan associated with the first discrete label;
• adapting waypoints of the selected pre-generated candidate motion plan according to the one or more present observations of the robot in the workcell to generate a final motion plan to be executed by the robot;
• obtaining one or more updated observations of a robot in a workcell, each updated observation having a respective updated observation value;
• classifying one or more of the updated observation values, including assigning a second discrete label of the predetermined number of possible discrete labels to a continuous observation value in the updated observation values;
• determining whether the selected pre-generated candidate motion plan matches the second discrete label assigned to the continuous observation value in the updated observation values;
• if the selected pre-generated candidate motion plan matches the second discrete label assigned to the continuous observation value in the updated observation values, adapting the selected pre-generated candidate motion plan according to the one or more updated observations of the robot in the workcell to generate an updated final motion plan to be executed by the robot, and
• if the selected pre-generated candidate motion plan does not match the second discrete label assigned to the continuous observation value in the updated observation values, selecting another pre-generated candidate motion plan from the library of pre-generated motion plans, and adapting the other pre-generated candidate motion plan according to the one or more updated observations of the robot in the workcell to generate an updated final motion plan to be executed by the robot.

One or more non-transitory computer storage media encoded with computer program instructions that when executed by one or more computers cause the one or more computers to perform operations comprising:

• maintaining a library of pre-generated motion plans for performing a particular task in a workcell, wherein each pre-generated motion plan is associated with a different discrete label that represents a continuous interval in the workcell, wherein each pre-generated motion plan comprises a plurality of waypoints that collectively define a path to be taken by a robot to perform the particular task and one or more actions to be performed by the robot when the robot is at each of one or more of the plurality of waypoints;
• obtaining one or more present observations of a robot in the workcell, each present observation having a respective observation value;
• classifying one or more of the observation values, including assigning a first discrete label of a predetermined number of possible discrete labels to a continuous observation value in the observation values, each possible discrete label representing a respective interval within a possible range for the continuous observation value;
• selecting, from the library of pre-generated motion plans associated with the different discrete labels, a pre-generated candidate motion plan associated with the first discrete label;
• adapting waypoints of the selected pre-generated candidate motion plan according to the one or more present observations of the robot in the workcell to generate a final motion plan to be executed by the robot;
• obtaining one or more updated observations of a robot in a workcell, each updated observation having a respective updated observation value;
• classifying one or more of the updated observation values, including assigning a second discrete label of the predetermined number of possible discrete labels to a continuous observation value in the updated observation values;
• determining whether the selected pre-generated candidate motion plan matches the second discrete label assigned to the continuous observation value in the updated observation values;
• if the selected pre-generated candidate motion plan matches the second discrete label assigned to the continuous observation value in the updated observation values, adapting the selected pre-generated candidate motion plan according to the one or more updated observations of the robot in the workcell to generate an updated final motion plan to be executed by the robot, and
• if the selected pre-generated candidate motion plan does not match the second discrete label assigned to the continuous observation value in the updated observation values, selecting another pre-generated candidate motion plan from the library of pre-generated motion plans, and adapting the other pre-generated candidate motion plan according to the one or more updated observations of the robot in the workcell to generate an updated final motion plan to be executed by the robot.