Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for training an autoencoder. One of the methods includes receiving a pair of images, wherein a first image of the pair represents a first state of a working environment at a first time step earlier than that of a second image of the pair. A first and a second latent representation is generated respectively for the first and the second image by the autoencoder; A predicted reward for an action executed in the first state is generated by a reward prediction neural network. A predicted next latent representation for the first state and the action is generated by a dynamics prediction neural network. An overall loss is determined based on the predicted reward, the predicted next latent representation, and the second latent representation. Model parameters of the autoencoder are updated to reduce the overall loss.

Abstract: A connection structure for a composite backrest and a method for making the same are provided. The connection structure includes two frames, a backrest body, and metal bushings. The metal bushings include a group of left metal bushings and a group of right metal bushings. The two frames are respectively fixed to the left side and the right side of the backrest body by the left metal bushings and the right metal bushings. The method includes: punching the backrest body and the frames; placing the frames into the backrest body; processing the metal bushings; sandblasting or sanding the external surface of the metal bushing; coating a configured structural adhesive on the metal bushing, and oppositely bonding the metal bushing; fixing the bonded metal bushing by a tool; and placing the fixed metal bushing, the composite backrest and the two frames into a drying oven for heating.

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: Generating and utilizing action image(s) that represent a candidate pose (e.g., a candidate end effector pose), in determining whether to utilize the candidate pose in performance of a robotic task. The action image(s) and corresponding current image(s) can be processed, using a trained critic network, to generate a value that indicates a probability of success of the robotic task if component(s) of the robot are traversed to the particular pose. When the value satisfies one or more conditions (e.g., satisfies a threshold), the robot can be controlled to cause the component(s) to traverse to the particular pose in performing the robotic task.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for using learnable robotic control plans. One of the methods comprises obtaining a learnable robotic control plan comprising data defining a state machine that includes a plurality of states and a plurality of transitions between states, wherein: one or more states are learnable states, and each learnable state comprises data defining (i) one or more learnable parameters of the learnable state and (ii) a machine learning procedure for automatically learning a respective value for each learnable parameter of the learnable state; and processing the learnable robotic control plan to generate a specific robotic control plan, comprising: obtaining data characterizing a robotic execution environment; and for each learnable state, executing, using the obtained data, the respective machine learning procedures defined by the learnable state to generate a respective value for each learnable parameter of the learnable state.

Abstract: A mold for an ultra-thick walled U-shaped composite product with a deep cavity includes an upper mold cavity, a lower mold cavity, a slidable side-drawing insert, and an auxiliary mold clamping structure. The auxiliary mold clamping structure includes a first auxiliary mold clamping insert and a second auxiliary mold clamping insert. The first auxiliary mold clamping insert is fixed on the lower mold cavity, and the second auxiliary mold clamping insert is fixed on the slidable side-drawing insert. The slidable side-drawing insert is located on a side of an integral structure formed after the upper mold cavity is engaged with the lower mold cavity, and the slidable side-drawing insert longitudinally slides along the lower mold cavity. The ultra-thick walled U-shaped composite product is formed and located between the upper mold cavity, the lower mold cavity, and the slidable side-drawing insert.

Abstract: A connection structure for a composite backrest and a method for making the same are provided. The connection structure includes two frames, a backrest body, and metal bushings. The metal bushings include a group of left metal bushings and a group of right metal bushings. The two frames are respectively fixed to the left side and the right side of the backrest body by the left metal bushings and the right metal bushings. The method includes: punching the backrest body and the frames; placing the frames into the backrest body; processing the metal bushings; sandblasting or sanding the external surface of the metal bushing; coating a configured structural adhesive on the metal bushing, and oppositely bonding the metal bushing; fixing the bonded metal bushing by a tool; and placing the fixed metal bushing, the composite backrest and the two frames into a drying oven for heating.

Abstract: Cascading variational autoencoder (“VAE”) models can be used to detect robot collisions while a robot is performing a task. For a current state of the robot, various implementations include a VAE used to generate a latent space of the current state, and a predictor network used to generate a predicted latent space for the current state. A collision can be determined based on a difference between the latent space for the current state and the predicted latent space for the current state.

Abstract: Cascading variational autoencoder (“VAE”) models can be used to detect robot collisions while a robot is performing a task. For a current state of the robot, various implementations include a VAE used to generate a latent space of the current state, and a predictor network used to generate a predicted latent space for the current state. A collision can be determined based on a difference between the latent space for the current state and the predicted latent space for the current state.

Abstract: Systems and methods for controlling robots including industrial robots. A method includes executing (402) a program (550) to control a robot (102) by the robot control system (120, 500). The method includes receiving (404) robot state information (554). The method includes receiving (406) force torque feedback (556) inputs from a sensor (554) on the robot (102). The method includes producing (410) a robot control command for the robot (102) based on the robot state information (554) and the force torque feedback (556) inputs. The method includes controlling (412) the robot (102) using the robot control command.

Abstract: Generating and utilizing action image(s) that represent a candidate pose (e.g., a candidate end effector pose), in determining whether to utilize the candidate pose in performance of a robotic task. The action image(s) and corresponding current image(s) can be processed, using a trained critic network, to generate a value that indicates a probability of success of the robotic task if component(s) of the robot are traversed to the particular pose. When the value satisfies one or more conditions (e.g., satisfies a threshold), the robot can be controlled to cause the component(s) to traverse to the particular pose in performing the robotic task.

Abstract: A process for manufacturing an apron board of a high-speed rail equipment cabin using a composite material is disclosed. The material includes aramid fiber honeycomb, PET foam, 3K twill carbon fiber flame retardant prepreg, unidirectional carbon fiber flame retardant prepreg, glass fiber flame retardant prepreg, aramid flame retardant prepreg, and 300 g/m2 single component medium temperature curing blue epoxy adhesive. The process includes manufacturing an apron main plate (3); manufacturing apron-board trim strips (1, 2), wherein there are two apron-board trim strips (1) and two apron-board trim strips (2); and obtaining the apron board through the apron main plate (3) and the apron-board trim strips (1, 2), wherein the two apron-board trim strips (1) are respectively stuck at two opposite sides of the apron main plate (3), the two apron-board trim strips (2) are respectively stuck at another two opposite sides of the apron main plate (3).

Abstract: A method for molding a flame-retardant bending beam integrated with a three-dimensional (3D) nylon air duct and a production mould thereof are provided. The method includes steps of: (1), producing the 3D nylon air duct; (2), selecting a flame-retardant carbon fiber pre-impregnated material: (3), laminating the pre-impregnated material and pre-molding; and (4), curing and molding composite material. The production mould includes an upper mould, a lower mould, sliding blocks and an air nozzle port. Compared with a traditional bending beam of metal molding, the flame-retardant bending beam integrated with the 3D nylon air duct achieves a same flame-retardant effect, and meanwhile has simple operation, flexible product shape, higher strength, lighter weight and corrosion resistance.

Abstract: A process for manufacturing a base board of a high-speed rail equipment cabin using a composite material is disclosed. The composite material includes: aramid honeycomb, PET foam, 3K twill carbon fiber flame retardant prepreg, unidirectional carbon fiber flame retardant prepreg, glass fiber flame retardant prepreg, aramid flame retardant prepreg, and 300 g/cm2 single component medium temperature curing blue epoxy adhesive. The process includes manufacturing a base-board main plate (1), a base-board handle (2) and two base-board sliders (3). While installation, the base-board handle (2) is stuck to one side of the base-board main plate (1), and the two base-board sliders (3) are respectively stuck to another two opposite sides of the base-board main plate (1). The weight of the base board made from the composite material is 35%-40% lower than the base board made from the aluminum alloy material, which leads to a good prospect of application.

Abstract: A method for molding a flame-retardant bending beam integrated with a three-dimensional (3D) nylon air duct and a production mould thereof are provided. The method includes steps of: (1), producing the 3D nylon air duct; (2), selecting a flame-retardant carbon fiber pre-impregnated material; (3), laminating the pre-impregnated material and pre-molding; and (4), curing and molding composite material.

Abstract: A process for manufacturing an apron board of a high-speed rail equipment cabin using a composite material is disclosed. The material includes aramid fiber honeycomb, PET foam, 3K twill carbon fiber flame retardant prepreg, unidirectional carbon fiber flame retardant prepreg, glass fiber flame retardant prepreg, aramid flame retardant prepreg, and 300 g/m2 single component medium temperature curing blue epoxy adhesive. The process includes manufacturing an apron main plate (3); manufacturing apron-board trim strips (1, 2), wherein there are two apron-board trim strips (1) and two apron-board trim strips (2); and obtaining the apron board through the apron main plate (3) and the apron-board trim strips (1, 2), wherein the two apron-board trim strips (1) are respectively stuck at two opposite sides of the apron main plate (3), the two apron-board trim strips (2) are respectively stuck at another two opposite sides of the apron main plate (3).

Abstract: A process for manufacturing a base board of a high-speed rail equipment cabin using a composite material is disclosed. The composite material includes: aramid honeycomb, PET foam, 3K twill carbon fiber flame retardant prepreg, unidirectional carbon fiber flame retardant prepreg, glass fiber flame retardant prepreg, aramid flame retardant prepreg, and 300 g/cm2 single component medium temperature curing blue epoxy adhesive. The process includes manufacturing a base-board main plate (1), a base-board handle (2) and two base-board sliders (3). While installation, the base-board handle (2) is stuck to one side of the base-board main plate (1), and the two base-board sliders (3) are respectively stuck to another two opposite sides of the base-board main plate (1). The weight of the base board made from the composite material is 35%-40% lower than the base board made from the aluminum alloy material, which leads to a good prospect of application.