Aptima at CogSci Virtual 2020

How do you ensure the reliability of an intelligent agent?

Aptima’s Tim Halverson and Mary Freiman with their L3Harris partner and AFRL customer presented current work in cognitive science at COGSCI 2020 Virtual Meeting (29 July – 1 August 2020). An ‘on-demand’ service is available to all registered attendees, where video content of the presentations can be viewed at any time, up to 30 days post conference.

Tim, Mary, and their L3Harris and AFRL colleagues are developing an intelligent agent that operates a remotely piloted aircraft with two human teammates that communicate using text chat. Together, the two human teammates (a photographer and navigator) and the synthetic agent (the pilot) must fly the aircraft to specified locations and photograph targets on the ground. While the task is well-defined, there are potentially numerous and unpredictable inputs during each varied 40-minute missions. To assure the reliability of agent behavior, the team must be able to run a large number of missions and analyze agent behavior at milliseconds resolution. To solve these challenges, Aptima, L3Harris, and AFRL have developed a set of tools that form a framework for verifying complex agent behavior as development progresses. View Poster (PDF)