🤖 AI Summary
This work addresses the challenge of reliably inferring latent environmental states and their associated uncertainties in reinforcement learning under partially observable or adversarially perturbed observations. It introduces, for the first time, adversarial observation perturbations that satisfy likelihood consistency constraints into linear probabilistic state-space models. By combining Bayesian inference with constrained optimization, the study analyzes how such perturbations affect latent state estimation and subsequent policy decisions. The research elucidates the propagation pathways through which observation perturbations influence both latent state beliefs and policy outputs, enabling the development of a more robust reinforcement learning framework. This approach provides a theoretical foundation and an effective design paradigm for safety-critical applications—such as robotics—to withstand sensor noise, failures, and adversarial attacks.
📝 Abstract
Decision-making under partial or adversarial observability requires accurate inference of the environment's latent state and its associated uncertainty. This work analyses adversarial attacks on linear probabilistic state-space models, commonly integrated within reinforcement learning architectures, where the attacker alters observations under likelihood constraints that ensure the perturbations remains consistent. We analyze how such adversarial yet realistic observation shifts influence the latent state and influence policy decisions. This perspective provides a principled pathway toward building more robust reinforcement learning systems, with direct relevance to safety-critical domains such as robotics, where reliable operation under sensor noise, partial failures, and adversarial conditions is essential.