Machine-learning-enabled tool changer for mushroom crop management system
A robotic mushroom crop manager periodically or continuously receives mushroom bed data corresponding to a mushroom bed including growing mushrooms at a plurality of times. A trained mushroom bed model is used to process the mushroom bed data to generate mushroom bed state vectors respectively characterizing corresponding states of the mushroom bed at the plurality of times. Crop management equipment is controlled to perform a crop management program comprising a sequence of actions to be performed by crop management equipment comprising, for each current action in the sequence of actions, selecting, based on corresponding a current mushroom bed state vector, a selected crop management tool from a plurality of crop management tools. The crop management equipment is controlled to use the selected crop management tool to perform the current action on the mushroom bed.
1. A robotic mushroom crop manager comprising:
at least one processor;
a communications interface; and
at least one computer-readable medium storing instructions executable by the at least one processor to cause the robotic mushroom crop manager:
to periodically or continuously receive mushroom bed data corresponding to a mushroom bed including growing mushrooms at a plurality of times;
to use a trained mushroom bed model to process the mushroom bed data to generate mushroom bed state vectors respectively characterizing corresponding states of the mushroom bed at the plurality of times;
to control crop management equipment to perform a crop management program comprising a sequence of actions to be performed by the crop management equipment comprising, for each current action in the sequence of actions, selecting, based on a corresponding current mushroom bed state vector, a selected crop management tool from a plurality of crop management tools, wherein selecting the selected crop management tool comprises:
accessing a selection library datastore storing predetermined tool selection data associating predetermined optimal ones of the plurality of crop management tools with respectively corresponding combinations of actions in the crop management program and preconfigured ranges of the mushroom bed state vectors;
for the current action, determining whether the selection library datastore stores predetermined tool selection data wherein the current mushroom bed state vector falls within one of the preconfigured ranges of the mushroom bed state vectors; and
when, for the current action, the selection library datastore stores predetermined tool selection data wherein the current mushroom bed state vector falls within one of the preconfigured ranges of the mushroom bed state vectors, using as the selected crop management tool to perform the current action the optimal one of the plurality of crop management tools identified by the predetermined tool selection data; and
to control the crop management equipment to use the selected crop management tool to perform the current action on the mushroom bed.
2. The robotic mushroom crop manager of claim 1 , wherein selecting the selected crop management tool further comprises:
when, for the current action, the selection library datastore is free from predetermined tool selection data wherein the current mushroom bed state vector falls within one of the preconfigured ranges of the mushroom bed state vectors, accessing a robotic mushroom crop manager (RMCM) list datastore storing records corresponding respectively to one or more other like robotic mushroom crop managers, each record storing an identifier and a communication address of the corresponding other like robotic mushroom crop manager, and selecting one of the other like robotic mushroom crop managers based on preconfigured selection parameters;
generating a tool selection request message encoding in association the current action and the current mushroom bed state vector, and using the communications interface to communicate the tool selection request message to the communication address of the selected one of the other like robotic mushroom crop managers;
when, for the current action, the robotic mushroom crop manager receives from the selected other like robotic mushroom manager a tool selection reply message corresponding to the tool selection request message encoding an optimal tool selection, using as the selected crop management tool to perform the current action the optimal one of the plurality of crop management tools identified by the tool selection reply message.
3. The robotic mushroom crop manager of claim 2 , wherein selecting the selected crop management tool further comprises:
when, for the current action, the robotic mushroom crop manager does not receive from the selected other like robotic mushroom manager the tool selection reply message corresponding to the tool selection request message encoding an optimal tool selection within a preconfigured interval, generating a further tool selection request message encoding in association the current action and the current mushroom bed state vector, and using the communications interface to communicate the further tool selection request to a remote crop management tool coordinator; and
when, for the current action, the robotic mushroom crop manager receives from the crop management tool coordinator a tool selection reply message corresponding to the further tool selection request message encoding an optimal tool selection, using as the selected crop management tool to perform the current action the optimal one of the plurality of crop management tools identified by the tool selection reply message.
4. The robotic mushroom crop manager of claim 3 , wherein selecting the selected crop management tool further comprises:
when, for the current action, the robotic mushroom crop manager does not receive from the crop management tool coordinator the tool selection reply message corresponding to the tool selection request message encoding an optimal tool selection within a further preconfigured interval, using a local trained tool selection model to generate based on the current mushroom bed state vector the selected one of the plurality of crop management tools.
5. The robotic mushroom crop manager of claim 4 , wherein the at least one computer-readable medium further stores instructions executable by the at least one processor to cause the robotic mushroom crop manager:
for each current action, to compare a pre-action mushroom bed state vector corresponding to a first time before performance of the current action with a post-action mushroom bed state vector corresponding to a second time after performance of the current action to determine an actual outcome of the current action;
to compare the actual outcome of the current action with a predefined desired outcome of the current action based on predefined parameters of the mushroom bed in accordance with the crop management program; and
when a difference between the actual outcome and the desired outcome falls within a predefined threshold, to store in the selection library datastore predetermined tool selection data associating the selected one of the plurality of crop management tools used to perform the current action on the mushroom bed with the current action and the pre-action mushroom bed state vector.
6. The robotic mushroom crop manager of claim 5 , wherein the at least one computer-readable medium further stores instructions executable by the at least one processor to cause the robotic mushroom crop manager:
when a difference between the actual outcome and the desired outcome falls within the preconfigured threshold, to train the local tool selection model based on the current action on the mushroom bed and the pre-action mushroom bed state vector.
7. The robotic mushroom crop manager of claim 6 , wherein the at least one computer-readable medium further stores instructions executable by the at least one processor to cause the robotic mushroom crop manager:
when a difference between the actual outcome and the desired outcome falls within the preconfigured threshold, to generate an update message encoding the selected one of the plurality of crop management tools used to perform the current action on the mushroom bed and the pre-action mushroom bed state vector; and
to use the communications interface to communicate the update message to the crop management tool coordinator.
8. The robotic mushroom crop manager of claim 1 comprising:
a controller comprising:
the at least one processor;
the communications interface; and
the at least one computer-readable medium; and
the crop management equipment comprising:
a crop management device operable to perform the sequence of actions on the mushroom bed; and
sensors operable to collect the mushroom bed data, the sensors comprising at least one optimal imager operable to collect images of the mushroom bed.
9. A robotic mushroom crop manager comprising:
at least one processor;
a communications interface; and
at least one computer-readable medium storing instructions executable by the at least one processor to cause the robotic mushroom crop manager:
to periodically or continuously receive mushroom bed data corresponding to a mushroom bed including growing mushrooms at a plurality of times;
to use a trained mushroom bed model to process the mushroom bed data to generate mushroom bed state vectors respectively characterizing corresponding states of the mushroom bed at the plurality of times;
to control crop management equipment to perform a crop management program comprising a sequence of actions to be performed by the crop management equipment comprising, for each current action in the sequence of actions, selecting, based on a corresponding current mushroom bed state vector, a selected crop management tool from a plurality of crop management tools; and
to control the crop management equipment to use the selected crop management tool to perform the current action on the mushroom bed;
a controller comprising:
the at least one processor;
the communications interface; and
the at least one computer-readable medium; and
the crop management equipment comprising:
a crop management device operable to perform the sequence of actions on the mushroom bed; and
sensors operable to collect the mushroom bed data, the sensors comprising at least one optimal imager operable to collect images of the mushroom bed;
wherein:
the crop management device comprises a robotic arm comprising an end effector operable to selectively couple with each of the plurality of crop management tools to perform each current action using the corresponding selected crop management tool;
the at least one optimal imager comprises:
a harvesting imager mounted at an underside of the robotic arm adjacent the end effector and operable to collect harvesting imager images of a plan or overhead view of the mushroom bed beneath the end effector;
a trimming imager mounted at or adjacent the robotic arm and oriented and operable to collect trimming imager images of an elevation, side, or perspective view of the end effector and the mushroom bed adjacent the end effector; and
the mushroom bed data comprises the harvesting imager images and the trimming imager images.
10. The robotic mushroom crop manager of claim 9 , wherein:
the current action comprises harvesting a selected one of the mushrooms;
the pre-action mushroom bed state vector is generated based on harvesting imager images of the selected mushroom collected before harvesting of the selected mushroom using the selected one of the crop management tools;
the post-action mushroom bed state vector is generated based on harvesting imager images of the selected mushroom after harvesting of the selected mushroom using the selected one of the crop management tools; and
the desired outcome comprises a maximum degree of damage to a cap or stem of the selected mushroom.
11. The robotic mushroom crop manager of claim 10 , wherein:
the desired outcome further comprises a maximum degree of damage to a cap or stem of other ones of the mushrooms adjacent the selected mushroom in the mushroom bed.
12. The robotic mushroom crop manager of claim 11 , wherein:
the desired outcome further comprises a maximum degree of disturbance to growing medium in the mushroom bed adjacent the selected mushroom.
13. The robotic mushroom crop manager of claim 1 , wherein:
at least some of the crop management tools comprise suction harvesters operable to apply suction to respective heads of at least some of the mushrooms to lift the mushrooms during harvesting.
14. An automated mushroom crop management system comprising:
a plurality of the robotic mushroom crop managers, each of the plurality of the robotic mushroom crop managers comprising:
at least one processor;
a communications interface; and
at least one computer-readable medium storing instructions executable by the at least one processor to cause the robotic mushroom crop manager:
to periodically or continuously receive mushroom bed data corresponding to a mushroom bed including growing mushrooms at a plurality of times;
to use a trained mushroom bed model to process the mushroom bed data to generate mushroom bed state vectors respectively characterizing corresponding states of the mushroom bed at the plurality of times;
to control crop management equipment to perform a crop management program comprising a sequence of actions to be performed by the crop management equipment comprising, for each current action in the sequence of actions, selecting, based on a corresponding current mushroom bed state vector, a selected crop management tool from a plurality of crop management tools, wherein selecting the selected crop management tool comprises:
accessing a selection library datastore storing predetermined tool selection data associating predetermined optimal ones of the plurality of crop management tools with respectively corresponding combinations of actions in the crop management program and preconfigured ranges of the mushroom bed state vectors;
for the current action, determining whether the selection library datastore stores predetermined tool selection data wherein the current mushroom bed state vector falls within one of the preconfigured ranges of the mushroom bed state vectors; and
when, for the current action, the selection library datastore stores predetermined tool selection data wherein the current mushroom bed state vector falls within one of the preconfigured ranges of the mushroom bed state vectors, using as the selected crop management tool to perform the current action the optimal one of the plurality of crop management tools identified by the predetermined tool selection data;
when, for the current action, the selection library datastore is free from predetermined tool selection data wherein the current mushroom bed state vector falls within one of the preconfigured ranges of the mushroom bed state vectors, accessing a robotic mushroom crop manager (RMCM) list datastore storing records corresponding respectively to one or more other like robotic mushroom crop managers, each record storing an identifier and a communication address of the corresponding other like robotic mushroom crop manager, and selecting one of the other like robotic mushroom crop managers based on preconfigured selection parameters;
generating a tool selection request message encoding in association the current action and the current mushroom bed state vector, and using the communications interface to communicate the tool selection request message to the communication address of the selected one of the other like robotic mushroom crop managers;
when, for the current action, the robotic mushroom crop manager receives from the selected other like robotic mushroom manager a tool selection reply message corresponding to the tool selection request message encoding an optimal tool selection, using as the selected crop management tool to perform the current action the optimal one of the plurality of crop management tools identified by the tool selection reply message
when, for the current action, the robotic mushroom crop manager does not receive from the selected other like robotic mushroom manager the tool selection reply message corresponding to the tool selection request message encoding an optimal tool selection within a preconfigured interval, generating a further tool selection request message encoding in association the current action and the current mushroom bed state vector, and using the communications interface to communicate the further tool selection request to a remote crop management tool coordinator; and
when, for the current action, the robotic mushroom crop manager receives from the crop management tool coordinator a tool selection reply message corresponding to the further tool selection request message encoding an optimal tool selection, using as the selected crop management tool to perform the current action the optimal one of the plurality of crop management tools identified by the tool selection reply message
to control the crop management equipment to use the selected crop management tool to perform the current action on the mushroom bed; and
the remote crop management tool coordinator, wherein the crop management tool coordinator comprises:
at least one further processor;
a further communications interface; and
at least one further computer-readable medium storing instructions executable by the at least one further processor to cause the crop management tool coordinator:
to use the further communications interface to receive the further tool selection request from a requesting robotic mushroom crop manager;
to access a further selection library datastore storing predetermined tool selection data associating predetermined optimal ones of the plurality of crop management tools with respectively corresponding actions in the crop management program and preconfigured ranges of the mushroom bed state vectors;
for the current action, to determine whether the further selection library datastore stores predetermined tool selection data wherein the current mushroom bed state vector falls within the preconfigured ranges of the mushroom bed state vectors; and
when, for the current action, the further selection library datastore stores predetermined tool selection data wherein the current mushroom bed state vector falls within the preconfigured ranges of the mushroom bed state vectors, to generate the tool selection reply message encoding the optimal tool selection, and using the further communications interface to communicate the tool selection reply message to the requesting robotic mushroom crop manager.
15. The automated mushroom crop management system of claim 14 , wherein the at least one further computer-readable medium further stores instructions executable by the at least one further processor to cause the crop management tool coordinator:
when, for the current action, the further selection library datastore does not store predetermined tool selection data wherein the current mushroom bed state vector falls within the preconfigured ranges of the mushroom bed state vectors, to use a further local trained tool selection model to generate the optimal tool selection based on the current mushroom bed state vector;
to generate the tool selection reply message encoding the optimal tool selection; and
to use the further communications interface to communicate the tool selection reply message to the requesting robotic mushroom crop manager.
16. The automated mushroom crop management system of claim 15 , wherein the at least one further computer-readable medium further stores instructions executable by the at least one further processor to cause the crop management tool coordinator:
to store in the further selection library datastore predetermined tool selection data associating the selected one of the plurality of crop management tools used to perform the current action on the mushroom bed with the current action and the pre-action mushroom bed state vector.
17. The automated mushroom crop management system of claim 16 , wherein the at least one further computer-readable medium further stores instructions executable by the at least one further processor to cause the crop management tool coordinator:
to train the further local tool selection model based on the current action on the mushroom bed and the pre-action mushroom bed state vector.