Search papers, labs, and topics across Lattice.
This paper introduces RecoverFly, a failure-aware reinforcement learning framework designed to enhance the performance of end-to-end UAV vision-language-action (UAV-VLA) policies in aerial navigation tasks. By employing token-level RL and a two-stage long-tail scene curriculum, RecoverFly effectively addresses issues of sample inefficiency and policy distribution shifts, leading to improved corrective learning and scene adaptation. Experimental results on the TravelUAV benchmark show that RecoverFly outperforms previous methods, achieving significant improvements in success rates across various test conditions with a reduced training budget.
RecoverFly boosts UAV navigation success rates by up to 8.37 percentage points while using only 30% of the training data, showcasing its efficiency and robustness.
Unmanned aerial vehicle vision-language navigation (UAV-VLN) requires agents to translate visual observations and language instructions into reliable flight actions in complex environments. Although recent end-to-end UAV vision-language-action (UAV-VLA) policies reduce reliance on separately designed perception, planning, and control modules, their behavior-cloning objectives provide limited corrective supervision for interactive closed-loop execution. Reinforcement learning (RL) offers a promising solution, while its effectiveness is constrained by inefficient use of samples, long-tailed scene distributions, and policy distribution shift during optimization. To this end, we propose RecoverFly, a failure-aware RL post-training framework for end-to-end UAV-VLA policies. Specifically, RecoverFly adapts token-level RL for stable optimization of grammar-constrained autoregressive UAV actions, revisits unresolved failure cases to strengthen corrective learning and sample utilization, and combines a two-stage long-tail scene curriculum with reference-policy regularization to improve scene adaptation while preserving acquired capabilities. Experiments on the TravelUAV benchmark demonstrate that RecoverFly achieves the best performance on the seen, unseen-map, and unseen-object splits. Moreover, compared to the AerialVLA initialization, RecoverFly improves success rate by 3.12 to 8.37 percentage points under a total rollout budget of about 30\% of the training-set size, validating its effectiveness, robustness, and generalization capabilities.
