Machine learning for optimizing tool path planning in autonomous earth moving vehicles

Abstract

An autonomous earth moving system can select an action for an earth moving vehicle (EMV) to autonomously perform using a tool (such as an excavator bucket). The system then generates a set of candidate tool paths, each illustrating a potential path for the tool to trace as the earth moving vehicle performs the action. In some cases, the system uses an online learning model iteratively trained to determine which candidate tool path best satisfies one or more metrics measuring the success of the action. The earth moving vehicle the executes the earth moving action using the selected tool path and measures the results of the action. In some implementations, the autonomous earth moving system updates the machine learning model based on the result of the executed action.