Qichao Zhang

Yupeng Zheng, Chengliang Zhong, Pengfei Li et al.

Self-supervised depth estimation draws a lot of attention recently as it can promote the 3D sensing capabilities of self-driving vehicles. However, it intrinsically relies upon the photometric consistency assumption, which hardly holds during nighttime. Although various supervised nighttime image enhancement methods have been proposed, their generalization performance in challenging driving scenarios is not satisfactory. To this end, we propose the first method that jointly learns a nighttime image enhancer and a depth estimator, without using ground truth for either task. Our method tightly entangles two self-supervised tasks using a newly proposed uncertain pixel masking strategy. This strategy originates from the observation that nighttime images not only suffer from underexposed regions but also from overexposed regions. By fitting a bridge-shaped curve to the illumination map distribution, both regions are suppressed and two tasks are bridged naturally. We benchmark the method on two established datasets: nuScenes and RobotCar and demonstrate state-of-the-art performance on both of them. Detailed ablations also reveal the mechanism of our proposal. Last but not least, to mitigate the problem of sparse ground truth of existing datasets, we provide a new photo-realistically enhanced nighttime dataset based upon CARLA. It brings meaningful new challenges to the community. Codes, data, and models are available at https://github.com/ucaszyp/STEPS.

Minghui Jia, Qichao Zhang, Ali Luo et al.

Due to the limited generalization and interpretability of deep learning classifiers, The final vetting of rare celestial object candidates still relies on expert visual inspection--a manually intensive process. In this process, astronomers leverage specialized tools to analyze spectra and construct reliable catalogs. However, this practice has become the primary bottleneck, as it is fundamentally incapable of scaling with the data deluge from modern spectroscopic surveys. To bridge this gap, we propose Spec-o3, a tool-augmented vision-language agent that performs astronomer-aligned spectral inspection via interleaved multimodal chain-of-thought reasoning. Spec-o3 is trained with a two-stage post-training recipe: cold-start supervised fine-tuning on expert inspection trajectories followed by outcome-based reinforcement learning on rare-type verification tasks. Evaluated on five rare-object identification tasks from LAMOST, Spec-o3 establishes a new State-of-the-Art, boosting the macro-F1 score from 28.3 to 76.5 with a 7B parameter base model and outperforming both proprietary VLMs and specialized deep models. Crucially, the agent demonstrates strong generalization to unseen inspection tasks across survey shifts (from LAMOST to SDSS/DESI). Expert evaluations confirm that its reasoning traces are coherent and physically consistent, supporting transparent and trustworthy decision-making. Code, data, and models are available at https://github.com/Maxwell-Jia/spec-o3.

Yaocheng Zhang, Jiajun Chai, Songjun Tu et al.

On-policy distillation (OPD) provides dense teacher feedback along rollouts generated by the student and has emerged as a promising post-training paradigm for long-horizon reasoning. However, standard OPD typically generates full rollouts during training, which is computationally expensive and may expose the student to unreliable teacher feedback at late rollout positions, especially during early training. We identify the rollout horizon as a key bottleneck in OPD that substantially impacts training efficiency. Unlike Reinforcement Learning with Verifiable Rewards (RLVR), OPD does not require a complete trajectory or a final answer reward to provide learning signals. This observation suggests that full rollouts may not always be necessary for effective OPD. Motivated by this insight, we propose two simple horizon-control strategies: Progressive OPD (POPD), which gradually expands the rollout horizon during training, and Truncated OPD (TOPD), which permanently performs distillation on reliable truncated rollouts. Experiments on mathematical reasoning show that POPD improves the training efficiency of OPD by up to 3$\times$, while TOPD matches OPD performance using only 10\% of the rollout horizon, leading to substantial wall-clock and memory reductions. These results demonstrate that controlling the rollout horizon offers a simple and practical path to more efficient OPD.

Ding Li, Qichao Zhang, Shuai Lu et al.

This paper aims to tackle the interactive behavior prediction task, and proposes a novel Conditional Goal-oriented Trajectory Prediction (CGTP) framework to jointly generate scene-compliant trajectories of two interacting agents. Our CGTP framework is an end to end and interpretable model, including three main stages: context encoding, goal interactive prediction and trajectory interactive prediction. First, a Goals-of-Interest Network (GoINet) is designed to extract the interactive features between agent-to-agent and agent-to-goals using a graph-based vectorized representation. Further, the Conditional Goal Prediction Network (CGPNet) focuses on goal interactive prediction via a combined form of marginal and conditional goal predictors. Finally, the Goaloriented Trajectory Forecasting Network (GTFNet) is proposed to implement trajectory interactive prediction via the conditional goal-oriented predictors, with the predicted future states of the other interacting agent taken as inputs. In addition, a new goal interactive loss is developed to better learn the joint probability distribution over goal candidates between two interacting agents. In the end, the proposed method is conducted on Argoverse motion forecasting dataset, In-house cut-in dataset, and Waymo open motion dataset. The comparative results demonstrate the superior performance of our proposed CGTP model than the mainstream prediction methods.

Junli Wang, Xueyi Liu, Yinan Zheng et al.

Generative models have shown great potential in trajectory planning. Recent studies demonstrate that anchor-guided generative models are effective in modeling the uncertainty of driving behaviors and improving overall performance. However, these methods rely on discrete anchor vocabularies that must sufficiently cover the trajectory distribution during testing to ensure robustness, inducing an inherent trade-off between vocabulary size and model performance. To overcome this limitation, we propose MeanFuser, an end-to-end autonomous driving method that enhances both efficiency and robustness through three key designs. (1) We introduce Gaussian Mixture Noise (GMN) to guide generative sampling, enabling a continuous representation of the trajectory space and eliminating the dependency on discrete anchor vocabularies. (2) We adapt ``MeanFlow Identity" to end-to-end planning, which models the mean velocity field between GMN and trajectory distribution instead of the instantaneous velocity field used in vanilla flow matching methods, effectively eliminating numerical errors from ODE solvers and significantly accelerating inference. (3) We design a lightweight Adaptive Reconstruction Module (ARM) that enables the model to implicitly select from all sampled proposals or reconstruct a new trajectory when none is satisfactory via attention weights. Experiments on the NAVSIM closed-loop benchmark demonstrate that MeanFuser achieves outstanding performance without the supervision of the PDM Score. and exceptional inference efficiency, offering a robust and efficient solution for end-to-end autonomous driving. Our code and model are available at https://github.com/wjl2244/MeanFuser.

Songjun Tu, Chengdong Xu, Qichao Zhang et al.

Agentic reinforcement learning (RL) can benefit substantially from reusable experience, yet existing skill-based methods mainly extract trajectory-level guidance and often lack principled mechanisms for maintaining an evolving skill memory. We propose D2Skill, a dynamic dual-granularity skill bank for agentic RL that organizes reusable experience into task skills for high-level guidance and step skills for fine-grained decision support and error correction. D2Skill jointly trains the policy and skill bank through paired baseline and skill-injected rollouts under the same policy, using their performance gap to derive hindsight utility signals for both skill updating and policy optimization. Built entirely from training-time experience, the skill bank is continuously expanded through reflection and maintained with utility-aware retrieval and pruning. Experiments on ALFWorld and WebShop with Qwen2.5-7B-Instruct and Qwen3-4B-Instruct-2507 show that D2Skill consistently improves success rates over skill-free baselines by 10-20 points. Further ablations and analyses show that both dual-granularity skill modeling and dynamic skill maintenance are critical to these gains, while the learned skills exhibit higher utility, transfer across evaluation settings, and introduce only modest training overhead.

Junjie Wang, Yao Mu, Dong Li et al.

The latent world model provides a promising way to learn policies in a compact latent space for tasks with high-dimensional observations, however, its generalization across diverse environments with unseen dynamics remains challenging. Although the recurrent structure utilized in current advances helps to capture local dynamics, modeling only state transitions without an explicit understanding of environmental context limits the generalization ability of the dynamics model. To address this issue, we propose a Prototypical Context-Aware Dynamics (ProtoCAD) model, which captures the local dynamics by time consistent latent context and enables dynamics generalization in high-dimensional control tasks. ProtoCAD extracts useful contextual information with the help of the prototypes clustered over batch and benefits model-based RL in two folds: 1) It utilizes a temporally consistent prototypical regularizer that encourages the prototype assignments produced for different time parts of the same latent trajectory to be temporally consistent instead of comparing the features; 2) A context representation is designed which combines both the projection embedding of latent states and aggregated prototypes and can significantly improve the dynamics generalization ability. Extensive experiments show that ProtoCAD surpasses existing methods in terms of dynamics generalization. Compared with the recurrent-based model RSSM, ProtoCAD delivers 13.2% and 26.7% better mean and median performance across all dynamics generalization tasks.

Ziyan Wang, Peng Chen, Ding Li et al. · cmu

Learning diverse and high-fidelity traffic simulations from human driving demonstrations is crucial for autonomous driving evaluation. The recent next-token prediction (NTP) paradigm, widely adopted in large language models (LLMs), has been applied to traffic simulation and achieves iterative improvements via supervised fine-tuning (SFT). However, such methods limit active exploration of potentially valuable motion tokens, particularly in suboptimal regions. Entropy patterns provide a promising perspective for enabling exploration driven by motion token uncertainty. Motivated by this insight, we propose a novel tokenized traffic simulation policy, R1Sim, which represents an initial attempt to explore reinforcement learning based on motion token entropy patterns, and systematically analyzes the impact of different motion tokens on simulation outcomes. Specifically, we introduce an entropy-guided adaptive sampling mechanism that focuses on previously overlooked motion tokens with high uncertainty yet high potential. We further optimize motion behaviors using Group Relative Policy Optimization (GRPO), guided by a safety-aware reward design. Overall, these components enable a balanced exploration-exploitation trade-off through diverse high-uncertainty sampling and group-wise comparative estimation, resulting in realistic, safe, and diverse multi-agent behaviors. Extensive experiments on the Waymo Sim Agent benchmark demonstrate that R1Sim achieves competitive performance compared to state-of-the-art methods.

Yaocheng Zhang, Yuanheng Zhu, Wenyue Chong et al.

Deep search agents have emerged as a promising paradigm for addressing complex information-seeking tasks, but their training remains challenging due to sparse rewards, weak credit assignment, and limited labeled data. Self-play offers a scalable route to reduce data dependence, but conventional self-play optimizes students only through sparse outcome rewards, leading to low learning efficiency. In this work, we observe that self-play naturally produces a question construction path (QCP) during task generation, an intermediate artifact that captures the reverse solution process. This reveals a new source of privileged information for self-distillation: self-play can itself provide high-quality privileged context for the teacher model in a low-cost and scalable manner, without relying on human feedback or curated privileged information. Leveraging this insight, we propose Privileged Information Self-Play ($π$-Play), a multi-agent self-evolution framework. In $π$-Play, an examiner generates tasks together with their QCPs, and a teacher model leverages QCP as privileged context to densely supervise a student via self-distillation. This design transforms conventional sparse-reward self-play into a dense-feedback self-evolution loop. Extensive experiments show that data-free $π$-Play surpasses fully supervised search agents and improves evolutionary efficiency by 2-3$\times$ over conventional self-play.

Zebin Xing, Yupeng Zheng, Qichao Zhang et al.

End-to-end autonomous driving has emerged as a pivotal direction in the field of autonomous systems. Recent works have demonstrated impressive performance by incorporating high-level guidance signals to steer low-level trajectory planners. However, their potential is often constrained by inaccurate high-level guidance and the computational overhead of complex guidance modules. To address these limitations, we propose Mimir, a novel hierarchical dual-system framework capable of generating robust trajectories relying on goal points with uncertainty estimation: (1) Unlike previous approaches that deterministically model, we estimate goal point uncertainty with a Laplace distribution to enhance robustness; (2) To overcome the slow inference speed of the guidance system, we introduce a multi-rate guidance mechanism that predicts extended goal points in advance. Validated on challenging Navhard and Navtest benchmarks, Mimir surpasses previous state-of-the-art methods with a 20% improvement in the driving score EPDMS, while achieving 1.6 times improvement in high-level module inference speed without compromising accuracy. The code and models will be released soon to promote reproducibility and further development. The code is available at https://github.com/ZebinX/Mimir-Uncertainty-Driving

Linbo Wang, Yupeng Zheng, Qiang Chen et al. · tsinghua

We introduce Latent-WAM, an efficient end-to-end autonomous driving framework that achieves strong trajectory planning through spatially-aware and dynamics-informed latent world representations. Existing world-model-based planners suffer from inadequately compressed representations, limited spatial understanding, and underutilized temporal dynamics, resulting in sub-optimal planning under constrained data and compute budgets. Latent-WAM addresses these limitations with two core modules: a Spatial-Aware Compressive World Encoder (SCWE) that distills geometric knowledge from a foundation model and compresses multi-view images into compact scene tokens via learnable queries, and a Dynamic Latent World Model (DLWM) that employs a causal Transformer to autoregressively predict future world status conditioned on historical visual and motion representations. Extensive experiments on NAVSIM v2 and HUGSIM demonstrate new state-of-the-art results: 89.3 EPDMS on NAVSIM v2 and 28.9 HD-Score on HUGSIM, surpassing the best prior perception-free method by 3.2 EPDMS with significantly less training data and a compact 104M-parameter model.

Jingbo Sun, Wenyue Chong, Songjun Tu et al.

Agentic retrieval-augmented generation (RAG) systems enable large language models (LLMs) to solve complex tasks through multi-step interaction with external retrieval tools. However, such multi-step interaction often involves redundant search steps, incurring substantial computational cost and latency. Prior work limits search depth (i.e., the number of search steps) to reduce cost, but this often leads to underexploration of complex questions. To address this, we first investigate how search depth affects accuracy and find a minimal sufficient search depth that defines an accuracy-efficiency trade-off, jointly determined by question complexity and the agent's capability. Furthermore, we propose AutoSearch, a reinforcement learning (RL) framework that evaluates each search step via self-generated intermediate answers. By a self-answering mechanism, AutoSearch identifies the minimal sufficient search depth and promotes efficient search by rewarding its attainment while penalizing over-searching. In addition, reward mechanisms are introduced to stabilize search behavior and improve answer quality on complex questions. Extensive experiments on multiple benchmarks show that AutoSearch achieves a superior accuracy-efficiency trade-off, alleviating over-searching while preserving search quality.

Yinfeng Gao, Qichao Zhang, Deqing Liu et al.

End-to-end autonomous driving policies based on Imitation Learning (IL) often struggle in closed-loop execution due to the misalignment between inadequate open-loop training objectives and real driving requirements. While Reinforcement Learning (RL) offers a solution by directly optimizing driving goals via reward signals, the rendering-based training environments introduce the rendering gap and are inefficient due to high computational costs. To overcome these challenges, we present a novel Pseudo-simulation-based RL method for closed-loop end-to-end autonomous driving, PerlAD. Based on offline datasets, PerlAD constructs a pseudo-simulation that operates in vector space, enabling efficient, rendering-free trial-and-error training. To bridge the gap between static datasets and dynamic closed-loop environments, PerlAD introduces a prediction world model that generates reactive agent trajectories conditioned on the ego vehicle's plan. Furthermore, to facilitate efficient planning, PerlAD utilizes a hierarchical decoupled planner that combines IL for lateral path generation and RL for longitudinal speed optimization. Comprehensive experimental results demonstrate that PerlAD achieves state-of-the-art performance on the Bench2Drive benchmark, surpassing the previous E2E RL method by 10.29% in Driving Score without requiring expensive online interactions. Additional evaluations on the DOS benchmark further confirm its reliability in handling safety-critical occlusion scenarios.

Yinfeng Gao, Deqing Liu, Qichao Zhang et al.

Current Vision-Language-Action (VLA) paradigms in end-to-end autonomous driving rely on offline training from static datasets, leaving them vulnerable to distribution shift. Recent post-training methods use takeover data to mitigate this by augmenting the dataset with high-quality expert takeover samples, yet they suffer from two key limitations: supervision restricted to the period after the takeover moments leads to policies with limited safety margins, and passive preference optimization lacks active exploration for optimal performance. In this paper, we propose TakeVLA, a novel VLA post-training framework that overcomes these shortcomings through two complementary innovations. First, we introduce pre-takeover language supervision, which allows the VLA to learn from mistakes proactively. By explicitly teaching the model about what to do in error-prone situations, we cultivate a precautionary mindset that anticipates hazards early and substantially enlarges safety margins. Second, we propose Scenario Dreaming, a reinforcement fine-tuning paradigm that operates in reconstruceted takeover scenarios, encouraging active exploration beyond mere preference fitting. Experiments on the Bench2Drive benchmark demonstrate that TakeVLA achieves state-of-the-art closed-loop performance, surpassing the strong VLA baseline SimLingo by 4.93 in driving score, with an enhanced safety margin as evidenced by an 11.76% increase in average TTC.

Yaocheng Zhang, Haohuan Huang, Zijun Song et al.

Tool-Integrated Reasoning (TIR) with search engines enables large language models to iteratively retrieve up-to-date external knowledge, enhancing adaptability and generalization in complex question-answering tasks. However, existing search agent pipelines typically depend on reinforcement learning based optimization, which often suffers from sparse outcome rewards, leading to inefficient exploration and unstable training. We introduce CriticSearch, a fine-grained credit-assignment framework that supplies dense, turn-level feedback via a retrospective critic mechanism. During training, a frozen, asymmetric critique LLM retrospectively evaluates each turn using privileged information from the full trajectory and gold answers, converting these assessments into stable, dense rewards that guide policy improvement. Experimental results across diverse multi-hop reasoning benchmarks demonstrate that CriticSearch consistently outperforms existing baselines, achieving faster convergence, improved training stability, and higher performance.

Yupeng Zheng, Xiang Li, Pengfei Li et al. · tsinghua

Monocular Semantic Occupancy Prediction aims to infer the complete 3D geometry and semantic information of scenes from only 2D images. It has garnered significant attention, particularly due to its potential to enhance the 3D perception of autonomous vehicles. However, existing methods rely on a complex cascaded framework with relatively limited information to restore 3D scenes, including a dependency on supervision solely on the whole network's output, single-frame input, and the utilization of a small backbone. These challenges, in turn, hinder the optimization of the framework and yield inferior prediction results, particularly concerning smaller and long-tailed objects. To address these issues, we propose MonoOcc. In particular, we (i) improve the monocular occupancy prediction framework by proposing an auxiliary semantic loss as supervision to the shallow layers of the framework and an image-conditioned cross-attention module to refine voxel features with visual clues, and (ii) employ a distillation module that transfers temporal information and richer knowledge from a larger image backbone to the monocular semantic occupancy prediction framework with low cost of hardware. With these advantages, our method yields state-of-the-art performance on the camera-based SemanticKITTI Scene Completion benchmark. Codes and models can be accessed at https://github.com/ucaszyp/MonoOcc

Junli Wang, Zhihua Hua, Xueyi Liu et al.

Existing imitation learning methods for end-to-end autonomous driving predominantly learn from successful demonstrations by minimizing geometric deviations from expert trajectories. This paradigm implicitly assumes that spatial proximity implies behavioral safety, leading to a critical objective mismatch: trajectories with nearly identical imitation losses may exhibit drastically different safety outcomes, where one remains recoverable while the other results in collision. To address this limitation, we propose BeyondDrive, a failure-aware imitation learning framework that jointly learns from successful and failed driving behaviors. First, we introduce a flow matching-based negative trajectory generator that synthesizes safety-critical yet expert-proximate trajectories, enabling explicit modeling of safety asymmetry. Second, we develop a diversity-aware sampling strategy that mitigates mode collapse and improves coverage of diverse failure modes during negative trajectory generation. Third, we propose a Repulsive Distance Loss that simultaneously attracts predictions toward expert demonstrations while repelling them from hard negative trajectories, thereby establishing discriminative safety boundaries in trajectory space. Applied to the uni-modal baseline Latent TransFuser, BeyondDrive achieves 89.7 PDMS on the NAVSIMv1 closed-loop benchmark, outperforming prior state-of-the-art methods. Moreover, BeyondDrive generalizes effectively across different autonomous driving architectures, including multi-modal planners, and further demonstrates strong zero-shot transferability on the HUGSIM benchmark.

Pengxuan Yang, Yupeng Zheng, Deheng Qian et al.

We introduce DreamerAD, the first latent world model framework that enables efficient reinforcement learning for autonomous driving by compressing diffusion sampling from 100 steps to 1 - achieving 80x speedup while maintaining visual interpretability. Training RL policies on real-world driving data incurs prohibitive costs and safety risks. While existing pixel-level diffusion world models enable safe imagination-based training, they suffer from multi-step diffusion inference latency (2s/frame) that prevents high-frequency RL interaction. Our approach leverages denoised latent features from video generation models through three key mechanisms: (1) shortcut forcing that reduces sampling complexity via recursive multi-resolution step compression, (2) an autoregressive dense reward model operating directly on latent representations for fine-grained credit assignment, and (3) Gaussian vocabulary sampling for GRPO that constrains exploration to physically plausible trajectories. DreamerAD achieves 87.7 EPDMS on NavSim v2, establishing state-of-the-art performance and demonstrating that latent-space RL is effective for autonomous driving.

Yupeng Zheng, Xiang Li, Songen Gu et al.

Recent advances in Vision-Language-Action (VLA) models have opened new avenues for robot manipulation, yet existing methods exhibit limited efficiency and a lack of high-level knowledge and spatial awareness. To address these challenges, we propose PokeVLA, a lightweight yet powerful foundation model for embodied manipulation that effectively infuses vision-language understanding into action learning. Our framework introduces a two-stage training paradigm: first, we pre-train a compact vision-language model (PokeVLM) on a curated multimodal dataset of 2.4M samples encompassing spatial grounding, affordance, and embodied reasoning tasks; second, we inject manipulation-relevant representations into the action space through multi-view goal-aware semantics learning, geometry alignment, and a novel action expert. Extensive experiments demonstrate state-of-the-art performance on the LIBERO-Plus benchmark and in real-world deployment, outperforming comparable baselines in success rate and robustness under diverse perturbations. To foster reproducibility and community progress, we will open-source our code, model weights, and the scripts for the curated pre-training dataset. Project page: https://getterupper.github.io/PokeVLA

Pengxuan Yang, Ben Lu, Zhongpu Xia et al.

Latent World Models enhance scene representation through temporal self-supervised learning, presenting a perception annotation-free paradigm for end-to-end autonomous driving. However, the reconstruction-oriented representation learning tangles perception with planning tasks, leading to suboptimal optimization for planning. To address this challenge, we propose WorldRFT, a planning-oriented latent world model framework that aligns scene representation learning with planning via a hierarchical planning decomposition and local-aware interactive refinement mechanism, augmented by reinforcement learning fine-tuning (RFT) to enhance safety-critical policy performance. Specifically, WorldRFT integrates a vision-geometry foundation model to improve 3D spatial awareness, employs hierarchical planning task decomposition to guide representation optimization, and utilizes local-aware iterative refinement to derive a planning-oriented driving policy. Furthermore, we introduce Group Relative Policy Optimization (GRPO), which applies trajectory Gaussianization and collision-aware rewards to fine-tune the driving policy, yielding systematic improvements in safety. WorldRFT achieves state-of-the-art (SOTA) performance on both open-loop nuScenes and closed-loop NavSim benchmarks. On nuScenes, it reduces collision rates by 83% (0.30% -> 0.05%). On NavSim, using camera-only sensors input, it attains competitive performance with the LiDAR-based SOTA method DiffusionDrive (87.8 vs. 88.1 PDMS).

Yupeng Zheng, Pengxuan Yang, Zebin Xing et al.

End-to-end autonomous driving directly generates planning trajectories from raw sensor data, yet it typically relies on costly perception supervision to extract scene information. A critical research challenge arises: constructing an informative driving world model to enable perception annotation-free, end-to-end planning via self-supervised learning. In this paper, we present World4Drive, an end-to-end autonomous driving framework that employs vision foundation models to build latent world models for generating and evaluating multi-modal planning trajectories. Specifically, World4Drive first extracts scene features, including driving intention and world latent representations enriched with spatial-semantic priors provided by vision foundation models. It then generates multi-modal planning trajectories based on current scene features and driving intentions and predicts multiple intention-driven future states within the latent space. Finally, it introduces a world model selector module to evaluate and select the best trajectory. We achieve perception annotation-free, end-to-end planning through self-supervised alignment between actual future observations and predicted observations reconstructed from the latent space. World4Drive achieves state-of-the-art performance without manual perception annotations on both the open-loop nuScenes and closed-loop NavSim benchmarks, demonstrating an 18.1\% relative reduction in L2 error, 46.7% lower collision rate, and 3.75 faster training convergence. Codes will be accessed at https://github.com/ucaszyp/World4Drive.

Pengxuan Yang, Yupeng Zheng, Qichao Zhang et al.

End-to-end autonomous driving aims to produce planning trajectories from raw sensors directly. Currently, most approaches integrate perception, prediction, and planning modules into a fully differentiable network, promising great scalability. However, these methods typically rely on deterministic modeling of online maps in the perception module for guiding or constraining vehicle planning, which may incorporate erroneous perception information and further compromise planning safety. To address this issue, we delve into the importance of online map uncertainty for enhancing autonomous driving safety and propose a novel paradigm named UncAD. Specifically, UncAD first estimates the uncertainty of the online map in the perception module. It then leverages the uncertainty to guide motion prediction and planning modules to produce multi-modal trajectories. Finally, to achieve safer autonomous driving, UncAD proposes an uncertainty-collision-aware planning selection strategy according to the online map uncertainty to evaluate and select the best trajectory. In this study, we incorporate UncAD into various state-of-the-art (SOTA) end-to-end methods. Experiments on the nuScenes dataset show that integrating UncAD, with only a 1.9% increase in parameters, can reduce collision rates by up to 26% and drivable area conflict rate by up to 42%. Codes, pre-trained models, and demo videos can be accessed at https://github.com/pengxuanyang/UncAD.

Zebin Xing, Pengxuan Yang, Linbo Wang et al.

Current autonomous driving systems often favor end-to-end frameworks, which take sensor inputs like images and learn to map them into trajectory space via neural networks. Previous work has demonstrated that models can achieve better planning performance when provided with a prior distribution of possible trajectories. However, these approaches often overlook two critical aspects: 1) The appropriate trajectory prior can vary significantly across different driving scenarios. 2) Their trajectory evaluation mechanism lacks policy-driven refinement, remaining constrained by the limitations of one-stage supervised training. To address these issues, we explore improvements in two key areas. For problem 1, we employ MoE to apply different trajectory priors tailored to different scenarios. For problem 2, we utilize Reinforcement Learning to fine-tune the trajectory scoring mechanism. Additionally, we integrate models with different perception backbones to enhance perceptual features. Our integrated model achieved a score of 51.08 on the navsim ICCV benchmark, securing third place.

Deqing Liu, Yinfeng Gao, Deheng Qian et al.

Existing end-to-end autonomous driving methods typically rely on imitation learning (IL) but face a key challenge: the misalignment between open-loop training and closed-loop deployment. This misalignment often triggers driver-initiated takeovers and system disengagements during closed-loop execution. How to leverage those expert takeover data from disengagement scenarios and effectively expand the IL policy's capability presents a valuable yet unexplored challenge. In this paper, we propose TakeAD, a novel preference-based post-optimization framework that fine-tunes the pre-trained IL policy with this disengagement data to enhance the closed-loop driving performance. First, we design an efficient expert takeover data collection pipeline inspired by human takeover mechanisms in real-world autonomous driving systems. Then, this post optimization framework integrates iterative Dataset Aggregation (DAgger) for imitation learning with Direct Preference Optimization (DPO) for preference alignment. The DAgger stage equips the policy with fundamental capabilities to handle disengagement states through direct imitation of expert interventions. Subsequently, the DPO stage refines the policy's behavior to better align with expert preferences in disengagement scenarios. Through multiple iterations, the policy progressively learns recovery strategies for disengagement states, thereby mitigating the open-loop gap. Experiments on the closed-loop Bench2Drive benchmark demonstrate our method's effectiveness compared with pure IL methods, with comprehensive ablations confirming the contribution of each component.

Chao Yang, Yong Fan, Qichao Zhang et al.

Recently, the transfer application of diffusion models in super-resolu-tion tasks has faced the problem ofdecreased fidelity. Due to the inherent randomsampling characteristics ofdiffusion models, direct application in super-resolu-tion tasks can result in generated details deviating from the true distribution ofhigh-resolution images. To address this, we propose DeltaDiff, a novel frame.work that constrains the difusion process, its essence is to establish a determin-istic mapping path between HR and LR, rather than the random noise disturbanceprocess oftraditional difusion models. Theoretical analysis demonstrates a 25%reduction in diffusion entropy in the residual space compared to pixel-space diffiusion, effectively suppressing irrelevant noise interference. The experimentalresults show that our method surpasses state-of-the-art models and generates re-sults with better fidelity. This work establishes a new low-rank constrained par-adigm for applying diffusion models to image reconstruction tasks, balancingstochastic generation with structural fidelity. Our code and model are publiclyavailable at https://github.com/continueyang/DeltaDiff .

Yuqi Liu, Qichao Zhang, Dongbin Zhao

In recent years, control under urban intersection scenarios becomes an emerging research topic. In such scenarios, the autonomous vehicle confronts complicated situations since it must deal with the interaction with social vehicles timely while obeying the traffic rules. Generally, the autonomous vehicle is supposed to avoid collisions while pursuing better efficiency. The existing work fails to provide a framework that emphasizes the integrity of the scenarios while being able to deploy and test reinforcement learning(RL) methods. Specifically, we propose a benchmark for training and testing RL-based autonomous driving agents in complex intersection scenarios, which is called RL-CIS. Then, a set of baselines are deployed consists of various algorithms. The test benchmark and baselines are to provide a fair and comprehensive training and testing platform for the study of RL for autonomous driving in the intersection scenario, advancing the progress of RL-based methods for intersection autonomous driving control. The code of our proposed framework can be found at https://github.com/liuyuqi123/ComplexUrbanScenarios.

Yuqian Fu, Tinghong Chen, Jiajun Chai et al.

Large language models (LLMs) have achieved remarkable progress in reasoning tasks, yet the optimal integration of Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) remains a fundamental challenge. Through comprehensive analysis of token distributions, learning dynamics, and integration mechanisms from entropy-based perspectives, we reveal key differences between these paradigms: SFT induces coarse-grained global changes to LLM policy distributions, while RL performs fine-grained selective optimizations, with entropy serving as a critical indicator of training effectiveness. Building on these observations, we propose Supervised Reinforcement Fine-Tuning (SRFT), a single-stage method that unifies both fine-tuning paradigms through entropy-aware weighting mechanisms. Our approach simultaneously applies SFT and RL to directly optimize the LLM using demonstrations and self-exploration rollouts rather than through two-stage sequential methods. Extensive experiments show that SRFT achieves 59.1% average accuracy, outperforming zero-RL methods by 9.0% on five mathematical reasoning benchmarks and 10.9% on three out-of-distribution benchmarks.

Songjun Tu, Jiahao Lin, Xiangyu Tian et al.

Recent advancements in post-training methodologies for large language models (LLMs) have highlighted reinforcement learning (RL) as a critical component for enhancing reasoning. However, the substantial computational costs associated with RL-based approaches have led to growing interest in alternative paradigms, such as Direct Preference Optimization (DPO). In this study, we investigate the effectiveness of DPO in facilitating self-improvement for LLMs through iterative preference-based learning. We demonstrate that a single round of DPO with coarse filtering significantly enhances mathematical reasoning performance, particularly for strong base model. Furthermore, we design an iterative enhancement framework for both the generator and the reward model (RM), enabling their mutual improvement through online interaction across multiple rounds of DPO. Finally, with simple verifiable rewards, our model DPO-VP achieves RL-level performance with significantly lower computational overhead. These findings highlight DPO as a scalable and cost-effective alternative to RL, offering a practical solution for enhancing LLM reasoning in resource-constrained situations.

Yuqian Fu, Yuanheng Zhu, Jiajun Chai et al.

Ensembling large language models (LLMs) can effectively combine diverse strengths of different models, offering a promising approach to enhance performance across various tasks. However, existing methods typically rely on fixed weighting strategies that fail to adapt to the dynamic, context-dependent characteristics of LLM capabilities. In this work, we propose Reinforcement Learning-Assisted Ensemble for LLMs (RLAE), a novel framework that reformulates LLM ensemble through the lens of a Markov Decision Process (MDP). Our approach introduces a RL agent that dynamically adjusts ensemble weights by considering both input context and intermediate generation states, with the agent being trained using rewards that directly correspond to the quality of final outputs. We implement RLAE using both single-agent and multi-agent reinforcement learning algorithms ($\text{RLAE}_\text{PPO}$ and $\text{RLAE}_\text{MAPPO}$ ), demonstrating substantial improvements over conventional ensemble methods. Extensive evaluations on a diverse set of tasks show that RLAE outperforms existing approaches by up to $3.3\%$ accuracy points, offering a more effective framework for LLM ensembling. Furthermore, our method exhibits superior generalization capabilities across different tasks without the need for retraining, while simultaneously achieving lower time latency.

Songjun Tu, Jingbo Sun, Qichao Zhang et al.

Preference-based reinforcement learning (PbRL) provides a powerful paradigm to avoid meticulous reward engineering by learning rewards based on human preferences. However, real-time human feedback is hard to obtain in online tasks. Most work suppose there is a "scripted teacher" that utilizes privileged predefined reward to provide preference feedback. In this paper, we propose a RL Self-augmented Large Language Model Feedback (RL-SaLLM-F) technique that does not rely on privileged information for online PbRL. RL-SaLLM-F leverages the reflective and discriminative capabilities of LLM to generate self-augmented trajectories and provide preference labels for reward learning. First, we identify an failure issue in LLM-based preference discrimination, specifically "query ambiguity", in online PbRL. Then LLM is employed to provide preference labels and generate self-augmented imagined trajectories that better achieve the task goal, thereby enhancing the quality and efficiency of feedback. Additionally, a double-check mechanism is introduced to mitigate randomness in the preference labels, improving the reliability of LLM feedback. The experiment across multiple tasks in the MetaWorld benchmark demonstrates the specific contributions of each proposed module in RL-SaLLM-F, and shows that self-augmented LLM feedback can effectively replace the impractical "scripted teacher" feedback. In summary, RL-SaLLM-F introduces a new direction of feedback acquisition in online PbRL that does not rely on any online privileged information, offering an efficient and lightweight solution with LLM-driven feedback.

Yinfeng Gao, Qichao Zhang, Da-wei Ding et al.

It is still a challenging topic to make reactive driving behaviors in complex urban environments as road users' intentions are unknown. Model-based reinforcement learning (MBRL) offers great potential to learn a reactive policy by constructing a world model that can provide informative states and imagination training. However, a critical limitation in relevant research lies in the scene-level reconstruction representation learning, which may overlook key interactive vehicles and hardly model the interactive features among vehicles and their long-term intentions. Therefore, this paper presents a novel MBRL method with a predictive individual world model (PIWM) for autonomous driving. PIWM describes the driving environment from an individual-level perspective and captures vehicles' interactive relations and their intentions via trajectory prediction task. Meanwhile, a behavior policy is learned jointly with PIWM. It is trained in PIWM's imagination and effectively navigates in the urban driving scenes leveraging intention-aware latent states. The proposed method is trained and evaluated on simulation environments built upon real-world challenging interactive scenarios. Compared with popular model-free and state-of-the-art model-based reinforcement learning methods, experimental results show that the proposed method achieves the best performance in terms of safety and efficiency.

Jingbo Sun, Qichao Zhang, Songjun Tu et al.

Zero-shot unsupervised reinforcement learning (URL) offers a promising direction for building generalist agents capable of generalizing to unseen tasks without additional supervision. Among existing approaches, successor representations (SR) have emerged as a prominent paradigm due to their effectiveness in structured, low-dimensional settings. However, SR methods struggle to scale to high-dimensional visual environments. Through empirical analysis, we identify two key limitations of SR in visual URL: (1) SR objectives often lead to suboptimal representations that attend to dynamics-irrelevant regions, resulting in inaccurate successor measures and degraded task generalization; and (2) these flawed representations hinder SR policies from modeling multi-modal skill-conditioned action distributions and ensuring skill controllability. To address these limitations, we propose Saliency-Guided Representation with Consistency Policy Learning (SRCP), a novel framework that improves zero-shot generalization of SR methods in visual URL. SRCP decouples representation learning from successor training by introducing a saliency-guided dynamics task to capture dynamics-relevant representations, thereby improving successor measure and task generalization. Moreover, it integrates a fast-sampling consistency policy with URL-specific classifier-free guidance and tailored training objectives to improve skill-conditioned policy modeling and controllability. Extensive experiments on 16 tasks across 4 datasets from the ExORL benchmark demonstrate that SRCP achieves state-of-the-art zero-shot generalization in visual URL and is compatible with various SR methods.

Songjun Tu, Jingbo Sun, Qichao Zhang et al.

Offline preference-based reinforcement learning (PbRL) typically operates in two phases: first, use human preferences to learn a reward model and annotate rewards for a reward-free offline dataset; second, learn a policy by optimizing the learned reward via offline RL. However, accurately modeling step-wise rewards from trajectory-level preference feedback presents inherent challenges. The reward bias introduced, particularly the overestimation of predicted rewards, leads to optimistic trajectory stitching, which undermines the pessimism mechanism critical to the offline RL phase. To address this challenge, we propose In-Dataset Trajectory Return Regularization (DTR) for offline PbRL, which leverages conditional sequence modeling to mitigate the risk of learning inaccurate trajectory stitching under reward bias. Specifically, DTR employs Decision Transformer and TD-Learning to strike a balance between maintaining fidelity to the behavior policy with high in-dataset trajectory returns and selecting optimal actions based on high reward labels. Additionally, we introduce an ensemble normalization technique that effectively integrates multiple reward models, balancing the tradeoff between reward differentiation and accuracy. Empirical evaluations on various benchmarks demonstrate the superiority of DTR over other state-of-the-art baselines.

Songjun Tu, Qichao Zhang, Jingbo Sun et al.

While multimodal large language models excel at tasks that integrate visual perception with symbolic reasoning, their performance is often undermined by a critical vulnerability: perception-induced errors that propagate through the reasoning chain. Current reinforcement learning (RL) fine-tuning methods, while enhancing reasoning abilities, largely fail to address the underlying misalignment between visual grounding and the subsequent reasoning process. To address this challenge, we propose \textbf{Caption-Regularized Policy Optimization (CapPO)}, a novel RL framework that explicitly enforces perceptual consistency during policy optimization. CapPO integrates two key mechanisms: (1) a caption-based consistency regularization, which minimizes the divergence between responses conditioned on raw images and those conditioned on captions, thereby anchoring reasoning to semantically faithful visual content; and (2) a KL-weighted advantage estimation scheme, which adaptively scales reinforcement signals to strengthen perceptually consistent trajectories while suppressing spurious correlations. Extensive experiments on five math-focused and five general reasoning benchmarks demonstrate that CapPO achieves competitive performance, yielding gains of +6.0% accuracy on math-related tasks and +2.4% on general reasoning tasks over the base Qwen2.5-VL-7B model. Moreover, ablation studies further confirm the effectiveness of each component, while error analysis reveals that CapPO significantly reduces perception-related mistakes compared with baselines. Overall, CapPO provides a simple yet effective framework for improving multimodal reasoning.

Jingbo Sun, Songjun Tu, Qichao Zhang et al.

Generalizing policies to unseen scenarios remains a critical challenge in visual reinforcement learning, where agents often overfit to the specific visual observations of the training environment. In unseen environments, distracting pixels may lead agents to extract representations containing task-irrelevant information. As a result, agents may deviate from the optimal behaviors learned during training, thereby hindering visual generalization.To address this issue, we propose the Salience-Invariant Consistent Policy Learning (SCPL) algorithm, an efficient framework for zero-shot generalization. Our approach introduces a novel value consistency module alongside a dynamics module to effectively capture task-relevant representations. The value consistency module, guided by saliency, ensures the agent focuses on task-relevant pixels in both original and perturbed observations, while the dynamics module uses augmented data to help the encoder capture dynamic- and reward-relevant representations. Additionally, our theoretical analysis highlights the importance of policy consistency for generalization. To strengthen this, we introduce a policy consistency module with a KL divergence constraint to maintain consistent policies across original and perturbed observations.Extensive experiments on the DMC-GB, Robotic Manipulation, and CARLA benchmarks demonstrate that SCPL significantly outperforms state-of-the-art methods in terms of generalization. Notably, SCPL achieves average performance improvements of 14\%, 39\%, and 69\% in the challenging DMC video hard setting, the Robotic hard setting, and the CARLA benchmark, respectively.Project Page: https://sites.google.com/view/scpl-rl.

Qichao Zhang, Yinfeng Gao, Yikang Zhang et al.

Realistic and diverse simulation scenarios with reactive and feasible agent behaviors can be used for validation and verification of self-driving system performance without relying on expensive and time-consuming real-world testing. Existing simulators rely on heuristic-based behavior models for background vehicles, which cannot capture the complex interactive behaviors in real-world scenarios. To bridge the gap between simulation and the real world, we propose TrajGen, a two-stage trajectory generation framework, which can capture more realistic behaviors directly from human demonstration. In particular, TrajGen consists of the multi-modal trajectory prediction stage and the reinforcement learning based trajectory modification stage. In the first stage, we propose a novel auxiliary RouteLoss for the trajectory prediction model to generate multi-modal diverse trajectories in the drivable area. In the second stage, reinforcement learning is used to track the predicted trajectories while avoiding collisions, which can improve the feasibility of generated trajectories. In addition, we develop a data-driven simulator I-Sim that can be used to train reinforcement learning models in parallel based on naturalistic driving data. The vehicle model in I-Sim can guarantee that the generated trajectories by TrajGen satisfy vehicle kinematic constraints. Finally, we give comprehensive metrics to evaluate generated trajectories for simulation scenarios, which shows that TrajGen outperforms either trajectory prediction or inverse reinforcement learning in terms of fidelity, reactivity, feasibility, and diversity.

Yuqi Liu, Qichao Zhang, Dongbin Zhao

Multi-task intersection navigation including the unprotected turning left, turning right, and going straight in dense traffic is still a challenging task for autonomous driving. For the human driver, the negotiation skill with other interactive vehicles is the key to guarantee safety and efficiency. However, it is hard to balance the safety and efficiency of the autonomous vehicle for multi-task intersection navigation. In this paper, we formulate a multi-task safe reinforcement learning with social attention to improve the safety and efficiency when interacting with other traffic participants. Specifically, the social attention module is used to focus on the states of negotiation vehicles. In addition, a safety layer is added to the multi-task reinforcement learning framework to guarantee safe negotiation. We compare the experiments in the simulator SUMO with abundant traffic flows and CARLA with high-fidelity vehicle models, which both show that the proposed algorithm can improve safety with consistent traffic efficiency for multi-task intersection navigation.

Junjie Wang, Qichao Zhang, Dongbin Zhao

The development of autonomous driving has attracted extensive attention in recent years, and it is essential to evaluate the performance of autonomous driving. However, testing on the road is expensive and inefficient. Virtual testing is the primary way to validate and verify self-driving cars, and the basis of virtual testing is to build simulation scenarios. In this paper, we propose a training, testing, and evaluation pipeline for the lane-changing task from the perspective of deep reinforcement learning. First, we design lane change scenarios for training and testing, where the test scenarios include stochastic and deterministic parts. Then, we deploy a set of benchmarks consisting of learning and non-learning approaches. We train several state-of-the-art deep reinforcement learning methods in the designed training scenarios and provide the benchmark metrics evaluation results of the trained models in the test scenarios. The designed lane-changing scenarios and benchmarks are both opened to provide a consistent experimental environment for the lane-changing task.

Junjie Wang, Qichao Zhang, Dongbin Zhao et al.

Existing model-based value expansion methods typically leverage a world model for value estimation with a fixed rollout horizon to assist policy learning. However, the fixed rollout with an inaccurate model has a potential to harm the learning process. In this paper, we investigate the idea of using the model knowledge for value expansion adaptively. We propose a novel method called Dynamic-horizon Model-based Value Expansion (DMVE) to adjust the world model usage with different rollout horizons. Inspired by reconstruction-based techniques that can be applied for visual data novelty detection, we utilize a world model with a reconstruction module for image feature extraction, in order to acquire more precise value estimation. The raw and the reconstructed images are both used to determine the appropriate horizon for adaptive value expansion. On several benchmark visual control tasks, experimental results show that DMVE outperforms all baselines in sample efficiency and final performance, indicating that DMVE can achieve more effective and accurate value estimation than state-of-the-art model-based methods.

Haoran Li, Qichao Zhang, Dongbin Zhao

This paper investigates the automatic exploration problem under the unknown environment, which is the key point of applying the robotic system to some social tasks. The solution to this problem via stacking decision rules is impossible to cover various environments and sensor properties. Learning based control methods are adaptive for these scenarios. However, these methods are damaged by low learning efficiency and awkward transferability from simulation to reality. In this paper, we construct a general exploration framework via decomposing the exploration process into the decision, planning, and mapping modules, which increases the modularity of the robotic system. Based on this framework, we propose a deep reinforcement learning based decision algorithm which uses a deep neural network to learning exploration strategy from the partial map. The results show that this proposed algorithm has better learning efficiency and adaptability for unknown environments. In addition, we conduct the experiments on the physical robot, and the results suggest that the learned policy can be well transfered from simulation to the real robot.

Haoran Li, Yaran Chen, Qichao Zhang et al.

Multi-sensor fusion-based road segmentation plays an important role in the intelligent driving system since it provides a drivable area. The existing mainstream fusion method is mainly to feature fusion in the image space domain which causes the perspective compression of the road and damages the performance of the distant road. Considering the bird's eye views(BEV) of the LiDAR remains the space structure in horizontal plane, this paper proposes a bidirectional fusion network(BiFNet) to fuse the image and BEV of the point cloud. The network consists of two modules: 1) Dense space transformation module, which solves the mutual conversion between camera image space and BEV space. 2) Context-based feature fusion module, which fuses the different sensors information based on the scenes from corresponding features.This method has achieved competitive results on KITTI dataset.

Dong Li, Qichao Zhang, Dongbin Zhao et al.

Visual navigation in complex environments is inefficient with traditional reactive policy or general-purposed recurrent policy. To address the long-term memory issue, this paper proposes a graph attention memory (GAM) architecture consisting of memory construction module, graph attention module and control module. The memory construction module builds the topological graph based on supervised learning by taking the exploration prior. Then, guided attention features are extracted with the graph attention module. Finally, the deep reinforcement learning based control module makes decisions based on visual observations and guided attention features. Detailed convergence analysis of GAM is presented in this paper. We evaluate GAM-based navigation system in two complex 3D environments. Experimental results show that the GAM-based navigation system significantly improves learning efficiency and outperforms all baselines in average success rate.

Junjie Wang, Qichao Zhang, Dongbin Zhao et al.

Autonomous driving decision-making is a great challenge due to the complexity and uncertainty of the traffic environment. Combined with the rule-based constraints, a Deep Q-Network (DQN) based method is applied for autonomous driving lane change decision-making task in this study. Through the combination of high-level lateral decision-making and low-level rule-based trajectory modification, a safe and efficient lane change behavior can be achieved. With the setting of our state representation and reward function, the trained agent is able to take appropriate actions in a real-world-like simulator. The generated policy is evaluated on the simulator for 10 times, and the results demonstrate that the proposed rule-based DQN method outperforms the rule-based approach and the DQN method.

Dong Li, Dongbin Zhao, Qichao Zhang et al.

This paper investigates the vision-based autonomous driving with deep learning and reinforcement learning methods. Different from the end-to-end learning method, our method breaks the vision-based lateral control system down into a perception module and a control module. The perception module which is based on a multi-task learning neural network first takes a driver-view image as its input and predicts the track features. The control module which is based on reinforcement learning then makes a control decision based on these features. In order to improve the data efficiency, we propose visual TORCS (VTORCS), a deep reinforcement learning environment which is based on the open racing car simulator (TORCS). By means of the provided functions, one can train an agent with the input of an image or various physical sensor measurement, or evaluate the perception algorithm on this simulator. The trained reinforcement learning controller outperforms the linear quadratic regulator (LQR) controller and model predictive control (MPC) controller on different tracks. The experiments demonstrate that the perception module shows promising performance and the controller is capable of controlling the vehicle drive well along the track center with visual input.