Automated Training And Use Of Predictive Dynamics Models For Autonomous Control Of Powered Earth-Moving Vehicles

A system, comprising:

• one or more hardware processors of one or more computing systems;
• and
• one or more memories with stored instructions that, when executed by at least one of the one or more hardware processors, cause at least one of the one or more computing systems to perform automated operations including at least:
• obtaining actual operational data from multiple earth-moving vehicles on one or more sites, wherein the actual operational data includes, for each of the multiple earth-moving vehicles, first data indicating inputs supplied by at least one human operator to physical controls of that earth-moving vehicle to cause performance of one or more tasks on an associated one site of the one or more sites on which that earth-moving vehicle is located, and second data gathered from sensors on that earth-moving vehicle during the performance of the one or more tasks, wherein the second data that is gathered from each of the multiple earth-moving vehicles includes vehicle data regarding vehicle status and movements of components of that earth-moving vehicle, the sensors of that earth-moving vehicle including one or more GPS antennas on a body of that earth-moving vehicle and one or more inclinometers mounted on at least one of the components of that earth-moving vehicle;
• preparing the actual operational data for use in training a model predictive control (MPC) vehicle dynamics model, including translating initial actual vehicle positions and orientations of the multiple earth-moving vehicles in the actual operational data to a single vehicle position and orientation, and translating subsequent actual vehicle positions and orientations of the multiple earth-moving vehicles in the actual operational data to additional vehicle positions and orientations that are relative to the single vehicle position and orientation;
• training the MPC vehicle dynamics model using the prepared actual operational data to encode information about movements of the components of the multiple earth-moving vehicles to implement the performance of the one or more tasks;
• and
• providing the trained MPC vehicle dynamics model for use in controlling autonomous operations on one or more additional earth-moving vehicles by predicting additional movements of additional components of the one or more additional earth-moving vehicles relative to the single vehicle position and orientation for additional inputs to additional physical controls of the one or more additional earth-moving vehicles.

A non-transitory computer-readable medium having stored contents that cause one or more computing systems to perform automated operations including at least:

• obtaining, by the one or more computing systems, actual operational data from multiple earth-moving vehicles on one or more sites, wherein the actual operational data includes, for each of the multiple earth-moving vehicles, first data indicating inputs supplied by at least one human operator to physical controls of that earth-moving vehicle to cause performance of one or more tasks on an associated one site of the one or more sites on which that earth-moving vehicle is located, and second data gathered from sensors on that earth-moving vehicle during the performance of the one or more tasks, wherein the second data that is gathered from each of the multiple earth-moving vehicles includes vehicle data regarding vehicle status and movements of components of that earth-moving vehicle, the sensors of that earth-moving vehicle including at least one of one or more GPS antennas on a body of that earth-moving vehicle or one or more inclinometers mounted on at least one of the components of that earth-moving vehicle;
• training, by one or more computing systems, a vehicle behavior and environment dynamics model using the actual operational data to encode information about the movements of the components of the multiple earth-moving vehicles in response to the inputs supplied to the physical controls of the multiple earth-moving vehicles;
• and
• providing, by one or more computing systems, the trained vehicle behavior and environment dynamics model for use in controlling autonomous operations on one or more additional earth-moving vehicles by predicting additional movements of additional components of the one or more additional earth-moving vehicles in response to additional inputs to additional physical controls of the one or more additional earth-moving vehicles.