New deceptive path planning algorithm outsmarts adversarial observers with linear programming

In adversarial environments, mobile agents often need to reach a goal while avoiding detection or interception. Traditional path planning assumes passive observers that simply infer intent from trajectory. This paper flips that: observers are strategic actors with limited defensive resources. The agent privately selects a goal and plans a path that deliberately misleads the observer about its true intention. This creates a dynamic asymmetric-information game where each side optimizes its strategy. The authors model this as a linear programming problem combined with the Double Oracle algorithm, which iteratively expands strategies to find equilibrium.

The key innovation is treating deception as a quantifiable metric. They define risk (probability of being caught) and deception effectiveness (how much the observer's allocation diverges from optimal). Numerical examples show the approach works efficiently. Accepted to American Control Conference 2026, this work has implications for drone swarms, military logistics, and cybersecurity where deception is critical. The linear programming formulation makes it computationally tractable for real-time applications.