Intelligent Automation, Inc. — Department of Defense STTR Phase I: A17A-T019

Description

For operational robots to be truly effective in the battlefield, they would need to be integrated with intelligent decision-making capabilities. In particular, the following capabilities would help them to deal with the challenging real-world problems of off-road navigation, namely, traversability assessment (Capability 1), optimal trajectory computation (Capability 2), and optimal maneuver selection (Capability 3). In this STTR effort, we propose to develop ORION, a computational framework that provides an operational robot with intelligent capabilities to effectively deal with real-world problems of off-road navigation. ORION treats Capability 1 as a classification problem, and addresses it by employing state-of-art deep learning techniques. It treats Capability 2 as a heuristic motion planning problem, and solves it by applying state-of-art classical artificial intelligence (AI) techniques of sampling-based kinodynamic planning. Finally, it treats Capability 3 as a policy mapping problem, and solves it by utilizing standard reinforcement learning.