Abstract: A method includes: receiving data identifying, for each of one or more objects, a respective target location to which a robotic agent interacting with a real-world environment should move the object; causing the robotic agent to move the one or more objects to the one or more target locations by repeatedly performing the following: receiving a current image of a current state of the real-world environment; determining, from the current image, a next sequence of actions to be performed by the robotic agent using a next image prediction neural network that predicts future images based on a current action and an action to be performed by the robotic agent; and directing the robotic agent to perform the next sequence of actions.

Abstract: A method includes: receiving data identifying, for each of one or more objects, a respective target location to which a robotic agent interacting with a real-world environment should move the object; causing the robotic agent to move the one or more objects to the one or more target locations by repeatedly performing the following: receiving a current image of a current state of the real-world environment; determining, from the current image, a next sequence of actions to be performed by the robotic agent using a next image prediction neural network that predicts future images based on a current action and an action to be performed by the robotic agent; and directing the robotic agent to perform the next sequence of actions.

Abstract: One aspect provides a machine-learned video prediction model configured to receive and process one or more previous video frames to generate one or more predicted subsequent video frames, wherein the machine-learned video prediction model comprises a convolutional variational auto encoder, and wherein the convolutional variational auto encoder comprises an encoder portion comprising one or more encoding cells and a decoder portion comprising one or more decoding cells.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training a robotic control policy to perform a particular task. One of the methods includes performing a meta reinforcement learning phase including using training data collected for a plurality of different robotic control tasks and updating a robotic control policy according to the training data, wherein the robotic control policy is conditioned on an encoder network that is trained to predict which task is being performed from a context of a robotic operating environment; and performing an adaptation phase using a plurality of demonstrations for the particular task, including iteratively updating the encoder network after processing each demonstration of the plurality of demonstrations, thereby training the encoder network to learn environmental features of successful task runs.

Abstract: Optical tactile sensors are provided that include a scaffolding structure, a transparent elastomer material covering at least an end portion of the scaffolding structure, and one or multiple cameras situated on the end portion of the scaffolding structure and embedded within the transparent elastomer material, wherein the one or multiple cameras are situated so as to provide an extended, e.g., up to 360°, field of view about the end portion of the scaffolding structure.

Abstract: Methods, computer systems, and apparatus, including computer programs encoded on computer storage media, for generating and searching reinforcement learning algorithms. In some implementations, a computer-implemented system generates a sequence of candidate reinforcement learning algorithms. Each candidate reinforcement learning algorithm in the sequence is configured to receive an input environment state characterizing a state of an environment and to generate an output that specifies an action to be performed by an agent interacting with the environment. For each candidate reinforcement learning algorithm in the sequence, the system performs a performance evaluation for a set of a plurality of training environments. For each training environment, the system adjusts a set of environment-specific parameters of the candidate reinforcement learning algorithm by performing training of the candidate reinforcement learning algorithm to control a corresponding agent in the training environment.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for off-policy evaluation of a control policy. One of the methods includes obtaining policy data specifying a control policy for controlling a source agent interacting with a source environment to perform a particular task; obtaining a validation data set generated from interactions of a target agent in a target environment; determining a performance estimate that represents an estimate of a performance of the control policy in controlling the target agent to perform the particular task in the target environment; and determining, based on the performance estimate, whether to deploy the control policy for controlling the target agent to perform the particular task in the target environment.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for computing Q values for actions to be performed by an agent interacting with an environment from a continuous action space of actions. In one aspect, a system includes a value subnetwork configured to receive an observation characterizing a current state of the environment and process the observation to generate a value estimate; a policy subnetwork configured to receive the observation and process the observation to generate an ideal point in the continuous action space; and a subsystem configured to receive a particular point in the continuous action space representing a particular action; generate an advantage estimate for the particular action; and generate a Q value for the particular action that is an estimate of an expected return resulting from the agent performing the particular action when the environment is in the current state.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training an action selection neural network that is used to control a robotic agent interacting with a real-world environment.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for computing Q values for actions to be performed by an agent interacting with an environment from a continuous action space of actions. In one aspect, a system includes a value subnetwork configured to receive an observation characterizing a current state of the environment and process the observation to generate a value estimate; a policy subnetwork configured to receive the observation and process the observation to generate an ideal point in the continuous action space; and a subsystem configured to receive a particular point in the continuous action space representing a particular action; generate an advantage estimate for the particular action; and generate a Q value for the particular action that is an estimate of an expected return resulting from the agent performing the particular action when the environment is in the current state.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, of training a global policy neural network. One of the methods includes initializing an instance of the robotic task for multiple local workers, generating a trajectory of state-action pairs by selecting actions to be performed by the robotic agent while performing the instance of the robotic task, optimizing a local policy controller on the trajectory, generating an optimized trajectory using the optimized local controller, and storing the optimized trajectory in a replay memory associated with the local worker. The method includes sampling, for multiple global workers, an optimized trajectory from one of one or more replay memories associated with the global worker, and training the replica of the global policy neural network maintained by the global worker on the sampled optimized trajectory to determine delta values for the parameters of the global policy neural network.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for navigation using visual inputs. One of the systems includes a mapping subsystem configured to, at each time step of a plurality of time steps, generate a characterization of an environment from an image of the environment at the time step, wherein the characterization comprises an environment map identifying locations in the environment having a particular characteristic, and wherein generating the characterization comprises, for each time step: obtaining the image of the environment at the time step, processing the image to generate a first initial characterization for the time step, obtaining a final characterization for a previous time step, processing the characterization for the previous time step to generate a second initial characterization for the time step, and combining the first initial characterization and the second initial characterization to generate a final characterization for the time step.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, of training a global policy neural network. One of the methods includes initializing a plurality of instances of the robotic task. For each instance of the robotic task, the method includes generating a trajectory of state-action pairs by selecting actions to be performed by the robotic agent while performing the instance of the robotic task in accordance with current values of the parameters of the global policy neural network, and optimizing a local policy controller that is specific to the instance on the trajectory of state-action pairs for the instance. The method further includes generating training data for the global policy neural network using the local policy controllers, and training the global policy neural network on the training data to adjust the current values of the parameters of the global policy neural network.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for off-policy evaluation of a control policy. One of the methods includes obtaining policy data specifying a control policy for controlling a source agent interacting with a source environment to perform a particular task; obtaining a validation data set generated from interactions of a target agent in a target environment; determining a performance estimate that represents an estimate of a performance of the control policy in controlling the target agent to perform the particular task in the target environment; and determining, based on the performance estimate, whether to deploy the control policy for controlling the target agent to perform the particular task in the target environment.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training an action selection neural network that is used to control a robotic agent interacting with a real-world environment.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training an object representation neural network. One of the methods includes obtaining training sets of images, each training set comprising: (i) a before image of a before scene of the environment, (ii) an after image of an after scene of the environment after the robot has removed a particular object, and (iii) an object image of the particular object, and training the object representation neural network on the batch of training data, comprising determining an update to the object representation parameters that encourages the vector embedding of the particular object in each training set to be closer to a difference between (i) the vector embedding of the after scene in the training set and (ii) the vector embedding of the before scene in the training set.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training an object representation neural network. One of the methods includes obtaining training sets of images, each training set comprising: (i) a before image of a before scene of the environment, (ii) an after image of an after scene of the environment after the robot has removed a particular object, and (iii) an object image of the particular object, and training the object representation neural network on the batch of training data, comprising determining an update to the object representation parameters that encourages the vector embedding of the particular object in each training set to be closer to a difference between (i) the vector embedding of the after scene in the training set and (ii) the vector embedding of the before scene in the training set.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for navigation using visual inputs. One of the systems includes a mapping subsystem configured to, at each time step of a plurality of time steps, generate a characterization of an environment from an image of the environment at the time step, wherein the characterization comprises an environment map identifying locations in the environment having a particular characteristic, and wherein generating the characterization comprises, for each time step: obtaining the image of the environment at the time step, processing the image to generate a first initial characterization for the time step, obtaining a final characterization for a previous time step, processing the characterization for the previous time step to generate a second initial characterization for the time step, and combining the first initial characterization and the second initial characterization to generate a final characterization for the time step.

Abstract: A method includes: receiving data identifying, for each of one or more objects, a respective target location to which a robotic agent interacting with a real-world environment should move the object; causing the robotic agent to move the one or more objects to the one or more target locations by repeatedly performing the following: receiving a current image of a current state of the real-world environment; determining, from the current image, a next sequence of actions to be performed by the robotic agent using a next image prediction neural network that predicts future images based on a current action and an action to be performed by the robotic agent; and directing the robotic agent to perform the next sequence of actions.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for computing Q values for actions to be performed by an agent interacting with an environment from a continuous action space of actions. In one aspect, a system includes a value subnetwork configured to receive an observation characterizing a current state of the environment and process the observation to generate a value estimate; a policy subnetwork configured to receive the observation and process the observation to generate an ideal point in the continuous action space; and a subsystem configured to receive a particular point in the continuous action space representing a particular action; generate an advantage estimate for the particular action; and generate a Q value for the particular action that is an estimate of an expected return resulting from the agent performing the particular action when the environment is in the current state.