Fast, Safe, and Proactive Runtime Planning and Control of Autonomous Ground Vehicles in Changing Environments

Abstract

Autonomous ground vehicles (UGVs) traversing paths in complex environments may have to adapt to changing terrain characteristics, including different friction, inclines, and obstacle configurations. In order to maintain safety, vehicles must make adjustments guided by runtime predictions of future velocities. To this end, we present a neural network-based framework for the proactive planning and control of an autonomous mobile robot navigating through different terrains. Using our approach, the mobile robot continually monitors the environment and the planned path ahead to accurately adjust its speed for successful navigation toward a desired goal. The target speed is selected by optimizing two criteria: (1) minimizing the rate of change between predicted and current vehicle speed and (2) maximizing the speed while staying within a safe distance from the desired path. Additionally, we introduce random noise into the network to model sensor uncertainty and reduce the risk of predicting unsafe speeds. We extensively tested and validated our framework on realistic simulations in Gazebo/ROS with a UGV navigating cluttered environments with different terrain frictions and slopes.