Deceptive Path Planning Against Learnable Observers

Researchers introduce Repeated Deceptive Path Planning (RDPP), a new formulation addressing the limitation of existing deceptive path planning (DPP) methods that assume static, non-learning observers. In real-world scenarios like critical goods transportation or military operations, adversaries can adapt by learning from historical trajectories. The study shows that current DPP methods fail under learnable observers as they cannot adapt to evolving predictions. Incremental updates cause accumulative lag, degrading deception. To solve this, the authors propose Deceptive Meta Planning (DeMP), a two-level optimization framework combining episode-level adaptation for short-term policy adjustment against updated observers.