Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating a final classification output for an image of eye tissue. The image is provided as input to each of one or more segmentation neural networks to obtain one or more segmentation maps of the eye tissue in the image. A respective classification input is generated from each of the segmentation maps. For each of the segmentation maps, the classification input for the segmentation map is provided as input to each of one or more classification neural networks to obtain, for each segmentation map, a respective classification output from each classification neural network. A final classification output for the image is generated from the respective classification outputs for each of the segmentation maps.

Abstract: Mobile self-balancing robots for telepresence are provided. The robots comprise a base, a head, and a shaft therebetween. The shaft can be telescoping to allow the head to be extended above the base to about the height of a normal sized person, or can be retracted to make the robot more compact for easier storage and transportation. The head includes components for telepresence such as cameras, a speaker, a microphone, a laser pointer, and a display screen, as well as protection from impacts and falls. The base provides locomotion and balance, and a narrow shaft between the head and base minimizes the robot's weight and reduces the likelihood of collisions with protrusions like table edges.

Abstract: Systems and methods as described for providing visual telepresence to an operator of a remotely controlled robot. The robot includes both video cameras and pose sensors. The system can also comprise a head-tracking sensor to monitor the orientation of the operator's head. These signals can be used to aim the video cameras. The controller receives both the video signals and the pose sensor signals from the robot, and optionally receives head-tracking signals from the head-tracking sensor. The controller stitches together the various video signals to form a composite video signal that maps to a robot view. The controller renders an image to a display from that portion of the composite video signal that maps to an operator view. The relationship of the operator view to the robot view is varied according to the signals from the pose sensors and the head-tracking sensor.

Abstract: Systems and methods are provided for projecting a stabilized laser beam from a robot being controlled through a human interface. The laser beam can be stabilized through the use of optics that continuously adjust the aim of the laser beam in response to measurements of the orientation of the robot. The human interface allows the operator to both observe the visual environment of the robot and also to aim the laser beam. The projected laser beam allows the operator to communicate from the perspective of the robot by pointing to objects or locations within the robot's visual environment, creating symbols with the laser beam, and make gestures with the laser beam.

Abstract: Mobile self-balancing robots for telepresence are provided. The robots comprise a base, a head, and a shaft therebetween. The shaft can be telescoping to allow the head to be extended above the base to about the height of a normal sized person, or can be retracted to make the robot more compact for easier storage and transportation. The head includes components for telepresence such as cameras, a speaker, a microphone, a laser pointer, and a display screen, as well as protection from impacts and falls. The base provides locomotion and balance, and a narrow shaft between the head and base minimizes the robot's weight and reduces the likelihood of collisions with protrusions like table edges.

Abstract: Systems and methods as described for providing visual telepresence to an operator of a remotely controlled robot. The robot includes both video cameras and pose sensors. The system can also comprise a head-tracking sensor to monitor the orientation of the operator's head. These signals can be used to aim the video cameras. The controller receives both the video signals and the pose sensor signals from the robot, and optionally receives head-tracking signals from the head-tracking sensor. The controller stitches together the various video signals to form a composite video signal that maps to a robot view. The controller renders an image to a display from that portion of the composite video signal that maps to an operator view. The relationship of the operator view to the robot view is varied according to the signals from the pose sensors and the head-tracking sensor.

Abstract: Robots comprising two links joined by a pivot joint are provided. In some cases, the pivot joint allows the robot to lean to either side. One link of the robot includes an electrically activated actuator such as an electric motor configured to rotate a pulley. A belt is engaged with the actuator, and the ends of the belt are coupled to the other link on either side of the pivot joint. Tensioners, such as springs, provide tension on either side of the belt. Actuating the actuator changes the position of the belt to respond to sloping surfaces and turns, for example.

Abstract: A power system is provided for robotic systems such as dynamically balancing robots, including inverted pendulum robots, robots that balance on two legs, and dynamically balancing robotic personal vehicles. Under normal operation, the power system provides operating power from a power source such as an internal battery system or an external AC power supply. The power system includes an energy storage device, such as an ultracapacitor assembly, that can provide emergency power sufficient to power components of the robotic system to a stable shutdown configuration in the event of a deleterious power event.

Abstract: A robotic hand and arm where the fingers are driven by rotational motors with drums, and pulled with cables using rolling friction. The hand extends into a robotic arm through a wrist wherein the wrist is controlled by pneumatic cylinders. Each finger preferably is provided with four manufactured parts and a single pulley. The thumb is provided with three pulleys for independent distal movement. Cables wraps around or over pulleys, eliminating tight bends. A glove is provided about the robotic hand which provides a compressive, liquid resistant membrane.

Abstract: A robotic hand and arm where the fingers are driven by rotational motors with drums, and pulled with cables using rolling friction. The hand extends into a robotic arm through a wrist wherein the wrist is controlled by pneumatic cylinders. Each finger preferably is provided with four manufactured parts and a single pulley. The thumb is provided with three pulleys for independent distal movement. Cables wraps around or over pulleys, eliminating tight bends. A glove is provided about the robotic hand which provides a compressive, liquid resistant membrane.