Nuscenes
PulseAugur coverage of Nuscenes — every cluster mentioning Nuscenes across labs, papers, and developer communities, ranked by signal.
-
HorizonDrive enables minute-scale driving simulation with self-correction
Researchers have developed HorizonDrive, a novel framework for autoregressive driving simulation that enables minute-scale rollouts with bounded memory. This approach trains a teacher model to recover from its own predi…
-
Co-Fusion4D framework boosts 3D object detection for autonomous driving
Researchers have developed Co-Fusion4D, a new framework designed to improve 3D object detection for autonomous driving by addressing spatiotemporal inconsistencies. The system uses a current-frame-centric approach that …
-
New VLMs boost autonomous driving efficiency and spatial reasoning
Researchers are developing advanced Vision-Language Models (VLMs) for autonomous driving, focusing on improving efficiency and spatial reasoning. New methods like Fast-dDrive aim to balance high-fidelity trajectory plan…
-
New metrics improve evaluation of autonomous driving map estimation
Researchers have developed new evaluation metrics, SOSPA and PLD, to more accurately assess online mapping systems used in autonomous driving. These metrics address limitations in current methods like Chamfer Distance a…
-
Fudan and Shanghai Jiao Tong propose spatial memory for autonomous driving
Researchers from Fudan University and Shanghai Jiao Tong University have developed a novel approach for autonomous driving that incorporates a "spatial memory" by retrieving historical geographic information. This metho…
-
Driving VLAs grounded with inverse kinematics achieve SOTA performance
Researchers have developed a new method for grounding driving vision-language models (VLAs) by reframing trajectory prediction as an inverse kinematics problem. This approach requires both current and future visual stat…
-
Language priors boost unsupervised 3D point cloud segmentation
Researchers have developed LangTail, a new framework designed to improve unsupervised 3D point cloud segmentation by addressing the issue of long-tail ambiguity. This problem occurs when minor object classes are overloo…
-
New HEAT model improves autonomous driving across diverse environments
Researchers have developed a new trajectory-guided learning paradigm called HEAT for end-to-end autonomous driving systems. This approach aims to improve performance across diverse and heterogeneous driving environments…
-
WeatherOcc3D uses VLM to improve 3D prediction in bad weather
Researchers have developed a new framework called WeatherOcc3D that uses Visual-Language Models (VLMs) to improve 3D semantic occupancy prediction in adverse weather conditions. The system leverages CLIP's latent space …
-
CLOVER framework enhances autonomous driving planning with closed-loop value estimation
Researchers have developed CLOVER, a novel framework designed to improve end-to-end autonomous driving planning systems. This approach addresses the common training-evaluation mismatch by generating diverse candidate tr…
-
New method improves HD map construction with cross-view supervision
Researchers have developed a new method called Cross-View Supervision (CVS) to improve the construction of high-definition maps using bird's-eye-view (BEV) representations from multiple cameras. Traditional methods stru…
-
Random-Set GNNs enhance uncertainty quantification in graph learning
Researchers have introduced Random-Set Graph Neural Networks (RS-GNNs) to address uncertainty quantification in graph learning. This new framework models node-level epistemic uncertainty using a belief function formalis…
-
Driving models' performance hinges on temporal sampling frequency
Researchers have investigated the impact of temporal sampling frequency on end-to-end autonomous driving trajectory prediction models. They found that while dense frame sampling is often assumed to improve performance, …
-
AI research advances autonomous driving perception and safety
Researchers are developing advanced AI techniques to improve autonomous driving systems. One approach, CaAD, focuses on causality-aware end-to-end modeling to better predict vehicle and agent interactions, showing stron…
-
New neuro-symbolic architecture improves autonomous driving scene understanding
Researchers have developed InfoCoordiBridge, a novel neuro-symbolic architecture designed to enhance the reliability of scene understanding in autonomous driving systems. This architecture addresses issues where languag…
-
BEVCALIB model uses bird's-eye view features for LiDAR-camera calibration
Researchers have developed BEVCALIB, a novel method for calibrating LiDAR and camera sensors, crucial for autonomous driving systems. This approach utilizes bird's-eye view (BEV) features extracted from both sensor type…
-
DynFlowDrive model enhances autonomous driving with flow-based dynamic world modeling
Researchers have introduced DynFlowDrive, a novel latent world model designed to enhance the reliability of autonomous driving systems. This model utilizes flow-based dynamics to predict future scene evolutions under va…
-
MapRF uses NeRF-guided self-training for weakly supervised HD map construction
Researchers have developed MapRF, a novel framework for constructing high-definition (HD) maps for autonomous driving systems using only 2D image labels. This weakly supervised approach leverages Neural Radiance Fields …
-
SimPB++ model unifies 2D and 3D object detection for autonomous driving
Researchers have developed SimPB++, an end-to-end model designed to simultaneously detect both 2D objects in perspective views and 3D objects in a bird's-eye view for multi-camera autonomous driving systems. The model e…
-
LiDAR-only HD map construction method enhances semantic cues via knowledge distillation
Researchers have developed LIE, a novel method for constructing High-Definition (HD) maps for autonomous driving using only LiDAR data. This approach overcomes the limitations of camera-based methods by leveraging knowl…