Visual Intelligence Grounded in Learning (VIGIL) | SRI International

Toggle Menu
illustration of a security camera with watching eyes

Visual Intelligence Grounded in Learning (VIGIL)

SRI’s visual intelligence system could enable a new era in unmanned robotic surveillance.

Artificial intelligence researchers at SRI are teaming with collaborators at the University of Leeds and the University of Maryland to enhance remote threat surveillance. The system under development, VIGIL, is being created through the Mind’s Eye program of the Defense Advanced Research Projects Agency (DARPA). VIGIL combines vision analysis, automatic reasoning, and machine learning to recognize interactions between people and objects, such as two people exchanging two boxes. The program goal is to produce a compact, field-portable system suitable for rapid, cost-effective deployment on unmanned vehicles.

VIGIL is being designed to use probabilistic reasoning, statistical and relational learning, and other artificial intelligence approaches to detect, recognize, and describe behaviors and activities seen in video. By enhancing unmanned robotic threat detection, the system would help keep personnel out of harm’s way, through minimizing the need to rely on advance scouts or reconnaissance teams.

For this project, SRI researchers and partners are working to solve several challenges:

  • Generate consistent interpretations from noisy visual input based on background knowledge about the structure of observed actions and objects
  • Recognize potential threats despite incomplete, noisy, or missing perceptual data
  • Learn new patterns and situation indicators with minimal human supervision

Related projects at SRI include the Cognitive Agent that Learns and Organizes (CALO) and Learning Applied to Ground Robots (LAGR), which uses real-time vision and learning to develop autonomous off-road navigation.