AI Vision Based Guidance

The USAF is constantly challenged to expand the possible course of actions available to the warfighter in complex and unpredictable battlespaces, offensively, defensively, and in support of both. Low-cost, autonomous air systems equipped with the vision based guidance systems can leverage advances in Artificial Intelligence and low-cost sensors to operate passively in GPS denied environments. These low-end systems can provide precision delivery of payloads to the warfighter or help guide UAS swarm technology to attack adversary weaknesses in unexpected ways by exploiting the use large numbers and delivery complexity. To help address these challenges, Defense Research Associates (DRA) is developing a high precision vision based guidance system (VBGS) for use on airdropped payloads operating in GPS denied environments. The VBGS will provide real-time directional guidance towards stationary and moving landing zones (LZ) previously identified in reference high altitude imagery. DRA's algorithm concept receives a high altitude reference image of the LZ and uses AI to locate the LZĺs current location using vision-based sensors and INS position. DRAĺs approach combines emerging and well-studied technologies in both AI object localization and computer vision to provide the maximum probability of finding an a priori LZ (identified in a high altitude reference image) with the LZ in the sensorĺs image space for providing continuous guidance. By incorporating these AI methods, DRAĺs algorithm is able to use scene context to estimate the LZĺs relative position even when the LZ is at a range at which it cannot be resolved visually, or is temporally obstructed, or outside the Field of View. When the LZ is resolvable, the algorithm is capable of focusing on specific LZ features, thus maximizing guidance precision without relying on unique signatures or non-passive emitted or reflected signals that disclose positions or require persistent line of sight between the LZ and the warfighter.