Tag
This paper proposes CRHT, a Continuous Regression Hybrid Transformer for vessel trajectory prediction using AIS data, featuring an online K-means cluster sampling strategy and a CNN-Transformer hybrid architecture to address geographic bias and improve short-term forecasting accuracy.
This paper introduces Hybrid Probabilistic Zonotopes (HProbZ), a neural network output head that jointly models discrete modes, bounded drift, and stochastic noise with closed-form likelihood, enabling identifiable uncertainty decomposition, observation-driven contraction, and multi-modal conformal coverage.
INTraJ is a unified framework for trajectory prediction that decomposes social influence into two stages: planning with future social information and local reaction from residuals, achieving state-of-the-art results on Argoverse 2, ETH/UCY, and SDD benchmarks.
This thesis introduces polynomial representations for long-term traffic scene prediction in autonomous driving, showing improved computational efficiency, generalization, and prediction plausibility over sequence-based baselines, validated on Argoverse 2 and Waymo Open datasets.
DreamTraj predicts 6-DoF object trajectories from a single RGB image and a language instruction by decoding internal video diffusion latents, eliminating the need for video, depth, or CAD models at inference. It introduces the MOVEdataset with fine-grained language-to-motion annotations and achieves state-of-the-art performance while running 4.6x faster than generate-then-extract pipelines.
This paper proposes the Conditional Informer, a Transformer-based encoder-decoder for ship trajectory prediction that conditions vessel dynamics on environmental contexts via a Conditional Attention mechanism, achieving 15.4% accuracy improvement and robust fallback via Modality Masking.
This paper introduces TrajRS, an extension of Randomized Smoothing that provides certified robust radii for pedestrian trajectory predictors, offering verifiable safety guarantees against adversarial perturbations.
This paper introduces a method called 'Learn to Cluster' to quantify and interpret social interactions among pedestrians for better trajectory prediction. It uses probabilistic latent variable generative learning to cluster social interactions without labels, improving robustness for autonomous driving and social robots.
This paper introduces M-CTX, an exact and scalable spatial context retrieval framework for trajectory analytics that reduces context construction time from 17 CPU-days to 1.8 hours on a 5.48M-anchor maritime corpus, by replacing brute-force stages with index-backed operators.
This paper compares 19 graph neural network layer types for modelling interactions in driving trajectory prediction, finding ARMA, Chebyshev, and topology-aware layers most effective and offering design principles for better prediction models.
Qwen-VLA is a unified vision-language-action model for embodied decision-making, integrating manipulation, navigation, and trajectory prediction across different robot platforms. It uses a DiT-based action decoder and embodiment-aware prompt conditioning, achieving strong performance and out-of-distribution generalization.
SleepWalk is a three-tier benchmark for evaluating vision-language models' ability to predict spatially coherent trajectories in 3D environments from textual instructions and visual observations, revealing systematic failures in grounded spatial reasoning under occlusions and multi-step instructions.
OneVL is a unified vision-language-action framework that compresses chain-of-thought reasoning into latent tokens supervised by both language and visual world model decoders, achieving state-of-the-art trajectory prediction accuracy for autonomous driving at answer-only inference latency. It is the first latent CoT method to surpass explicit CoT across four benchmarks.