PREDICTIVE ROBOTIC CONTROLLER APPARATUS AND METHODS

• US 20160303738A1
• Filed: 04/18/2016
• Published: 10/20/2016
• Est. Priority Date: 06/14/2013
• Status: Active Grant

• Current Assignee: Brain Corporation
• Original Assignee: Brain Corporation
• Inventors: Laurent, Patryk, Passot, Jean-Baptiste, Sinyavskiy, Oleg, Ponulak, Filip, Gabardos, Borja Ibarz, Izhikevich, Eugene
• Granted Patent: US 9,950,426 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: B25J 9/163 learning, adaptive, model based, rule based expert control
• CPC Class Codes: G05B 2219/39271 Ann artificial neural network, ffw-nn, feedforward neural network
• CPC Class Codes: G05B 2219/39289 Adaptive ann controller
• CPC Class Codes: G06N 3/008 based on physical entities controlled by simulated intelligence so as to replicate intelligent life forms, e.g. based on robots replicating pets or humans in their appearance or behaviour
• CPC Class Codes: G06N 3/049 Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs

Abstract

Robotic devices may be trained by a user guiding the robot along target action trajectory using an input signal. A robotic device may comprise an adaptive controller configured to generate control signal based on one or more of the user guidance, sensory input, performance measure, and/or other information. Training may comprise a plurality of trials, wherein for a given context the user and the robot'"'"'s controller may collaborate to develop an association between the context and the target action. Upon developing the association, the adaptive controller may be capable of generating the control signal and/or an action indication prior and/or in lieu of user input. The predictive control functionality attained by the controller may enable autonomous operation of robotic devices obviating a need for continuing user guidance.

Claims

1-22. -22. (canceled)

23. A non-transitory computer-readable storage medium having a plurality of instructions stored thereon, the instructions being executable by a processing apparatus to train a robotic apparatus, the instructions configured to, when executed by the processing apparatus, cause the robotic apparatus to:

perform a training action in a context determined from at least one or more inputs from one or more sensors, the training action performed based on at least a training instruction provided by a user;

perform a first action in the context, the first action determined by at least a user instruction and a first controller instruction indicative at least in part of a first predicted action determined from at least the context and the training action;

compute a learning parameter based on at least a discrepancy between the first action and the training action; and

perform a second action in the context, the second action determined by at least a second controller instruction generated from at least the learning parameter and a second predicted action determined from at least the context, wherein a discrepancy between the second action and the training action is less than the discrepancy between the first action and the training action.

24. The non-transitory computer-readable storage medium of claim 23, wherein the one or more sensors include a camera.

25. The non-transitory computer-readable storage medium of claim 23, wherein the training action, first action, first predicted action, second action, and second predicted action each comprise trajectory navigation.

26. The non-transitory computer-readable storage medium of claim 23, wherein the training action, first action, first predicted action, second action, and second predicted action each comprise manipulations of a motorized operational element of the robotic apparatus.

27. The non-transitory computer-readable storage medium of claim 23, wherein the context comprises an object representation and the training action comprises at least one of an object approach maneuver and an object avoidance maneuver.

28. The non-transitory computer-readable storage medium of claim 23, wherein the discrepancy between the first action and the training action is calculated from a proximity measure between the first action and the training action.

29. The non-transitory computer-readable storage medium of claim 23, wherein computing the learning parameter comprises using a supervised learning process configured based on at least the context and a combination of the first action and the training action.

30. A method of training a robotic apparatus comprising:

performing a training action in a context determined from at least one or more inputs from one or more sensors, the training action performed based on at least a training instruction provided by a user;

performing a first action in the context, the first action determined by at least a user instruction and a first controller instruction indicative at least in part of a first predicted action determined from at least the context and the training action; and

performing a second action in the context, the second action determined by at least a second controller instruction generated from at least the first action, the training action, and a second predicted action determined from at least the context, wherein a discrepancy between the second action and the training action is less than a discrepancy between the first action and the training action.

31. The method of claim 30, further comprising computing a learning parameter based on at least the discrepancy of the first action and the training action, and the second controller instruction is further generated from the learning parameter.

32. The method of claim 31, wherein computing the learning parameter comprises using a supervised learning process configured based on at least the context and a combination of the first action and the training action.

33. The method of claim 30, wherein the training action, first action, first predicted action, second action, and second predicted action each comprise navigating a trajectory.

34. The method of claim 30, wherein the training action, first action, first predicted action, second action, and second predicted action each comprise manipulating a motorized operational element of the robotic apparatus.

35. The method of claim 30, wherein determining the context comprises generating an object representation and the training action comprises at least one of performing an object approach maneuver and performing an object avoidance maneuver.

36. The method of claim 30, further comprising calculating the discrepancy between the first action and the training action from a proximity measure between the first action and the training action.

37. A trainable robotic apparatus comprising:

a sensor configured to generate a sensory input;

an interface configured to receive one or more user instructions;

a motorized operational element configured to perform actions; and

a processor configured to;

determine a context from the sensory input,

provide a training control signal based on at least a training instruction received through the interface, the training control signal causing the motorized operation element to perform a training action in the context,

provide a first control signal based on at least a user instruction received through the interface and a first predicted action determined from at least the context and the training action, the first control signal causing the motorized operation element to perform a first action in the context,

compute a learning parameter based on at least a discrepancy between the first action and the training action, and

provide a second control signal based on at least the learning parameter and a second predicted action based on at least the context, the second control signal causing the motorized operation element to perform a second action in the context, wherein a discrepancy between the second action and the training action is less than the discrepancy between the first action and the training action.

38. The trainable robotic apparatus of claim 37, wherein the sensor includes a camera.

39. The trainable robotic apparatus of claim 37, wherein the training action, first action, first predicted action, second action, and second predicted action each comprise navigating trajectories.

40. The trainable robotic apparatus of claim 37, wherein the context comprises an object representation and the training action comprises at least one of an object approach maneuver or an object avoidance maneuver.

41. The trainable robotic apparatus of claim 37, wherein the discrepancy between the first action and the training action is calculated from a proximity measure between the first action and the training action.

42. The trainable robotic apparatus of claim 37, wherein computing the learning parameter comprises using a supervised learning process configured based on at least the context and a combination of the first action and the training action.