Autonomous Driving Comparison Hub
17 papers - avg viability 6.3
Recent advancements in autonomous driving are increasingly focused on enhancing exploration and adaptability in vehicle decision-making systems. New frameworks like Curious-VLA and SAMoE-VLA are addressing the limitations of traditional imitation learning by incorporating diverse data sampling and scene-adaptive expert selection, which improve performance in complex driving environments. Additionally, the integration of natural language processing through modules like Talk2DM is facilitating more intuitive human-vehicle interactions, potentially transforming how drivers communicate with autonomous systems. The introduction of datasets such as RAID and ADAS-TO is further enriching the understanding of driver behavior and risk perception, crucial for developing safer autonomous systems. Meanwhile, techniques like CycleBEV are refining the transformation of visual data into actionable insights, enhancing semantic understanding from various viewpoints. Collectively, these developments indicate a shift towards more robust, context-aware autonomous driving solutions that promise to address critical commercial challenges, including safety and user experience in real-world applications.
Top Papers
- Devil is in Narrow Policy: Unleashing Exploration in Driving VLA Models(8.0)
Curious-VLA unlocks the exploratory potential of autonomous driving models by addressing the exploit-explore dilemma, achieving state-of-the-art results on the Navsim benchmark.
- Talk2DM: Enabling Natural Language Querying and Commonsense Reasoning for Vehicle-Road-Cloud Integrated Dynamic Maps with Large Language Models(8.0)
Talk2DM offers an advanced natural language interface to enhance vehicle-road-cloud dynamic map interaction for autonomous driving systems.
- SAMoE-VLA: A Scene Adaptive Mixture-of-Experts Vision-Language-Action Model for Autonomous Driving(8.0)
SAMoE-VLA is a scene-adaptive Vision-Language-Action model for autonomous driving that outperforms existing approaches with fewer parameters, offering a safer and more efficient driving experience.
- $M^2$-Occ: Resilient 3D Semantic Occupancy Prediction for Autonomous Driving with Incomplete Camera Inputs(8.0)
M^2-Occ enhances 3D semantic occupancy prediction for autonomous driving by effectively handling incomplete camera inputs.
- Driving on Registers(8.0)
DrivoR is a transformer-based autonomous driving system offering efficient, adaptive, end-to-end driving with high benchmark performance.
- Towards Driver Behavior Understanding: Weakly-Supervised Risk Perception in Driving Scenes(8.0)
A driving risk perception dataset and weakly-supervised framework to identify potential risk sources, enabling safer autonomous driving systems.
- Perception-Aware Multimodal Spatial Reasoning from Monocular Images(8.0)
Enhance autonomous driving spatial reasoning by equipping VLMs with object-centric grounding using visual reference tokens and multimodal chain-of-thought, outperforming existing methods on the SURDS benchmark.
- StyleVLA: Driving Style-Aware Vision Language Action Model for Autonomous Driving(8.0)
StyleVLA is a physics-informed Vision Language Action model that generates diverse and plausible driving behaviors tailored to individual driving styles.
- ADAS-TO: A Large-Scale Multimodal Naturalistic Dataset and Empirical Characterization of Human Takeovers during ADAS Engagement(7.0)
A dataset of autonomous driving takeovers paired with visual cues enables the development of early warning systems for safer transitions.
- CycleBEV: Regularizing View Transformation Networks via View Cycle Consistency for Bird's-Eye-View Semantic Segmentation(7.0)
Deploy CycleBEV to enhance existing VT models for improved BEV semantic segmentation in autonomous driving applications.