Methods, systems, and apparatus, including computer programs encoded on computer storage media, for integrating sensor streams for robotic demonstration learning. One of the methods includes selecting, by a learning system for a robot, a base update rate for combining multiple sensor streams into a task state representation. The learning system repeatedly generates the task state representation at the base update rate, including combining, during each time period defined by the update rate, the task state representation from most recently updated sensor data processed by the plurality of neural networks. The learning system repeatedly uses the task state representations to generate commands for the robot at the base update rate.

Systems, devices, and methods for training and operating (semi-)autonomous robots to complete multiple different work objectives are described. A robot control system stores a library of reusable work primitives each corresponding to a respective basic sub-task or sub-action that the robot is operative to autonomously perform. A work objective is analyzed to determine a sequence (i.e., a combination and/or permutation) of reusable work primitives that, when executed by the robot, will complete the work objective. The robot executes the sequence of reusable work primitives to complete the work objective. The reusable work primitives may include one or more reusable grasp primitives that enable(s) a robot's end effector to grasp objects. Simulated instances of real physical robots may be trained in simulated environments to develop control instructions that, once uploaded to the real physical robots, enable such real physical robots to autonomously perform reusable work primitives.

Implementations described herein relate to training and refining robotic control policies using imitation learning techniques. A robotic control policy can be initially trained based on human demonstrations of various robotic tasks. Further, the robotic control policy can be refined based on human interventions while a robot is performing a robotic task. In some implementations, the robotic control policy may determine whether the robot will fail in performance of the robotic task, and prompt a human to intervene in performance of the robotic task. In additional or alternative implementations, a representation of the sequence of actions can be visually rendered for presentation to the human can proactively intervene in performance of the robotic task.

Systems, devices, and methods for training and operating (semi-)autonomous robots to complete multiple different work objectives are described. A robot control system stores a library of reusable work primitives each corresponding to a respective basic sub-task or sub-action that the robot is operative to autonomously perform. A work objective is analyzed to determine a sequence (i.e., a combination and/or permutation) of reusable work primitives that, when executed by the robot, will complete the work objective. The robot executes the sequence of reusable work primitives to complete the work objective. The reusable work primitives may include one or more reusable grasp primitives that enable(s) a robot's end effector to grasp objects. Simulated instances of real physical robots may be trained in simulated environments to develop control instructions that, once uploaded to the real physical robots, enable such real physical robots to autonomously perform reusable work primitives.

Systems, devices, and methods for training and operating (semi-)autonomous robots to complete multiple different work objectives are described. A robot control system stores a library of reusable work primitives each corresponding to a respective basic sub-task or sub-action that the robot is operative to autonomously perform. A work objective is analyzed to determine a sequence (i.e., a combination and/or permutation) of reusable work primitives that, when executed by the robot, will complete the work objective. The robot executes the sequence of reusable work primitives to complete the work objective. The reusable work primitives may include one or more reusable grasp primitives that enable(s) a robot's end effector to grasp objects. Simulated instances of real physical robots may be trained in simulated environments to develop control instructions that, once uploaded to the real physical robots, enable such real physical robots to autonomously perform reusable work primitives.

Systems, devices, and methods for training and operating (semi-)autonomous robots to complete multiple different work objectives are described. A robot control system stores a library of reusable work primitives each corresponding to a respective basic sub-task or sub-action that the robot is operative to autonomously perform. A work objective is analyzed to determine a sequence (i.e., a combination and/or permutation) of reusable work primitives that, when executed by the robot, will complete the work objective. The robot executes the sequence of reusable work primitives to complete the work objective. The reusable work primitives may include one or more reusable grasp primitives that enable(s) a robot's end effector to grasp objects. Simulated instances of real physical robots may be trained in simulated environments to develop control instructions that, once uploaded to the real physical robots, enable such real physical robots to autonomously perform reusable work primitives.

Implementations described herein relate to training and refining robotic control policies using imitation learning techniques. A robotic control policy can be initially trained based on human demonstrations of various robotic tasks. Further, the robotic control policy can be refined based on human interventions while a robot is performing a robotic task. In some implementations, the robotic control policy may determine whether the robot will fail in performance of the robotic task, and prompt a human to intervene in performance of the robotic task. In additional or alternative implementations, a representation of the sequence of actions can be visually rendered for presentation to the human can proactively intervene in performance of the robotic task.

Various aspects of methods, systems, and use cases include techniques for training or using a model to control a robot. A method may include identifying a set of action primitives applicable to a set of robots, receiving information corresponding to a task (e.g., a collaborative task), and determining at least one action primitive based on the received information. The method may include training a model to control operations of at least one robot of the set of robots using the received information and the at least one action primitive.

A method of training a robot system for manipulation of objects, the robot system being able to perform a set of skills, wherein each skill is learned as a skill model, the method comprising: receiving physical input from a human trainer, regarding the skill to be learned by the robot; determining for the skill model a set of task parameters including determining for each task parameter of the set of task parameters if a task parameter is an attached task parameter, which is related to an object being part of said kinesthetic demonstration or if a task parameter is a free task parameter, which is not related to a physical object; obtaining data for each task parameter of the set of task parameters from the set of kinesthetic demonstrations, and training the skill model with the set of task parameters and the data obtained for each task parameter.

A skill-based robot programming apparatus is disclosed including a master DB for storing a work cell item including a robot or a peripheral and having a relevant parameter and a master skill, which is a set of commands for driving the work cell item, a work cell manager for selecting a work cell item to be used among the work cell items stored in the master DB, and inputting a parameter of the selected work cell item, a work cell engine for searching the master skill relevant to the selected work cell item from the master DB, and generating a user skill by applying at least one parameter among the parameters of the selected work cell item to the searched master skill, a user DB for storing the selected work cell item and the user skill, and a task builder for generating a task that the robot should work.