Diffusion-based generative models are emerging as powerful tools for long-horizon planning in reinforcement learning (RL), particularly with offline datasets. However, their performance is fundamentally limited by the quality and diversity of training data. This often restricts their generalization to tasks outside their training distribution or longer planning horizons. To overcome this challenge, we propose State-Covering Trajectory Stitching (SCoTS), a novel reward-free trajectory augmentation method that incrementally stitches together short trajectory segments, systematically generating diverse and extended trajectories. SCoTS first learns a temporal distance-preserving latent representation that captures the underlying temporal structure of the environment, then iteratively stitches trajectory segments guided by directional exploration and novelty to effectively cover and expand this latent space. We demonstrate that SCoTS significantly improves the performance and generalization capabilities of diffusion planners on offline goal-conditioned benchmarks requiring stitching and long-horizon reasoning. Furthermore, augmented trajectories generated by SCoTS significantly improve the performance of widely used offline goal-conditioned RL algorithms across diverse environments. Our code is available at https://github.com/leekwoon/scots/

Diffusion models have demonstrated strong capabilities for modeling humanlike driving behaviors in autonomous driving, but their iterative sampling process induces substantial latency, and operating directly on raw trajectory points forces the model to spend capacity on low-level kinematics, rather than high-level multi-modal semantics. To address these limitations, we propose LAtent Planner (LAP), a framework that plans in a V AE-learned latent space that disentangles high-level intents from low-level kinematics, enabling our planner to capture rich, multi-modal driving strategies. We further introduce a fine-grained feature distillation mechanism to guide a better interaction and fusion between the high-level semantic planning space and the vectorized scene context. Notably, LAP can produce high-quality plans in one single denoising step, substantially reducing computational overhead. Through extensive evaluations on the large-scale nuPlan benchmark, LAP achieves state-of-the-art closed-loop performance among learning-based planning methods, while demonstrating an inference speedup of at most 10 over previous SOT A approaches. A central challenge is handling the inherent uncertainty and behavioral multimodality of real-world traffic, where multiple distinct yet equally plausible maneuvers may be available (Y ang et al., 2023; Xiao et al., 2020). While early rule-based systems offered interpretability, their hand-crafted logic is brittle and fails to scale to the long-tail of open-world scenarios (Fan et al., 2018; Chen et al., 2024). Consequently, the field has shifted towards data-driven Imitation Learning (IL), which excels at capturing nuanced, human-like behaviors from large-scale datasets (Le Mero et al., 2022; Teng et al., 2022). However, the standard IL objective is notoriously susceptible to mode-averaging, where the model collapses multiple valid expert trajectories into a single, physically infeasible path, fundamentally failing to represent the multi-modal nature of human decision-making (Strohbeck et al., 2020). To overcome this limitation, Denoising Diffusion Probabilistic Models(DDPMs) have emerged as a powerful tool for modeling complex, multi-modal distributions (Liao et al., 2025; Ho et al., 2020). However, existing approaches models directly to raw trajectory waypoints are both computationally inefficient and conceptually flawed. This mirrors the core challenge of early image synthesis: operating in a high-dimensional pixel space expends vast model capacity on low-level details over high-level semantics (Rombach et al., 2022).

Training a policy with online rollouts can be costly, dangerous, and sample-inefficient [1]. Alternatively, offline reinforcement learning (RL) involves a policy trained exclusively with pre-collected data. Extracting effective polices without exploration or feedback from the environment is challenging for conventional off-policy and even specialized offline RL algorithms [2, 3]. Approaches to of-fline RL are also frequently faced with the problem of incomplete or undirected demonstrations [4, 5, 6]. Offline algorithms must compose sub-trajectories from training data to generate advantageous behaviors. Another challenge is high-dimensionality and long horizons, which make accurate planning and behavior cloning difficult [1]. Finally, sparse rewards pose a challenge to many training algorithms as they hinder accurate credit assignment to actions [7]. Diffusion models have emerged as a powerful framework for expressing complex, multi-modal distributions [8, 9]. Leveraging this model class, diffusion policies generate high fidelity actions and use a value function for action selection [10, 11, 12].

Diffusion models have recently gained prominence in offline reinforcement learning due to their ability to effectively learn high-performing, generalizable policies from static datasets. Diffusion-based planners facilitate long-horizon decision-making by generating high-quality trajectories through iterative denoising, guided by return-maximizing objectives. However, existing guided sampling strategies such as Classifier Guidance, Classifier-Free Guidance, and Monte Carlo Sample Selection either produce suboptimal multi-modal actions, struggle with distributional drift, or incur prohibitive inference-time costs. To address these challenges, we propose Prior Guidance (PG), a novel guided sampling framework that replaces the standard Gaussian prior of a behavior-cloned diffusion model with a learnable distribution, optimized via a behavior-regularized objective. PG directly generates high-value trajectories without costly reward optimization of the diffusion model itself, and eliminates the need to sample multiple candidates at inference for sample selection. We present an efficient training strategy that applies behavior regularization in latent space, and empirically demonstrate that PG outperforms state-of-the-art diffusion policies and planners across diverse long-horizon offline RL benchmarks.Our code is available at https://github.com/ku-dmlab/PG.

Diffusion-based generative models are emerging as powerful tools for long-horizon planning in reinforcement learning (RL), particularly with offline datasets. However, their performance is fundamentally limited by the quality and diversity of training data. This often restricts their generalization to tasks outside their training distribution or longer planning horizons. To overcome this challenge, we propose State-Covering Trajectory Stitching (SCoTS), a novel reward-free trajectory augmentation method that incrementally stitches together short trajectory segments, systematically generating diverse and extended trajectories. SCoTS first learns a temporal distance-preserving latent representation that captures the underlying temporal structure of the environment, then iteratively stitches trajectory segments guided by directional exploration and novelty to effectively cover and expand this latent space. We demonstrate that SCoTS significantly improves the performance and generalization capabilities of diffusion planners on offline goal-conditioned benchmarks requiring stitching and long-horizon reasoning. Furthermore, augmented trajectories generated by SCoTS significantly improve the performance of widely used offline goal-conditioned RL algorithms across diverse environments.

The most widely used techniques for learning dynamics models include autoregressive forward models (Deisenroth & Rasmussen, 2011; Hafner et al., 2019; Kaiser et al., 2020), which make predictions based on future time progression.

Recent studies have explored leveraging the world knowledge and cognitive capabilities of Vision-Language Models (VLMs) to address the long-tail problem in end-to-end autonomous driving. However, existing methods typically formulate trajectory planning as a language modeling task, where physical actions are output in the language space, potentially leading to issues such as format-violating outputs, infeasible actions, and slow inference speeds. In this paper, we propose ReCogDrive, a novel Reinforced Cognitive framework for end-to-end autonomous Driving, unifying driving understanding and planning by integrating an autoregressive model with a diffusion planner. First, to instill human driving cognition into the VLM, we introduce a hierarchical data pipeline that mimics the sequential cognitive process of human drivers through three stages: generation, refinement, and quality control. Building on this cognitive foundation, we then address the language-action mismatch by injecting the VLM's learned driving priors into a diffusion planner to efficiently generate continuous and stable trajectories. Furthermore, to enhance driving safety and reduce collisions, we introduce a Diffusion Group Relative Policy Optimization (DiffGRPO) stage, reinforcing the planner for enhanced safety and comfort. Extensive experiments on the NAVSIM and Bench2Drive benchmarks demonstrate that ReCogDrive achieves state-of-the-art performance. Additionally, qualitative results across diverse driving scenarios and DriveBench highlight the model's scene comprehension. All code, model weights, and datasets will be made publicly available to facilitate subsequent research.

Learning-based planners are sensitive to the long-tailed distribution of driving data. Common maneuvers dominate datasets, while dangerous or rare scenarios are sparse. This imbalance can bias models toward the frequent cases and degrade performance on critical scenarios. To tackle this problem, we compare balancing strategies for sampling training data and find reweighting by trajectory pattern an effective approach. We then present FlowDrive, a flow-matching trajectory planner that learns a conditional rectified flow to map noise directly to trajectory distributions with few flow-matching steps. We further introduce moderated, in-the-loop guidance that injects small perturbation between flow steps to systematically increase trajectory diversity while remaining scene-consistent. On nuPlan and the interaction-focused interPlan benchmarks, FlowDrive achieves state-of-the-art results among learning-based planners and approaches methods with rule-based refinements. After adding moderated guidance and light post-processing (FlowDrive*), it achieves overall state-of-the-art performance across nearly all benchmark splits.

Diffusion-based planners have gained significant recent attention for their robustness and performance in long-horizon tasks. However, most existing planners rely on a fixed, pre-specified horizon during both training and inference. This rigidity often produces length-mismatch (trajectories that are too short or too long) and brittle performance across instances with varying geometric or dynamical difficulty. In this paper, we introduce the Variable Horizon Diffuser (VHD) framework, which treats the horizon as a learned variable rather than a fixed hyperparameter. Given a start-goal pair, we first predict an instance-specific horizon using a learned Length Predictor model, which guides a Diffusion Planner to generate a trajectory of the desired length. Our design maintains compatibility with existing diffusion planners by controlling trajectory length through initial noise shaping and training on randomly cropped sub-trajectories, without requiring architectural changes. Empirically, VHD improves success rates and path efficiency in maze-navigation and robot-arm control benchmarks, showing greater robustness to horizon mismatch and unseen lengths, while keeping training simple and offline-only.