Methods, systems, and apparatus, including computer programs encoded on computer storage media, for distributed robotic demonstration learning. One of the methods includes receiving a skill template to be trained to cause a robot to perform a particular skill having a plurality of subtasks. One or more demonstration subtasks defined by the skill template are identified, wherein each demonstration subtask is an action to be refined using local demonstration data. On online execution system uploads sets of local demonstration data to a cloud-based training system. The cloud-based training system generates respective trained model parameters for each set of local demonstration data. The skill template is executed on the robot using the trained model parameters generated by the cloud-based training system.

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for distributing skill templates for robotic demonstration learning. One of the methods includes receiving, from the user device by a skill template distribution system, a selection of an available skill template. The skill template distribution system provides a skill template, wherein the skill template comprises information representing a state machine of one or more tasks, and wherein the skill template specifies which of the one or more tasks are demonstration subtasks requiring local demonstration data. The skill template distribution system trains a machine learning model for the demonstration subtask using a local demonstration data to generate learned parameter values.

A robot control device includes: a learned model created through learning work data composed of input and output data, the input data including states of a robot and the surroundings where humans operate the robot to perform a series of works, the output data including human operation corresponding to the case or movement of the robot caused thereby; a control data acquisition section that acquires control data by obtaining output data related to human operation or movement from the model, being presumed in response to and in accordance with the input data; a completion rate acquisition section acquiring a completion rate indicating to which progress level in the series of works the output data corresponds; and a certainty factor acquisition section that acquires a certainty factor indicating a probability of the presumption in a case where the model outputs the output data in response to the input data.

The present disclosure relates to a method for creating a robot control program for operating a machine tool, in particular a bending machine, having the steps of generating image material of a machining operation of a workpiece on the machine tool by means of at least one optical sensor; extracting at least one part of the workpiece and/or at least one part of a hand of an operator handling the workpiece from the image material; generating a trajectory and/or a sequence of movement points of at least one part of the workpiece and/or at least one part of a hand of an operator from the extracted image material; and creating a robot control program by reverse transformation of the trajectory and/or the sequence of movement points.

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for distributing skill templates for robotic demonstration learning. One of the methods includes receiving from the user device by a skill template distribution system, a selection of an available skill template. The skill template distribution system provides a skill template, wherein the skill template comprises information representing a state machine of one or more tasks, and wherein the skill template specifies which of the one or more tasks are demonstration subtasks requiring local demonstration data. The skill template distribution system trains a machine learning model for the demonstration subtask using a local demonstration data to generate learned parameter values.

A method for teaching a robot to perform an operation based on human demonstration with images from a camera. The method includes a teaching phase where a 2D or 3D camera detects a human hand grasping and moving a workpiece, and images of the hand and workpiece are analyzed to determine a robot gripper pose and positions which equate to the pose and positions of the hand and corresponding pose and positions of the workpiece. Robot programming commands are then generated from the computed gripper pose and position relative to the workpiece pose and position. In a replay phase, the camera identifies workpiece pose and position, and the programming commands cause the robot to move the gripper to pick, move and place the workpiece as demonstrated. A teleoperation mode is also disclosed, where camera images of a human hand are used to control movement of the robot in real time.

The present disclosure generally relates to performance recreation, and in particular, the recreation of observed human performance using reinforcement learning. In this regard, a first object is identified from a plurality of objects. The manipulation of the first object is tracked from a first position to a second position. A characterization of the manipulation is generated. A policy that controls a mechanical gripper to recreate the manipulation is generated based on an iteratively increasing cumulative award. The mechanical gripper iteratively recreates the manipulation to increase a cumulative award with each recreation.

A sortation system is disclosed that includes a programmable motion device including an end effector, a perception system for recognizing any of the identity, location, and orientation of an object presented in a plurality of objects, a grasp selection system for selecting a grasp location on the object, the grasp location being chosen to provide a secure grasp of the object by the end effector to permit the object to be moved from the plurality of objects to one of a plurality of destination locations, and a motion planning system for providing a motion path for the transport of the object when grasped by the end effector from the plurality of objects to the one of the plurality of destination locations, wherein the motion path is chosen to provide a path from the plurality of objects to the one of the plurality of destination locations.

A sortation system is disclosed that includes a programmable motion device including an end effector, a perception system for recognizing any of the identity, location, and orientation of an object presented in a plurality of objects, a grasp selection system for selecting a grasp location on the object, the grasp location being chosen to provide a secure grasp of the object by the end effector to permit the object to be moved from the plurality of objects to one of a plurality of destination locations, and a motion planning system for providing a motion path for the transport of the object when grasped by the end effector from the plurality of objects to the one of the plurality of destination locations, wherein the motion path is chosen to provide a path from the plurality of objects to the one of the plurality of destination locations.

Learning to effectively imitate human teleoperators, even in unseen, dynamic environments is a promising path to greater autonomy, enabling robots to steadily acquire complex skills from supervision. Various motion generation techniques are described herein that are rooted in contraction theory and sum-of-squares programming for learning a dynamical systems control policy in the form of a polynomial vector field from a given set of demonstrations. Notably, this vector field is provably optimal for the problem of minimizing imitation loss while providing certain continuous-time guarantees on the induced imitation behavior. Techniques herein generalize to new initial and goal poses of the robot and can adapt in real time to dynamic obstacles during execution, with convergence to teleoperator behavior within a well-defined safety tube.