Software compensated robotics

1. A robotic system comprising:

a movement generation device;

a robotic arm having a movement in a three-dimensional space responsive to the movement generation device;

an end effector coupled to the robotic arm and having a spatial pose adjustable, at least in part, by the movement of the robotic arm in the three-dimensional space;

one or more sensing devices configured to capture one or more images of a portion of an environment including the end effector; and

a neural network configured to output controls for the movement generation device in response to receiving images captured by the one or more sensing devices, wherein the neural network comprises a first processing block configured to estimate a position of an occluded object in an image of the portion of the environment and generate an image processing output that is representative of a state of a first object within the image and that includes the estimate of the position of the occluded object, wherein the first object is different from the occluded object.

2. The robotic system of claim 1 , wherein the neural network retains a memory of the position of the occluded object within a state of the first processing block.

3. The robotic system of claim 2 , wherein the first processing block comprises recurrent neural network layers having states dependent on previously processed images.

4. The robotic system of claim 1 , wherein the neural network retains a memory of the position of the occluded object within an external memory communicatively coupled to the neural network.

5. The robotic system of claim 1 , wherein the image processing output includes at least one of a representation of a distance between the end effector and the first object and a representation of a distance between two objects within the image.

6. The robotic system of claim 1 , wherein the neural network comprises a second processing block configured to generate command signals for movement of the end effector based on at least one of a goal for the end effector, the image processing output, or a time dependent internal state of the second processing block.

7. The robotic system of claim 6 , wherein the second processing block comprises recurrent neural network layers in which a state of the second processing block is changed through processing of the image processing output such that processing of the image processing output by the second processing block is dependent on prior states of the second processing block.

8. The robotic system of claim 7 , wherein the third processing block comprises recurrent neural network layers in which a state of the third processing block is changed through processing of the image processing output and the command signal such that processing of the image processing output and the command signal is dependent on prior states of the third processing block.

9. The robotic system of claim 6 , wherein the neural network comprises a third processing block configured to generate the control for the movement generation device based on the image processing output and the command signals.

10. The robotic system of claim 1 , wherein the neural network comprises a multiple of the first processing block, and wherein each first processing block is configured to receive images from one of the one or more sensing devices.

11. A method of operating a robotic system, the method comprising:

receiving a first image of a portion of an environment including an end effector, wherein the end effector is coupled to a robotic arm that is responsive to a movement generation device;

generating a first image processing output representative of a state of a first object within the first image, wherein generating the image processing output comprises estimating a position of an occluded object in the first image, wherein the first image processing output includes the estimate of the position of the occluded object, and wherein the first object is different from the occluded object;

generating command signals for movement of the end effector; and

providing an output for control of the movement generation device based on the command signals and the image processing output, wherein providing the output causes the robotic arm to adjust at least one component of a spatial pose of the end effector.

12. The method of claim 11 , wherein generating command signals for movement of the end effector comprises identifying one or more movement goals for the end effector based on the image processing output.

13. The method of claim 11 , further comprising:

receiving a task for the end effector; and

identifying a first movement goal from the task;

wherein generating the command signals for the movement of the end effector comprises selecting a first processing block of a neural network configured to perform the first movement goal, and wherein selecting the first processing block causes the first processing block to generate the command signals.

14. The method of claim 11 , wherein the image processing output is generated by a processing block of a neural network, and wherein generating the image processing output further comprises capturing a memory of the position of the occluded object in a state of the processing block.

15. The method of claim 11 , wherein receiving the first image of the portion of the environment comprises receiving a plurality of images of the portion of the environment; and

wherein the image processing output is representative of states of objects within the plurality of images.

16. The method of claim 11 , wherein the output for control of the movement generation device is provided by a neural network having recurrent nodes, and further comprising adjusting the output for hysteresis in positioning of the end effector based on the first image and the recurrent nodes of the neural network.

17. One or more non-transitory computer-readable storage media storing computer-executable instructions for causing a computing system to perform operations comprising:

receiving an image captured by one or more sensing devices configured to capture images of a portion of an environment including an end effector, wherein the end effector is coupled to a robotic arm;

generating, by a processing block of a neural network, an image processing output representative of a state of a first object within the image, wherein generating the image processing output comprises estimating a position of one or more occluded objects in the image, wherein the image processing output includes the estimate of the position of the one or more occluded objects, wherein the first object is different from the one or more occluded objects;

generating command signals for movement of the end effector based on a movement goal for the end effector; and

providing an output for control of a movement generation device coupled to the robotic arm based on the set of command signals and the set of image processing outputs, wherein the output causes the robotic arm to adjust at least one component of a spatial pose of the end effector.

18. The one or more non-transitory computer-readable storage media of claim 17 , wherein the neural network includes recurrent layers, and wherein generating the image processing output further comprises capturing a memory of the position of the one or more occluded objects in a state of the neural network.

19. The one or more non-transitory computer-readable storage media of claim 17 , wherein the output for control of the movement of the generation device is provided by a neural network having recurrent nodes, and wherein the operations further comprise adjusting the output for hysteresis in positioning of the end effector based on the image and the recurrent nodes of the neural network.