Curriculum-Adapted Robust Reinforcement Learning for UAV Deconfliction in Adversarial Environments
Researchers have developed a new curriculum-guided adaptation framework for reinforcement learning (RL) in autonomous UAVs. This approach aims to improve the robustness of UAV navigation against adversarial attacks, such as GNSS spoofing, which can corrupt value estimation and degrade performance. By progressively exposing the RL policy to increasing intensities of adversarial perturbations and aligning temporal-difference error distributions, the framework enhances transferability across different attack conditions. Evaluations in simulated UAV deconfliction environments demonstrated significant improvements in mission success rates and rewards compared to standard and existing robust RL baselines, even against previously unseen attack types. AI
IMPACT Enhances UAV resilience against cyberattacks, potentially improving safety and reliability in critical infrastructure and defense applications.