A machine learning system builds and uses computer models for controlling robotic performance of a task. Such computer models may be first trained using feedback on computer simulations of the robot performing the task, and then refined using feedback on real-world trials of the robot performing the task. Some examples of the computer models can be trained to automatically evaluate robotic task performance and provide the feedback. This feedback can be used by a machine learning system, for example an evolution strategies system or reinforcement learning system, to generate and refine the controller.

Methods, systems, and apparatus, including computer programs encoded on computer storage media for evaluating robot learning. In some implementations, one or more computers receive object classification examples from a plurality of robots. Each object classification example includes (i) an embedding that a robot generated using a machine learning model, and (ii) an object classification corresponding to the embedding. The object classification examples are evaluated based on a similarity of the received embeddings with respect to other embeddings. A subset of the object classification examples is selected based on the evaluation of the quality of the embeddings. The subset of the object classification examples is distributed to the robots in the plurality of robots.

A machine learning system builds and uses computer models for identifying how to evaluate the level of success reflected in a recorded observation of a task. Such computer models may be used to generate a policy for controlling a robotic system performing the task. The computer models can also be used to evaluate robotic task performance and provide feedback for recalibrating the robotic control policy.

A machine learning system builds and uses computer models for identifying how to evaluate the level of success reflected in a recorded observation of a task. Such computer models may be used to generate a policy for controlling a robotic system performing the task. The computer models can also be used to evaluate robotic task performance and provide feedback for recalibrating the robotic control policy.

A teaching device capable of teaching not only movement work but also more detailed working content. The teaching device is provided with input section for inputting work information such as work of pinching workpieces which is carried out by a robot arm at a working position. When carrying out motion capture by moving jig (an object which mimics the robot arm) which is provided with marker section, a user manipulate input section at an appropriate timing to input the working content to be performed by the robot arm as work information, and thus it is possible to set fine working content of the robot arm in teaching device. Accordingly, teaching device is capable of linking positional information of jig and the like and work information generating control information for controlling the robot arm.

An object attitude detection device includes a pick-up image acquisition unit, a template image acquisition unit, and an attitude decision unit. The pick-up image acquisition unit acquires a picked-up image of an object. The template image acquisition unit acquires a template image for each attitude of the object. The attitude decision unit decides an attitude of the object based on the template image having pixels. In the pixels, a distance between pixels forming a contour in the picked-up image and pixels forming a contour of the template image is shorter than a first threshold. Further, a degree of similarity between a gradient of the pixels forming the contour in the picked-up image and a gradient of the pixels forming the contour of the template image is higher than a second threshold.

Methods, systems, and apparatus, including computer-readable media storing executable instructions, for enhancing robot learning. In some implementations, a robot stores first embeddings generated using a first machine learning model, and the first embeddings include one or more first private embeddings that are not shared with other robots. The robot receives a second machine learning model from a server system over a communication network. The robot generates a second private embedding for each of the one or more first private embeddings using the second machine learning model. The robot adds the second private embeddings to the cache of the robot and removes the one or more first private embeddings from the cache of the robot.

To provide an action information learning device, robot control system and action information learning method for facilitating the performing of cooperative work by an operator with a robot. An action information learning device includes: a state information acquisition unit that acquires a state of a robot; an action information output unit for outputting an action, which is adjustment information for the state; a reward calculation section for acquiring determination information, which is information about a handover time related to handover of a workpiece, and calculating a value of reward in reinforcement learning based on the determination information thus acquired; and a value function update section for updating a value function by way of performing the reinforcement learning based on the value of reward calculated by the reward calculation section, the state and the action.

The present invention relates to a method, such as a surgical method for assisting a surgeon for placing screws in the spine using a robot attached to a passive structure. The present invention also related to a method, such as a surgical method for assisting a surgeon for removing volumes in the body of a patient using a robot attached to a passive structure and to a device to carry out said methods. The present invention further concerns a device suitable to carry out the methods according to the present invention.

A robotic system (new robot) operative for performing at least one task in an environment, the system comprising: learn-from-predecessor functionality governed by a data exchange protocol, which controls short-range wireless knowledge transfer from a short-range wireless transmitter in a predecessor robot system (old robot) to a short-range wireless receiver in said robotic system, said knowledge comprising at least one environment-specific datum previously stored by the predecessor robot.