Method of synthesising training datasets for autonomous robotic control
The present disclosure relates to a computer-implemented method of synthesising a training dataset for training a comparison function of a machine learning service in the detection of a multi-pick event in which a robotic manipulator erroneously picks two or more items concurrently, the method comprising: selecting first image data representative of an image of a plurality of items to be picked by the robotic manipulator; selecting intermediate image data representative of an image of the plurality of items following the removal, by the robotic manipulator, of an item from the plurality of items; selecting final image data representative of an image of the plurality of items following the removal, by the robotic manipulator, of another item from the plurality of items; and, generating the training dataset representative of a notional multi-pick event based on a sequence of unlabelled image data consisting of the first and final image data.
1 . A robotic picking system comprising:
a robotic manipulator;
a robotic control system;
a picking area from which the robotic manipulator picks items;
an imaging system; and,
a machine learning service trained to detect multi-pick events using a training dataset generated by a method of training a machine learning service, the method comprising:
selecting, from a dataset of unlabeled image data, first image data representative of an image of a plurality of items to be picked by the robotic manipulator;
selecting, from the dataset of unlabeled image data, intermediate image data representative of an image of the plurality of items following the removal, by the robotic manipulator, of an item from the plurality of items;
selecting, from the dataset of unlabeled image data, final image data representative of an image of the plurality of items following the removal, by the robotic manipulator, of another item from the plurality of items; and
generating the training dataset representative of a notional multi-pick event based on a sequence of unlabeled image data including the first and final image data,
wherein the robotic control system is configured to:
output a signal suitable for causing the robotic manipulator to pick an item from a plurality of items within the picking area;
before stowing the item, determine, based on a signal outputted from the machine learning service, whether or not the robotic manipulator has erroneously picked two items concurrently;
and, if it is determined that the robotic manipulator has picked two items concurrently, output a signal suitable for causing the robotic manipulator to return the items back in the picking area.
2 . The robotic picking system of claim 1 , wherein the dataset of unlabeled image data is representative of a sequence of images of the plurality of items to be picked by the robotic manipulator captured over a predetermined period of time.
3 . The robotic picking system of claim 1 , wherein the method of training a machine learning service comprises removing from the dataset of unlabeled image data all image data captured during a control cycle of the robotic manipulator in which the robotic manipulator failed to remove an item from the plurality of items.
4 . The robotic picking system of claim 1 , wherein the method of training a machine learning service comprises:
selecting from the dataset of unlabeled image data the intermediate unlabeled image data within a predetermined period following the removal of the item from the plurality of items; and
selecting from the dataset of unlabeled image data the final unlabeled image data within a predetermined period following the removal of the other item from the plurality of items.