In offline reinforcement learning (RL), addressing the out-of-distribution (OOD) action issue has been a focus, but we argue that there exists an OOD state issue that also impairs performance yet has been underexplored. Such an issue describes the scenario when the agent encounters states out of the offline dataset during the test phase, leading to uncontrolled behavior and performance degradation. To this end, we propose SCAS, a simple yet effective approach that unifies OOD state correction and OOD action suppression in offline RL. Technically, SCAS achieves value-aware OOD state correction, capable of correcting the agent from OOD states to high-value in-distribution states. Theoretical and empirical results show that SCAS also exhibits the effect of suppressing OOD actions. On standard offline RL benchmarks, SCAS achieves excellent performance without additional hyperparameter tuning. Moreover, benefiting from its OOD state correction feature, SCAS demonstrates enhanced robustness against environmental perturbations.

This paper introduces a novel hybrid architecture that enhances radar-based Dynamic Occupancy Grid Mapping (DOGM) for autonomous vehicles, integrating deep learning for state-classification. Traditional radar-based DOGM often faces challenges in accurately distinguishing between static and dynamic objects. Our approach addresses this limitation by introducing a neural network-based DOGM state correction mechanism, designed as a semantic segmentation task, to refine the accuracy of the occupancy grid. Additionally a heuristic fusion approach is proposed which allows to enhance performance without compromising on safety. We extensively evaluate this hybrid architecture on the NuScenes Dataset, focusing on its ability to improve dynamic object detection as well grid quality. The results show clear improvements in the detection capabilities of dynamic objects, highlighting the effectiveness of the deep learning-enhanced state correction in radar-based DOGM.