From Legible to Inscrutable Trajectories: (Il)legible Motion Planning Accounting for Multiple Observers
New AI system teaches robots to be deceptive in adversarial environments.
Researchers have developed DUBIOUS, a new motion planning algorithm that allows robots to generate trajectories that are legible to friendly observers while remaining inscrutable to adversarial ones. The system solves the Mixed-Motive Limited-Observability Legible Motion Planning (MMLO-LMP) problem, balancing communication with concealment based on each observer's motives and visibility limitations. This enables a single robot to simultaneously signal intentions to allies in cooperative settings while hiding them from opponents in adversarial scenarios like military operations.
Why It Matters
This breakthrough could revolutionize autonomous systems in security, defense, and competitive environments where strategic deception is crucial.