Unmanned Ground Vehicle Convoy Operations in Narrow Spaces (UGVCONS)

hptech.ai and Texas A&M University's RELLIS Campus, Bush Combat Development Complex and Unmanned System Lab are developing two Unmanned Ground Vehicle Convoy Operations in Narrow Spaces (UGVCONS) concepts by testing them on actual autonomous vehicles. Real vehicles will be used because current state-of-the-art simulations are not reliable enough to support evaluation of sensor performance to changing environment/weather conditions. Neither concept needs GPS, although it can be incorporated. Option A focuses on rapid deployment by circumventing the typical issues of autonomy by temporarily autonomously physically connecting the vehicles with a robot arm like device to share information over a wired connection (zero-EMF). Existing work on commercial trucks by hptech.ai will be translated and studied in the context of the ROUGE-Fires convoy in a variety of on/off-road conditions. The physical tether is a bolt-on solution much like a winch and is expected to track the leaderÆs path within inches. The system can also be connected/disconnected on-demand (button press) to transition from existing operating modes. If a connected vehicle slips into a ditch there is the ability to push/pull through the physical tether to get out, significantly increasing the convoyÆs mobility. The system is robust in all weather conditions and eliminates the need for high compute which enables reliable, explainable, cost effective and rapid deployment. Option B, Path Sharing, is an improvement on the typical delayed following leader-follower methods. The leader creates a map and path relative to its initial position utilizing all existing on-board sensors. The map is created with state-of-the-art simultaneous localization and mapping (SLAM) techniques that fuse visual, lidar and radar to improve odometry with or without using GPS. The high-definition map is then compressed into a costmap and radioed from the leader, along with the path traversed, to the followers so they can localize themselves and track the same path. In both options a model predictive control (MPC) approach will be utilized to handle any communication, computation or CAN-bus/actuation time delays and to minimize the path following cross track error.