Abstract: In various examples, a generalizable mobility model can receive a state and identifier of a robot, and generate an action for the robot based on the state and the identifier. The state can identify a position, environment, and navigation goal of the robot while the identifier can indicate a type of the robot. The generalizable mobility model can use the identifier to generate an action for the robot to reach the navigation goal from its position based on the type of the robot, such as a humanoid, quadruped, or wheeled robot. The generalizable mobility model can be a distilled combination of multiple robot type-specific models, and can use the identifier to mimic type-specific actions output by the multiple robot type-specific models to transmit to the robot to move the robot.

Abstract: In various examples, a single generalizable mobility model can be applied to a variety of robot types. Actions output by the generalizable mobility model can be based on actions generated using imitation learning, and the actions can be refined per type of robot using residual reinforcement learning. A type-specific model can be updated per type of robot using residual reinforcement learning, and can be combined and distilled into the single generalizable model. Actions output by the type-specific model can be combined with actions generated using imitation learning, and results of the action on a simulated robot can update weights of one layer of the type-specific model. By building upon an imitation learning model and refining each type-specific model using residual reinforcement learning, the generalizable mobility model can improve data and time efficiency compared to continuous multi-type training while also avoiding joint multi-type training complexities.

Abstract: In various examples, a technique for enhancement of latent states by visual reinforcement learning includes converting a set of sensory inputs obtained using one or more sensors of a machine at a current time step into a set of embedded features. The technique also includes, based at least on a latent state of a machine at a current time step, and the set of embedded features, using a machine learning model employing a reinforcement learning algorithm to implement an action policy. The technique further includes, based at least on the action policy, outputting an action associated with one or more future time steps for the machine to perform.

Abstract: A vehicle includes a ranged sensor that generates time-series data indicating positions of objects in an environment surrounding the vehicle, a user interface configured to warn the driver of a predicted collision between the vehicle and one of the objects in the environment, and at least one processor including an ECU operatively connected to the ranged sensor and the user interface. The processor records control inputs by the driver driving the vehicle, and develops a driver behavior model associated with the driver driving the vehicle based on the control inputs. The processor also predicts trajectories of the objects and the vehicle based on the time-series data and the driver behavior model, and predicts a collision between the vehicle and one of the objects based on the predicted trajectories. The processor also generates a warning indicating the predicted collision to the driver.

Abstract: A data processing service builds a container for a customer to run a trained large language model (LLM). The data processing service receives a trained LLM and a desired configuration from a user of a client device. Based on the desired configuration, the data processing service selects a hardware configuration and structures weights of the trained LLM based on the hardware configuration. The data processing service generates a container image reflecting the hardware configuration, registers the container image to a container registry, and generates a container from the container image as well as an application programming interface (API) endpoint for the container. The data processing service deploys the trained LLM in the API endpoint using the container such that the trained LLM is accessible through API calls.

Abstract: A vehicle includes a ranged sensor that generates time-series data indicating positions of objects in an environment surrounding the vehicle, a user interface configured to warn the driver of a predicted collision between the vehicle and one of the objects in the environment, and at least one processor including an ECU operatively connected to the ranged sensor and the user interface. The processor records control inputs by the driver driving the vehicle, and develops a driver behavior model associated with the driver driving the vehicle based on the control inputs. The processor also predicts trajectories of the objects and the vehicle based on the time-series data and the driver behavior model, and predicts a collision between the vehicle and one of the objects based on the predicted trajectories. The processor also generates a warning indicating the predicted collision to the driver.