Dense Monocular 3D Sensing for Autonomous Aerial Systems

The United States Air Force 2030 Science and Technology Strategy addresses a new era of great power competition where the Air Force can no longer assume technological superiority over potential adversaries. Recent conflicts have confirmed that rival powers are investing in artificial intelligence (AI), autonomous platforms and weapons, denial of GPS, and alternatives to GPS for their own forces. In response, the 2030 Strategy calls for several transformational capabilities to be driven by scientific and engineering advances. These capabilities will include Global Persistent Awareness, Resilient Information Sharing, and Rapid Effective Decision Making that in turn require advancements in sensing, self-contained positioning and navigation, AI, and autonomous systems among other technologies. Where possible the Air Force will leverage commercial investment in these technologies, which greatly exceeds the resources of the Defense Department.  Compound Eye is a leading commercial developer of 3D sensing and perception technology for autonomous vehicles and other robots. The company’s VIDAS technology (“Visual Inertial Distributed Aperture System”) uses off-the-shelf, automotive-grade cameras, inertial sensors, and AI to build a dense 3D representation of the world around a vehicle in real time. This information is critical for an autonomous system to avoid static and dynamic obstacles and can enable self-contained positioning, mitigating or eliminating the need for external positioning systems like GPS.  VIDAS has drawn interest from the DoD because it delivers the same information as active sensors such as lidar and radar for a fraction of the size, weight, power, and cost, all with zero signature. However, the current commercial version of VIDAS relies in part on wide baseline stereo sensing, which is not suitable for small unmanned platforms. For these applications AFRL (AFRL/RWWI) has challenged Compound Eye to develop a version of VIDAS that requires only a single camera.  The primary objectives of the proposed effort are:  1. Verify experimentally that monocular depth from optic flow and semantic cues can provide robust, accurate, real-time depth measurements   2. Develop and validate a monocular version of VIDAS for aerial platforms  3. Modify the commercial Developer Kit for VIDAS to support monocular depth and train AF Customers to use the technology