arXiv:2511.17150v1 Announce Type: new 
Abstract: Unlike discriminative approaches in autonomous driving that predict a fixed set of candidate trajectories of the ego vehicle, generative methods, such as diffusion models, learn the underlying distribution of future motion, enabling more flexible trajectory prediction. However, since these methods typically rely on denoising human-crafted trajectory anchors or random noise, there remains significant room for improvement. In this paper, we propose DiffRefiner, a novel two-stage trajectory prediction framework. The first stage uses a transformer-based Proposal Decoder to generate coarse trajectory predictions by regressing from sensor inputs using predefined trajectory anchors. The second stage applies a Diffusion Refiner that iteratively denoises and refines these initial predictions. In this way, we enhance the performance of diffusion-based planning by incorporating a discriminative trajectory proposal module, which provides strong guidance for the generative refinement process. Furthermore, we design a fine-grained denoising decoder to enhance scene compliance, enabling more accurate trajectory prediction through enhanced alignment with the surrounding environment. Experimental results demonstrate that DiffRefiner achieves state-of-the-art performance, attaining 87.4 EPDMS on NAVSIM v2, and 87.1 DS along with 71.4 SR on Bench2Drive, thereby setting new records on both public benchmarks. The effectiveness of each component is validated via ablation studies as well.

يقدم المقال DiffRefiner، وهو إطار عمل لتوقع المسارات على مرحلتين مصمم للقيادة الذاتية من البداية إلى النهاية. تولد المرحلة الأولى توقعات مسار خشنة باستخدام جهاز فك تشفير الاقتراحات المعتمد على المحولات، بينما تقوم المرحلة الثانية بتنقيح هذه التوقعات من خلال مُنقح الانتشار، مما يعزز مرونة ودقة تخطيط المسارات. يتناول هذا النهج القيود المفروضة على الأساليب التوليدية الحالية التي تعتمد على نقاط مرجعية محددة مسبقًا أو ضوضاء عشوائية.

El artículo presenta DiffRefiner, un marco de predicción de trayectorias en dos etapas diseñado para la conducción autónoma de extremo a extremo. La primera etapa genera predicciones de trayectorias gruesas utilizando un decodificador de propuestas basado en transformadores, mientras que la segunda etapa refina estas predicciones a través de un refinador de difusión, mejorando así la flexibilidad y precisión de la planificación de trayectorias. Este enfoque aborda las limitaciones de los métodos generativos actuales que dependen de anclajes predefinidos o ruido aleatorio.

L'article présente DiffRefiner, un cadre de prédiction de trajectoire en deux étapes conçu pour la conduite autonome de bout en bout. La première étape génère des prédictions de trajectoire grossières à l'aide d'un décodeur de proposition basé sur un transformateur, tandis que la seconde étape affine ces prédictions grâce à un affinement par diffusion, améliorant ainsi la flexibilité et la précision de la planification des trajectoires. Cette approche répond aux limites des méthodes génératives actuelles qui s'appuient sur des ancres prédéfinies ou du bruit aléatoire.

The paper introduces DiffRefiner, a two-stage trajectory prediction framework designed for end-to-end autonomous driving. The first stage generates coarse trajectory predictions using a transformer-based Proposal Decoder, while the second stage refines these predictions through a Diffusion Refiner, enhancing the flexibility and accuracy of trajectory planning. This approach addresses limitations in current generative methods that rely on predefined anchors or random noise.

DiffRefiner: Coarse to Fine Trajectory Planning via Diffusion Refinement with Semantic Interaction for End to End Autonomous Driving

arXiv:2511.18729v1 Announce Type: new 
Abstract: Driving planning is a critical component of end-to-end (E2E) autonomous driving. However, prevailing Imitative E2E Planners often suffer from multimodal trajectory mode collapse, failing to produce diverse trajectory proposals. Meanwhile, Generative E2E Planners struggle to incorporate crucial safety and physical constraints directly into the generative process, necessitating an additional optimization stage to refine their outputs. In this paper, we propose \textit{\textbf{GuideFlow}}, a novel planning framework that leverages Constrained Flow Matching. Concretely, \textit{\textbf{GuideFlow}} explicitly models the flow matching process, which inherently mitigates mode collapse and allows for flexible guidance from various conditioning signals. Our core contribution lies in directly enforcing explicit constraints within the flow matching generation process, rather than relying on implicit constraint encoding. Crucially, \textit{\textbf{GuideFlow}} unifies the training of the flow matching with the Energy-Based Model (EBM) to enhance the model's autonomous optimization capability to robustly satisfy physical constraints. Secondly, \textit{\textbf{GuideFlow}} parameterizes driving aggressiveness as a control signal during generation, enabling precise manipulation of trajectory style. Extensive evaluations on major driving benchmarks (Bench2Drive, NuScenes, NavSim and ADV-NuScenes) validate the effectiveness of \textit{\textbf{GuideFlow}}. Notably, on the NavSim test hard split (Navhard), \textit{\textbf{GuideFlow}} achieved SOTA with an EPDMS score of 43.0. The code will be released.

تم تقديم GuideFlow كإطار تخطيط جديد للقيادة الذاتية من النهاية إلى النهاية، حيث يعالج قيود المخططين E2E التقليديين والجيلين. من خلال استخدام مطابقة التدفق المقيد، يخفف GuideFlow بشكل فعال من انهيار الأنماط ويقوم بإدماج قيود السلامة والقيود الفيزيائية بشكل صريح في عملية توليد المسارات.

GuideFlow se ha presentado como un nuevo marco de planificación para la conducción autónoma de extremo a extremo, abordando las limitaciones de los planificadores E2E imitativos y generativos existentes. Al emplear la Correspondencia de Flujo Constricto, GuideFlow mitiga eficazmente el colapso de modos e incorpora restricciones de seguridad y físicas explícitas en el proceso de generación de trayectorias.

GuideFlow a été introduit comme un nouveau cadre de planification pour la conduite autonome de bout en bout, répondant aux limites des planificateurs E2E imitatifs et génératifs existants. En utilisant le Correspondance de Flux Contraint, GuideFlow atténue efficacement l'effondrement des modes et intègre des contraintes de sécurité et physiques explicites dans le processus de génération de trajectoires.

GuideFlow has been introduced as a novel planning framework for end-to-end autonomous driving, addressing the limitations of existing Imitative and Generative E2E Planners. By employing Constrained Flow Matching, GuideFlow effectively mitigates mode collapse and incorporates explicit safety and physical constraints into the trajectory generation process.

GuideFlow: Constraint-Guided Flow Matching for Planning in End-to-End Autonomous Driving

arXiv:2506.06659v3 Announce Type: replace-cross 
Abstract: Autonomous vehicles must navigate safely in complex driving environments. Imitating a single expert trajectory, as in regression-based approaches, usually does not explicitly assess the safety of the predicted trajectory. Selection-based methods address this by generating and scoring multiple trajectory candidates and predicting the safety score for each. However, they face optimization challenges in precisely selecting the best option from thousands of candidates and distinguishing subtle but safety-critical differences, especially in rare and challenging scenarios. We propose DriveSuprim to overcome these challenges and advance the selection-based paradigm through a coarse-to-fine paradigm for progressive candidate filtering, a rotation-based augmentation method to improve robustness in out-of-distribution scenarios, and a self-distillation framework to stabilize training. DriveSuprim achieves state-of-the-art performance, reaching 93.5% PDMS in NAVSIM v1 and 87.1% EPDMS in NAVSIM v2 without extra data, with 83.02 Driving Score and 60.00 Success Rate on the Bench2Drive benchmark, demonstrating superior planning capabilities in various driving scenarios.

تم تقديم DriveSuprim كنهج جديد لتحسين اختيار المسارات للمركبات المستقلة، حيث يتناول التحديات المتعلقة بالتنقل بأمان في بيئات القيادة المعقدة. تستخدم هذه الطريقة نموذجًا من الخشن إلى الدقيق لتصفية المرشحين وتدمج تحسينات قائمة على الدوران لتعزيز المتانة في السيناريوهات النادرة.

DriveSuprim se ha presentado como un enfoque novedoso para mejorar la selección de trayectorias para vehículos autónomos, abordando los desafíos de navegar de manera segura en entornos de conducción complejos. Este método emplea un paradigma de filtrado de candidatos de grueso a fino e incorpora una augmentación basada en rotación para mejorar la robustez en escenarios raros.

DriveSuprim a été introduit comme une nouvelle approche pour améliorer la sélection de trajectoires pour les véhicules autonomes, abordant les défis de la navigation en toute sécurité dans des environnements de conduite complexes. Cette méthode utilise un paradigme grossier à fin pour le filtrage des candidats et intègre une augmentation basée sur la rotation pour améliorer la robustesse dans des scénarios rares.

DriveSuprim has been introduced as a novel approach to enhance trajectory selection for autonomous vehicles, addressing the challenges of safely navigating complex driving environments. This method employs a coarse-to-fine paradigm for candidate filtering and incorporates rotation-based augmentation to improve robustness in rare scenarios.

DiffRefiner: Coarse to Fine Trajectory Planning via Diffusion Refinement with Semantic Interaction for End to End Autonomous Driving

Was this article worth reading? Share it