Wenhao Ding

Yan Wang, Wenjie Luo, Junjie Bai et al. · nvidia

End-to-end architectures trained via imitation learning have advanced autonomous driving by scaling model size and data, yet performance remains brittle in safety-critical long-tail scenarios where supervision is sparse and causal understanding is limited. To address this, we introduce Alpamayo-R1 (AR1), a vision-language-action model (VLA) that integrates Chain of Causation reasoning with trajectory planning to enhance decision-making in complex driving scenarios. Our approach features three key innovations: (1) the Chain of Causation (CoC) dataset, built through a hybrid auto-labeling and human-in-the-loop pipeline producing decision-grounded, causally linked reasoning traces aligned with driving behaviors; (2) a modular VLA architecture combining Cosmos-Reason, a Vision-Language Model pre-trained for Physical AI applications, with a diffusion-based trajectory decoder that generates dynamically feasible plans in real time; (3) a multi-stage training strategy using supervised fine-tuning to elicit reasoning and reinforcement learning (RL) to optimize reasoning quality via large reasoning model feedback and enforce reasoning-action consistency. Evaluation shows AR1 achieves up to a 12% improvement in planning accuracy on challenging cases compared to a trajectory-only baseline, with a 35% reduction in off-road rate and 25% reduction in close encounter rate in closed-loop simulation. RL post-training improves reasoning quality by 45% as measured by a large reasoning model critic and reasoning-action consistency by 37%. Model scaling from 0.5B to 7B parameters shows consistent improvements. On-vehicle road tests confirm real-time performance (99 ms latency) and successful urban deployment. By bridging interpretable reasoning with precise control, AR1 demonstrates a practical path towards Level 4 autonomous driving. We plan to release AR1 models and a subset of the CoC in a future update.

Haohong Lin, Wenhao Ding, Zuxin Liu et al. · cmu

In the domain of autonomous driving, the offline Reinforcement Learning~(RL) approaches exhibit notable efficacy in addressing sequential decision-making problems from offline datasets. However, maintaining safety in diverse safety-critical scenarios remains a significant challenge due to long-tailed and unforeseen scenarios absent from offline datasets. In this paper, we introduce the saFety-aware strUctured Scenario representatION (FUSION), a pioneering representation learning method in offline RL to facilitate the learning of a generalizable end-to-end driving policy by leveraging structured scenario information. FUSION capitalizes on the causal relationships between the decomposed reward, cost, state, and action space, constructing a framework for structured sequential reasoning in dynamic traffic environments. We conduct extensive evaluations in two typical real-world settings of the distribution shift in autonomous vehicles, demonstrating the good balance between safety cost and utility reward compared to the current state-of-the-art safe RL and IL baselines. Empirical evidence in various driving scenarios attests that FUSION significantly enhances the safety and generalizability of autonomous driving agents, even in the face of challenging and unseen environments. Furthermore, our ablation studies reveal noticeable improvements in the integration of causal representation into the offline safe RL algorithm. Our code implementation is available at: https://sites.google.com/view/safe-fusion/.

Junwon Seo, Sushant Veer, Ran Tian et al.

Video world models (WMs) have shown promise for policy evaluation and improvement by imagining realistic future observations conditioned on ego-robot actions. While WMs can model distributions over futures, policy evaluation and improvement typically rely on nominal imaginations, which can miss high-impact outcomes of robot actions unless prohibitively many samples are drawn. To enable robust policy evaluation and improvement over WM imaginations, we propose StressDream, which steers imaginations toward high-impact yet plausible outcomes specified at inference time by optimizing the initial noise of diffusion-based WMs. However, optimizing high-dimensional noise is challenging: the optimization must reason about nuanced, scene-dependent target events in generated videos while avoiding out-of-distribution (OOD) noise that yields implausible imaginations. We address this with two complementary objectives: a semantic objective with a Vision-Language Model that provides informative gradients by reasoning about the generated video, and a plausibility objective that prevents the optimized noise from drifting OOD. With state-of-the-art video world models for autonomous driving and robotic manipulation, we show that StressDream effectively steers imaginations toward high-impact yet plausible outcomes specified by text at inference time, such as task failures, enabling robust policy evaluation and improvement by identifying actions whose plausible futures include undesirable outcomes. Video results are available at https://junwon.me/StressDream/.

Aditi, Niket Agarwal, Arslan Ali et al.

We introduce Cosmos 3, a family of omnimodal world models designed to jointly process and generate language, image, video, audio, and action sequences within a unified mixture-of-transformers architecture. By supporting highly flexible input-output configurations, Cosmos 3 seamlessly unifies critical modalities for Physical AI -- effectively subsuming vision-language models, video generators, world simulators, and world-action models into a single framework. Our evaluation demonstrates that Cosmos 3 establishes a new state-of-the-art across a diverse suite of understanding and generation tasks, demonstrating omnimodal world models as scalable, general-purpose backbones for embodied agents. Our post-trained Cosmos 3 models were ranked as the best open-source Text-to-Image and Image-to-Video models by Artificial Analysis, and the best policy model by RoboArena at the time the technical report was written. To accelerate open research and deployment in Physical AI, we make our code, model checkpoints, curated synthetic datasets, and evaluation benchmark available under the Linux Foundation's OpenMDW-1.1 https://openmdw.ai/license/1-1/ License at https://github.com/nvidia/cosmos}{github.com/nvidia/cosmos and https://huggingface.co/collections/nvidia/cosmos3 . The project website is available at https://research.nvidia.com/labs/cosmos-lab/cosmos3 .

Wenhao Ding, Laixi Shi, Yuejie Chi et al. · cmu

Robustness has been extensively studied in reinforcement learning (RL) to handle various forms of uncertainty such as random perturbations, rare events, and malicious attacks. In this work, we consider one critical type of robustness against spurious correlation, where different portions of the state do not have correlations induced by unobserved confounders. These spurious correlations are ubiquitous in real-world tasks, for instance, a self-driving car usually observes heavy traffic in the daytime and light traffic at night due to unobservable human activity. A model that learns such useless or even harmful correlation could catastrophically fail when the confounder in the test case deviates from the training one. Although motivated, enabling robustness against spurious correlation poses significant challenges since the uncertainty set, shaped by the unobserved confounder and causal structure, is difficult to characterize and identify. Existing robust algorithms that assume simple and unstructured uncertainty sets are therefore inadequate to address this challenge. To solve this issue, we propose Robust State-Confounded Markov Decision Processes (RSC-MDPs) and theoretically demonstrate its superiority in avoiding learning spurious correlations compared with other robust RL counterparts. We also design an empirical algorithm to learn the robust optimal policy for RSC-MDPs, which outperforms all baselines in eight realistic self-driving and manipulation tasks.

Mengdi Xu, Zuxin Liu, Peide Huang et al. · cmu

A trustworthy reinforcement learning algorithm should be competent in solving challenging real-world problems, including {robustly} handling uncertainties, satisfying {safety} constraints to avoid catastrophic failures, and {generalizing} to unseen scenarios during deployments. This study aims to overview these main perspectives of trustworthy reinforcement learning considering its intrinsic vulnerabilities on robustness, safety, and generalizability. In particular, we give rigorous formulations, categorize corresponding methodologies, and discuss benchmarks for each perspective. Moreover, we provide an outlook section to spur promising future directions with a brief discussion on extrinsic vulnerabilities considering human feedback. We hope this survey could bring together separate threads of studies together in a unified framework and promote the trustworthiness of reinforcement learning.

Wenhao Ding, Haohong Lin, Bo Li et al. · cmu

As a pivotal component to attaining generalizable solutions in human intelligence, reasoning provides great potential for reinforcement learning (RL) agents' generalization towards varied goals by summarizing part-to-whole arguments and discovering cause-and-effect relations. However, how to discover and represent causalities remains a huge gap that hinders the development of causal RL. In this paper, we augment Goal-Conditioned RL (GCRL) with Causal Graph (CG), a structure built upon the relation between objects and events. We novelly formulate the GCRL problem into variational likelihood maximization with CG as latent variables. To optimize the derived objective, we propose a framework with theoretical performance guarantees that alternates between two steps: using interventional data to estimate the posterior of CG; using CG to learn generalizable models and interpretable policies. Due to the lack of public benchmarks that verify generalization capability under reasoning, we design nine tasks and then empirically show the effectiveness of the proposed method against five baselines on these tasks. Further theoretical analysis shows that our performance improvement is attributed to the virtuous cycle of causal discovery, transition modeling, and policy training, which aligns with the experimental evidence in extensive ablation studies.

Wenhao Ding, Nathalie Majcherczyk, Mohit Deshpande et al. · cmu

Active perception describes a broad class of techniques that couple planning and perception systems to move the robot in a way to give the robot more information about the environment. In most robotic systems, perception is typically independent of motion planning. For example, traditional object detection is passive: it operates only on the images it receives. However, we have a chance to improve the results if we allow planning to consume detection signals and move the robot to collect views that maximize the quality of the results. In this paper, we use reinforcement learning (RL) methods to control the robot in order to obtain images that maximize the detection quality. Specifically, we propose using a Decision Transformer with online fine-tuning, which first optimizes the policy with a pre-collected expert dataset and then improves the learned policy by exploring better solutions in the environment. We evaluate the performance of proposed method on an interactive dataset collected from an indoor scenario simulator. Experimental results demonstrate that our method outperforms all baselines, including expert policy and pure offline RL methods. We also provide exhaustive analyses of the reward distribution and observation space.

Yihang Yao, Zhepeng Cen, Wenhao Ding et al. · cmu

Offline safe reinforcement learning (RL) aims to train a policy that satisfies constraints using a pre-collected dataset. Most current methods struggle with the mismatch between imperfect demonstrations and the desired safe and rewarding performance. In this paper, we introduce OASIS (cOnditionAl diStributIon Shaping), a new paradigm in offline safe RL designed to overcome these critical limitations. OASIS utilizes a conditional diffusion model to synthesize offline datasets, thus shaping the data distribution toward a beneficial target domain. Our approach makes compliance with safety constraints through effective data utilization and regularization techniques to benefit offline safe RL training. Comprehensive evaluations on public benchmarks and varying datasets showcase OASIS's superiority in benefiting offline safe RL agents to achieve high-reward behavior while satisfying the safety constraints, outperforming established baselines. Furthermore, OASIS exhibits high data efficiency and robustness, making it suitable for real-world applications, particularly in tasks where safety is imperative and high-quality demonstrations are scarce.

Haohong Lin, Wenhao Ding, Jian Chen et al. · cmu

Offline model-based reinforcement learning (MBRL) enhances data efficiency by utilizing pre-collected datasets to learn models and policies, especially in scenarios where exploration is costly or infeasible. Nevertheless, its performance often suffers from the objective mismatch between model and policy learning, resulting in inferior performance despite accurate model predictions. This paper first identifies the primary source of this mismatch comes from the underlying confounders present in offline data for MBRL. Subsequently, we introduce \textbf{B}ilin\textbf{E}ar \textbf{CAUS}al r\textbf{E}presentation~(BECAUSE), an algorithm to capture causal representation for both states and actions to reduce the influence of the distribution shift, thus mitigating the objective mismatch problem. Comprehensive evaluations on 18 tasks that vary in data quality and environment context demonstrate the superior performance of BECAUSE over existing offline RL algorithms. We show the generalizability and robustness of BECAUSE under fewer samples or larger numbers of confounders. Additionally, we offer theoretical analysis of BECAUSE to prove its error bound and sample efficiency when integrating causal representation into offline MBRL.

Wenhao Ding, Qing He, Hanghang Tong et al.

As a fundamental mathmatical tool in many engineering disciplines, coupled differential equation groups are being widely used to model complex structures containing multiple physical quantities. Engineers constantly adjust structural parameters at the design stage, which requires a highly efficient solver. The rise of deep learning technologies has offered new perspectives on this task. Unfortunately, existing black-box models suffer from poor accuracy and robustness, while the advanced methodologies of single-output operator regression cannot deal with multiple quantities simultaneously. To address these challenges, we propose PINO-CDE, a deep learning framework for solving coupled differential equation groups (CDEs) along with an equation normalization algorithm for performance enhancing. Based on the theory of physics-informed neural operator (PINO), PINO-CDE uses a single network for all quantities in a CDEs, instead of training dozens, or even hundreds of networks as in the existing literature. We demonstrate the flexibility and feasibility of PINO-CDE for one toy example and two engineering applications: vehicle-track coupled dynamics (VTCD) and reliability assessment for a four-storey building (uncertainty propagation). The performance of VTCD indicates that PINO-CDE outperforms existing software and deep learning-based methods in terms of efficiency and precision, respectively. For the uncertainty propagation task, PINO-CDE provides higher-resolution results in less than a quarter of the time incurred when using the probability density evolution method (PDEM). This framework integrates engineering dynamics and deep learning technologies and may reveal a new concept for CDEs solving and uncertainty propagation.

Yingzi Ma, Yulong Cao, Wenhao Ding et al.

The autonomous driving community is increasingly focused on addressing the challenges posed by out-of-distribution (OOD) driving scenarios. A dominant research trend seeks to enhance end-to-end (E2E) driving systems by integrating vision-language models (VLMs), leveraging their rich world knowledge and reasoning abilities to improve generalization across diverse environments. However, most existing VLMs or vision-language agents (VLAs) are built upon autoregressive (AR) models. In this paper, we observe that existing AR-based VLMs -- limited by causal attention and sequential token generation -- often fail to maintain consistency and controllability between high-level reasoning and low-level planning. In contrast, recent discrete diffusion VLMs equipped with bidirectional attention exhibit superior controllability and reliability through iterative denoising. Building on these observations, we introduce dVLM-AD, a diffusion-based vision-language model that unifies perception, structured reasoning, and low-level planning for end-to-end driving. Evaluated on nuScenes and WOD-E2E, dVLM-AD yields more consistent reasoning-action pairs and achieves planning performance comparable to existing driving VLM/VLA systems despite a modest backbone, outperforming AR-based baselines with a 9 percent improvement in behavior-trajectory consistency and a 6 percent increase in RFS on long-tail WOD-E2E scenarios. These results suggest a controllable and reliable pathway for scalable end-to-end driving.

Yiying Sheng, Wenhao Ding, Dylan Roi et al.

Extensive studies suggested that fluid mechanical markers of intracranial aneurysms (IAs) derived from Computational Fluid Dynamics (CFD) can indicate disease progression risks, but to date this has not been translated clinically. This is because CFD requires specialized expertise and is time-consuming and low throughput, making it difficult to support clinical trials. A deep learning model that maps IA morphology to biomechanical markers can address this, enabling physicians to obtain these markers in real time without performing CFD. Here, we show that a Graph Transformer model that incorporates temporal information, which is supervised by large CFD data, can accurately predict Wall Shear Stress (WSS) across the cardiac cycle from IA surface meshes. The model effectively captures the temporal variations of the WSS pattern, achieving a Structural Similarity Index (SSIM) of up to 0.981 and a maximum-based relative L2 error of 2.8%. Ablation studies and SOTA comparison confirmed its optimality. Further, as pulsatile CFD data is computationally expensive to generate and sample sizes are limited, we engaged a strategy of injecting a large amount of steady-state CFD data, which are extremely low-cost to generate, as augmentation. This approach enhances network performance substantially when pulsatile CFD data sample size is small. Our study provides a proof of concept that temporal sequences cardiovascular fluid mechanical parameters can be computed in real time using a deep learning model from the geometric mesh, and this is achievable even with small pulsatile CFD sample size. Our approach is likely applicable to other cardiovascular scenarios.

Yihao Luo, Dario Sesia, Fanwen Wang et al.

Mesh reconstruction of the cardiac anatomy from medical images is useful for shape and motion measurements and biophysics simulations to facilitate the assessment of cardiac function and health. However, 3D medical images are often acquired as 2D slices that are sparsely sampled and noisy, and mesh reconstruction on such data is a challenging task. Traditional voxel-based approaches rely on pre- and post-processing that compromises image fidelity, while mesh-level deep learning approaches require mesh annotations that are difficult to get. Therefore, direct cross-domain supervision from 2D images to meshes is a key technique for advancing 3D learning in medical imaging, but it has not been well-developed. While there have been attempts to approximate the optimized meshes' slicing, few existing methods directly use 2D slices to supervise mesh reconstruction in a differentiable manner. Here, we propose a novel explicit differentiable voxelization and slicing (DVS) algorithm that allows gradient backpropagation to a mesh from its slices, facilitating refined mesh optimization directly supervised by the losses defined on 2D images. Further, we propose an innovative framework for extracting patient-specific left ventricle (LV) meshes from medical images by coupling DVS with a graph harmonic deformation (GHD) mesh morphing descriptor of cardiac shape that naturally preserves mesh quality and smoothness during optimization. Experimental results demonstrate that our method achieves state-of-the-art performance in cardiac mesh reconstruction tasks from CT and MRI, with an overall Dice score of 90% on multi-datasets, outperforming existing approaches. The proposed method can further quantify clinically useful parameters such as ejection fraction and global myocardial strains, closely matching the ground truth and surpassing the traditional voxel-based approach in sparse images.

Zhejun Zhang, Peter Karkus, Maximilian Igl et al.

Traffic simulation aims to learn a policy for traffic agents that, when unrolled in closed-loop, faithfully recovers the joint distribution of trajectories observed in the real world. Inspired by large language models, tokenized multi-agent policies have recently become the state-of-the-art in traffic simulation. However, they are typically trained through open-loop behavior cloning, and thus suffer from covariate shift when executed in closed-loop during simulation. In this work, we present Closest Among Top-K (CAT-K) rollouts, a simple yet effective closed-loop fine-tuning strategy to mitigate covariate shift. CAT-K fine-tuning only requires existing trajectory data, without reinforcement learning or generative adversarial imitation. Concretely, CAT-K fine-tuning enables a small 7M-parameter tokenized traffic simulation policy to outperform a 102M-parameter model from the same model family, achieving the top spot on the Waymo Sim Agent Challenge leaderboard at the time of submission. The code is available at https://github.com/NVlabs/catk.

Haohong Lin, Yunzhi Zhang, Wenhao Ding et al.

End-to-end (E2E) autonomous driving models have demonstrated strong performance in open-loop evaluations but often suffer from cascading errors and poor generalization in closed-loop settings. To address this gap, we propose Model-based Policy Adaptation (MPA), a general framework that enhances the robustness and safety of pretrained E2E driving agents during deployment. MPA first generates diverse counterfactual trajectories using a geometry-consistent simulation engine, exposing the agent to scenarios beyond the original dataset. Based on this generated data, MPA trains a diffusion-based policy adapter to refine the base policy's predictions and a multi-step Q value model to evaluate long-term outcomes. At inference time, the adapter proposes multiple trajectory candidates, and the Q value model selects the one with the highest expected utility. Experiments on the nuScenes benchmark using a photorealistic closed-loop simulator demonstrate that MPA significantly improves performance across in-domain, out-of-domain, and safety-critical scenarios. We further investigate how the scale of counterfactual data and inference-time guidance strategies affect overall effectiveness.

Miao Li, Wenhao Ding, Ding Zhao

The prominence of embodied Artificial Intelligence (AI), which empowers robots to navigate, perceive, and engage within virtual environments, has attracted significant attention, owing to the remarkable advances in computer vision and large language models. Privacy emerges as a pivotal concern within the realm of embodied AI, as the robot accesses substantial personal information. However, the issue of privacy leakage in embodied AI tasks, particularly concerning reinforcement learning algorithms, has not received adequate consideration in research. This paper aims to address this gap by proposing an attack on the training process of the value-based algorithm and the gradient-based algorithm, utilizing gradient inversion to reconstruct states, actions, and supervisory signals. The choice of using gradients for the attack is motivated by the fact that commonly employed federated learning techniques solely utilize gradients computed based on private user data to optimize models, without storing or transmitting the data to public servers. Nevertheless, these gradients contain sufficient information to potentially expose private data. To validate our approach, we conducted experiments on the AI2THOR simulator and evaluated our algorithm on active perception, a prevalent task in embodied AI. The experimental results demonstrate the effectiveness of our method in successfully reconstructing all information from the data in 120 room layouts. Check our website for videos.

Wenhao Ding, Choon Hwai Yap, Kangjun Ji et al.

A generative model for the mesh geometry of intracranial aneurysms (IA) is crucial for training networks to predict blood flow forces in real time, which is a key factor affecting disease progression. This need is necessitated by the absence of a large IA image datasets. Existing shape generation methods struggle to capture realistic IA features and ignore the relationship between IA pouches and parent vessels, limiting physiological realism and their generation cannot be controlled to have specific morphological measurements. We propose AneuG, a two-stage Variational Autoencoder (VAE)-based IA mesh generator. In the first stage, AneuG generates low-dimensional Graph Harmonic Deformation (GHD) tokens to encode and reconstruct aneurysm pouch shapes, constrained to morphing energy statistics truths. GHD enables more accurate shape encoding than alternatives. In the second stage, AneuG generates parent vessels conditioned on GHD tokens, by generating vascular centreline and propagating the cross-section. AneuG's IA shape generation can further be conditioned to have specific clinically relevant morphological measurements. This is useful for studies to understand shape variations represented by clinical measurements, and for flow simulation studies to understand effects of specific clinical shape parameters on fluid dynamics. Source code and implementation details are available at https://github.com/anonymousaneug/AneuG.

Hanjiang Hu, Baoquan Yang, Zhijian Qiao et al.

Different environments pose a great challenge to the outdoor robust visual perception for long-term autonomous driving, and the generalization of learning-based algorithms on different environments is still an open problem. Although monocular depth prediction has been well studied recently, few works focus on the robustness of learning-based depth prediction across different environments, e.g. changing illumination and seasons, owing to the lack of such a multi-environment real-world dataset and benchmark. To this end, the first cross-season monocular depth prediction dataset and benchmark, SeasonDepth, is introduced to benchmark the depth estimation performance under different environments. We investigate several state-of-the-art representative open-source supervised and self-supervised depth prediction methods using newly-formulated metrics. Through extensive experimental evaluation on the proposed dataset and cross-dataset evaluation with current autonomous driving datasets, the performance and robustness against the influence of multiple environments are analyzed qualitatively and quantitatively. We show that long-term monocular depth prediction is still challenging and believe our work can boost further research on the long-term robustness and generalization for outdoor visual perception. The dataset is available on https://seasondepth.github.io, and the benchmark toolkit is available on https://github.com/ SeasonDepth/SeasonDepth.

Wenhao Ding, Yulong Cao, Ding Zhao et al. · cmu

Simulation plays a crucial role in the development of autonomous vehicles (AVs) due to the potential risks associated with real-world testing. Although significant progress has been made in the visual aspects of simulators, generating complex behavior among agents remains a formidable challenge. It is not only imperative to ensure realism in the scenarios generated but also essential to incorporate preferences and conditions to facilitate controllable generation for AV training and evaluation. Traditional methods, mainly relying on memorizing the distribution of training datasets, often fall short in generating unseen scenarios. Inspired by the success of retrieval augmented generation in large language models, we present RealGen, a novel retrieval-based in-context learning framework for traffic scenario generation. RealGen synthesizes new scenarios by combining behaviors from multiple retrieved examples in a gradient-free way, which may originate from templates or tagged scenarios. This in-context learning framework endows versatile generative capabilities, including the ability to edit scenarios, compose various behaviors, and produce critical scenarios. Evaluations show that RealGen offers considerable flexibility and controllability, marking a new direction in the field of controllable traffic scenario generation. Check our project website for more information: https://realgen.github.io.

Yiqing Liang, Jielin Qiu, Wenhao Ding et al. · cmu

Reinforcement Learning with Verifiable Rewards (RLVR) has recently emerged as a powerful paradigm for post-training large language models (LLMs), achieving state-of-the-art performance on tasks with structured, verifiable answers. Applying RLVR to Multimodal LLMs (MLLMs) presents significant opportunities but is complicated by the broader, heterogeneous nature of vision-language tasks that demand nuanced visual, logical, and spatial capabilities. As such, training MLLMs using RLVR on multiple datasets could be beneficial but creates challenges with conflicting objectives from interaction among diverse datasets, highlighting the need for optimal dataset mixture strategies to improve generalization and reasoning. We introduce a systematic post-training framework for Multimodal LLM RLVR, featuring a rigorous data mixture problem formulation and benchmark implementation. Specifically, (1) We developed a multimodal RLVR framework for multi-dataset post-training by curating a dataset that contains different verifiable vision-language problems and enabling multi-domain online RL learning with different verifiable rewards; (2) We proposed a data mixture strategy that learns to predict the RL fine-tuning outcome from the data mixture distribution, and consequently optimizes the best mixture. Comprehensive experiments showcase that multi-domain RLVR training, when combined with mixture prediction strategies, can significantly boost MLLM general reasoning capacities. Our best mixture improves the post-trained model's accuracy on out-of-distribution benchmarks by an average of 5.24% compared to the same model post-trained with uniform data mixture, and by a total of 20.74% compared to the pre-finetuning baseline.

Wenhao Ding, Sushant Veer, Karen Leung et al.

Validating the safety and performance of an autonomous vehicle (AV) requires benchmarking on real-world driving logs. However, typical driving logs contain mostly uneventful scenarios with minimal interactions between road users. Identifying interactive scenarios in real-world driving logs enables the curation of datasets that amplify critical signals and provide a more accurate assessment of an AV's performance. In this paper, we present a novel metric that identifies interactive scenarios by measuring an AV's surprise potential on others. First, we identify three dimensions of the design space to describe a family of surprise potential measures. Second, we exhaustively evaluate and compare different instantiations of the surprise potential measure within this design space on the nuScenes dataset. To determine how well a surprise potential measure correctly identifies an interactive scenario, we use a reward model learned from human preferences to assess alignment with human intuition. Our proposed surprise potential, arising from this exhaustive comparative study, achieves a correlation of more than 0.82 with the human-aligned reward function, outperforming existing approaches. Lastly, we validate motion planners on curated interactive scenarios to demonstrate downstream applications.

Miao Li, Wenhao Ding, Haohong Lin et al. · cmu

Training and evaluating autonomous driving algorithms requires a diverse range of scenarios. However, most available datasets predominantly consist of normal driving behaviors demonstrated by human drivers, resulting in a limited number of safety-critical cases. This imbalance, often referred to as a long-tail distribution, restricts the ability of driving algorithms to learn from crucial scenarios involving risk or failure, scenarios that are essential for humans to develop driving skills efficiently. To generate such scenarios, we utilize Multi-modal Large Language Models to convert crash reports of accidents into a structured scenario format, which can be directly executed within simulations. Specifically, we introduce CrashAgent, a multi-agent framework designed to interpret multi-modal real-world traffic crash reports for the generation of both road layouts and the behaviors of the ego vehicle and surrounding traffic participants. We comprehensively evaluate the generated crash scenarios from multiple perspectives, including the accuracy of layout reconstruction, collision rate, and diversity. The resulting high-quality and large-scale crash dataset will be publicly available to support the development of safe driving algorithms in handling safety-critical situations.

Wenhao Ding, Sushant Veer, Yuxiao Chen et al.

Learning-based planners generate natural human-like driving behaviors by learning to reason about nuanced interactions from data, overcoming the rigid behaviors that arise from rule-based planners. Nonetheless, data-driven approaches often struggle with rare, safety-critical scenarios and offer limited controllability over the generated trajectories. To address these challenges, we propose RealDrive, a Retrieval-Augmented Generation (RAG) framework that initializes a diffusion-based planning policy by retrieving the most relevant expert demonstrations from the training dataset. By interpolating between current observations and retrieved examples through a denoising process, our approach enables fine-grained control and safe behavior across diverse scenarios, leveraging the strong prior provided by the retrieved scenario. Another key insight we produce is that a task-relevant retrieval model trained with planning-based objectives results in superior planning performance in our framework compared to a task-agnostic retriever. Experimental results demonstrate improved generalization to long-tail events and enhanced trajectory diversity compared to standard learning-based planners -- we observe a 40% reduction in collision rate on the Waymo Open Motion dataset with RAG.

Wenhao Ding, Tong Che, Ding Zhao et al.

Recently, reward-conditioned reinforcement learning (RCRL) has gained popularity due to its simplicity, flexibility, and off-policy nature. However, we will show that current RCRL approaches are fundamentally limited and fail to address two critical challenges of RCRL -- improving generalization on high reward-to-go (RTG) inputs, and avoiding out-of-distribution (OOD) RTG queries during testing time. To address these challenges when training vanilla RCRL architectures, we propose Bayesian Reparameterized RCRL (BR-RCRL), a novel set of inductive biases for RCRL inspired by Bayes' theorem. BR-RCRL removes a core obstacle preventing vanilla RCRL from generalizing on high RTG inputs -- a tendency that the model treats different RTG inputs as independent values, which we term ``RTG Independence". BR-RCRL also allows us to design an accompanying adaptive inference method, which maximizes total returns while avoiding OOD queries that yield unpredictable behaviors in vanilla RCRL methods. We show that BR-RCRL achieves state-of-the-art performance on the Gym-Mujoco and Atari offline RL benchmarks, improving upon vanilla RCRL by up to 11%.

Wenhao Ding, Chejian Xu, Mansur Arief et al.

Autonomous driving systems have witnessed a significant development during the past years thanks to the advance in machine learning-enabled sensing and decision-making algorithms. One critical challenge for their massive deployment in the real world is their safety evaluation. Most existing driving systems are still trained and evaluated on naturalistic scenarios collected from daily life or heuristically-generated adversarial ones. However, the large population of cars, in general, leads to an extremely low collision rate, indicating that the safety-critical scenarios are rare in the collected real-world data. Thus, methods to artificially generate scenarios become crucial to measure the risk and reduce the cost. In this survey, we focus on the algorithms of safety-critical scenario generation in autonomous driving. We first provide a comprehensive taxonomy of existing algorithms by dividing them into three categories: data-driven generation, adversarial generation, and knowledge-based generation. Then, we discuss useful tools for scenario generation, including simulation platforms and packages. Finally, we extend our discussion to five main challenges of current works -- fidelity, efficiency, diversity, transferability, controllability -- and research opportunities lighted up by these challenges.

Mansur Arief, Yuanlu Bai, Wenhao Ding et al.

Rare-event simulation techniques, such as importance sampling (IS), constitute powerful tools to speed up challenging estimation of rare catastrophic events. These techniques often leverage the knowledge and analysis on underlying system structures to endow desirable efficiency guarantees. However, black-box problems, especially those arising from recent safety-critical applications of AI-driven physical systems, can fundamentally undermine their efficiency guarantees and lead to dangerous under-estimation without diagnostically detected. We propose a framework called Deep Probabilistic Accelerated Evaluation (Deep-PrAE) to design statistically guaranteed IS, by converting black-box samplers that are versatile but could lack guarantees, into one with what we call a relaxed efficiency certificate that allows accurate estimation of bounds on the rare-event probability. We present the theory of Deep-PrAE that combines the dominating point concept with rare-event set learning via deep neural network classifiers, and demonstrate its effectiveness in numerical examples including the safety-testing of intelligent driving algorithms.

Wenhao Ding, Haohong Lin, Bo Li et al.

Generating safety-critical scenarios, which are crucial yet difficult to collect, provides an effective way to evaluate the robustness of autonomous driving systems. However, the diversity of scenarios and efficiency of generation methods are heavily restricted by the rareness and structure of safety-critical scenarios. Therefore, existing generative models that only estimate distributions from observational data are not satisfying to solve this problem. In this paper, we integrate causality as a prior into the scenario generation and propose a flow-based generative framework, Causal Autoregressive Flow (CausalAF). CausalAF encourages the generative model to uncover and follow the causal relationship among generated objects via novel causal masking operations instead of searching the sample only from observational data. By learning the cause-and-effect mechanism of how the generated scenario causes risk situations rather than just learning correlations from data, CausalAF significantly improves learning efficiency. Extensive experiments on three heterogeneous traffic scenarios illustrate that CausalAF requires much fewer optimization resources to effectively generate safety-critical scenarios. We also show that using generated scenarios as additional training samples empirically improves the robustness of autonomous driving algorithms.

Wenhao Ding, Haohong Lin, Bo Li et al.

Generating adversarial scenarios, which have the potential to fail autonomous driving systems, provides an effective way to improve robustness. Extending purely data-driven generative models, recent specialized models satisfy additional controllable requirements such as embedding a traffic sign in a driving scene by manipulating patterns implicitly in the neuron level. In this paper, we introduce a method to incorporate domain knowledge explicitly in the generation process to achieve the Semantically Adversarial Generation (SAG). To be consistent with the composition of driving scenes, we first categorize the knowledge into two types, the property of objects and the relationship among objects. We then propose a tree-structured variational auto-encoder (T-VAE) to learn hierarchical scene representation. By imposing semantic rules on the properties of nodes and edges in the tree structure, explicit knowledge integration enables controllable generation. We construct a synthetic example to illustrate the controllability and explainability of our method in a succinct setting. We further extend to realistic environments for autonomous vehicles: our method efficiently identifies adversarial driving scenes against different state-of-the-art 3D point cloud segmentation models and satisfies the traffic rules specified as the explicit knowledge.

Baiming Chen, Zuxin Liu, Jiacheng Zhu et al.

Safety is a critical concern when deploying reinforcement learning agents for realistic tasks. Recently, safe reinforcement learning algorithms have been developed to optimize the agent's performance while avoiding violations of safety constraints. However, few studies have addressed the non-stationary disturbances in the environments, which may cause catastrophic outcomes. In this paper, we propose the context-aware safe reinforcement learning (CASRL) method, a meta-learning framework to realize safe adaptation in non-stationary environments. We use a probabilistic latent variable model to achieve fast inference of the posterior environment transition distribution given the context data. Safety constraints are then evaluated with uncertainty-aware trajectory sampling. The high cost of safety violations leads to the rareness of unsafe records in the dataset. We address this issue by enabling prioritized sampling during model training and formulating prior safety constraints with domain knowledge during constrained planning. The algorithm is evaluated in realistic safety-critical environments with non-stationary disturbances. Results show that the proposed algorithm significantly outperforms existing baselines in terms of safety and robustness.

Wenhao Ding, Baiming Chen, Bo Li et al.

Existing neural network-based autonomous systems are shown to be vulnerable against adversarial attacks, therefore sophisticated evaluation on their robustness is of great importance. However, evaluating the robustness only under the worst-case scenarios based on known attacks is not comprehensive, not to mention that some of them even rarely occur in the real world. In addition, the distribution of safety-critical data is usually multimodal, while most traditional attacks and evaluation methods focus on a single modality. To solve the above challenges, we propose a flow-based multimodal safety-critical scenario generator for evaluating decisionmaking algorithms. The proposed generative model is optimized with weighted likelihood maximization and a gradient-based sampling procedure is integrated to improve the sampling efficiency. The safety-critical scenarios are generated by querying the task algorithms and the log-likelihood of the generated scenarios is in proportion to the risk level. Experiments on a self-driving task demonstrate our advantages in terms of testing efficiency and multimodal modeling capability. We evaluate six Reinforcement Learning algorithms with our generated traffic scenarios and provide empirical conclusions about their robustness.

Mansur Arief, Zhiyuan Huang, Guru Koushik Senthil Kumar et al.

Evaluating the reliability of intelligent physical systems against rare safety-critical events poses a huge testing burden for real-world applications. Simulation provides a useful platform to evaluate the extremal risks of these systems before their deployments. Importance Sampling (IS), while proven to be powerful for rare-event simulation, faces challenges in handling these learning-based systems due to their black-box nature that fundamentally undermines its efficiency guarantee, which can lead to under-estimation without diagnostically detected. We propose a framework called Deep Probabilistic Accelerated Evaluation (Deep-PrAE) to design statistically guaranteed IS, by converting black-box samplers that are versatile but could lack guarantees, into one with what we call a relaxed efficiency certificate that allows accurate estimation of bounds on the safety-critical event probability. We present the theory of Deep-PrAE that combines the dominating point concept with rare-event set learning via deep neural network classifiers, and demonstrate its effectiveness in numerical examples including the safety-testing of an intelligent driving algorithm.

Mengdi Xu, Wenhao Ding, Jiacheng Zhu et al.

Continuously learning to solve unseen tasks with limited experience has been extensively pursued in meta-learning and continual learning, but with restricted assumptions such as accessible task distributions, independently and identically distributed tasks, and clear task delineations. However, real-world physical tasks frequently violate these assumptions, resulting in performance degradation. This paper proposes a continual online model-based reinforcement learning approach that does not require pre-training to solve task-agnostic problems with unknown task boundaries. We maintain a mixture of experts to handle nonstationarity, and represent each different type of dynamics with a Gaussian Process to efficiently leverage collected data and expressively model uncertainty. We propose a transition prior to account for the temporal dependencies in streaming data and update the mixture online via sequential variational inference. Our approach reliably handles the task distribution shift by generating new models for never-before-seen dynamics and reusing old models for previously seen dynamics. In experiments, our approach outperforms alternative methods in non-stationary tasks, including classic control with changing dynamics and decision making in different driving scenarios.

Wenhao Ding, Baiming Chen, Minjun Xu et al.

Long-tail and rare event problems become crucial when autonomous driving algorithms are applied in the real world. For the purpose of evaluating systems in challenging settings, we propose a generative framework to create safety-critical scenarios for evaluating specific task algorithms. We first represent the traffic scenarios with a series of autoregressive building blocks and generate diverse scenarios by sampling from the joint distribution of these blocks. We then train the generative model as an agent (or a generator) to investigate the risky distribution parameters for a given driving algorithm being evaluated. We regard the task algorithm as an environment (or a discriminator) that returns a reward to the agent when a risky scenario is generated. Through the experiments conducted on several scenarios in the simulation, we demonstrate that the proposed framework generates safety-critical scenarios more efficiently than grid search or human design methods. Another advantage of this method is its adaptiveness to the routes and parameters.

Wenhao Ding, Liang He

The goal of acoustic (or sound) events detection (AED or SED) is to predict the temporal position of target events in given audio segments. This task plays a significant role in safety monitoring, acoustic early warning and other scenarios. However, the deficiency of data and diversity of acoustic event sources make the AED task a tough issue, especially for prevalent data-driven methods. In this paper, we start by analyzing acoustic events according to their time-frequency domain properties, showing that different acoustic events have different time-frequency scale characteristics. Inspired by the analysis, we propose an adaptive multi-scale detection (AdaMD) method. By taking advantage of the hourglass neural network and gated recurrent unit (GRU) module, our AdaMD produces multiple predictions at different temporal and frequency resolutions. An adaptive training algorithm is subsequently adopted to combine multi-scale predictions to enhance its overall capability. Experimental results on Detection and Classification of Acoustic Scenes and Events 2017 (DCASE 2017) Task 2, DCASE 2016 Task 3 and DCASE 2017 Task 3 demonstrate that the AdaMD outperforms published state-of-the-art competitors in terms of the metrics of event error rate (ER) and F1-score. The verification experiment on our collected factory mechanical dataset also proves the noise-resistant capability of the AdaMD, providing the possibility for it to be deployed in the complex environment.

Wenhao Ding, Mengdi Xu, Ding Zhao

Naturalistic driving trajectories are crucial for the performance of autonomous driving algorithms. However, most of the data is collected in safe scenarios leading to the duplication of trajectories which are easy to be handled by currently developed algorithms. When considering safety, testing algorithms in near-miss scenarios that rarely show up in off-the-shelf datasets is a vital part of the evaluation. As a remedy, we propose a near-miss data synthesizing framework based on Variational Bayesian methods and term it as Conditional Multiple Trajectory Synthesizer (CMTS). We leverage a generative model conditioned on road maps to bridge safe and collision driving data by representing their distribution in the latent space. By sampling from the near-miss distribution, we can synthesize safety-critical data crucial for understanding traffic scenarios but not shown in neither the original dataset nor the collision dataset. Our experimental results demonstrate that the augmented dataset covers more kinds of driving scenarios, especially the near-miss ones, which help improve the trajectory prediction accuracy and the capability of dealing with risky driving scenarios.

Jingyang Zhang, Wenhao Ding, Jintao Kang et al.

Most attention-based methods only concentrate along the time axis, which is insufficient for Acoustic Event Detection (AED). Meanwhile, previous methods for AED rarely considered that target events possess distinct temporal and frequential scales. In this work, we propose a Multi-Scale Time-Frequency Attention (MTFA) module for AED. MTFA gathers information at multiple resolutions to generate a time-frequency attention mask which tells the model where to focus along both time and frequency axis. With MTFA, the model could capture the characteristics of target events with different scales. We demonstrate the proposed method on Task 2 of Detection and Classification of Acoustic Scenes and Events (DCASE) 2017 Challenge. Our method achieves competitive results on both development dataset and evaluation dataset.

Wenhao Ding, Wenshuo Wang, Ding Zhao

Generating multi-vehicle trajectories from existing limited data can provide rich resources for autonomous vehicle development and testing. This paper introduces a multi-vehicle trajectory generator (MTG) that can encode multi-vehicle interaction scenarios (called driving encounters) into an interpretable representation from which new driving encounter scenarios are generated by sampling. The MTG consists of a bi-directional encoder and a multi-branch decoder. A new disentanglement metric is then developed for model analyses and comparisons in terms of model robustness and the independence of the latent codes. Comparison of our proposed MTG with $β$-VAE and InfoGAN demonstrates that the MTG has stronger capability to purposely generate rational vehicle-to-vehicle encounters through operating the disentangled latent codes. Thus the MTG could provide more data for engineers and researchers to develop testing and evaluation scenarios for autonomous vehicles.

Wenhao Ding, Shuaijun Li, Huihuan Qian

In this paper, we propose a navigation algorithm oriented to multi-agent environment. This algorithm is expressed as a hierarchical framework that contains a Hidden Markov Model (HMM) and a Deep Reinforcement Learning (DRL) structure. For simplification, we term our method Hierarchical Navigation Reinforcement Network (HNRN). In high- level architecture, we train an HMM to evaluate the agent's perception to obtain a score. According to this score, adaptive control action will be chosen. While in low-level architecture, two sub-systems are introduced, one is a differential target- driven system, which aims at heading to the target; the other is a collision avoidance DRL system, which is used for avoiding dynamic obstacles. The advantage of this hierarchical structure is decoupling the target-driven and collision avoidance tasks, leading to a faster and more stable model to be trained. The experiments indicate that our algorithm has higher learning efficiency and rate of success than traditional Velocity Obstacle (VO) algorithms or hybrid DRL method.

Wenhao Ding, Liang He

In this paper, we propose an enhanced triplet method that improves the encoding process of embeddings by jointly utilizing generative adversarial mechanism and multitasking optimization. We extend our triplet encoder with Generative Adversarial Networks (GANs) and softmax loss function. GAN is introduced for increasing the generality and diversity of samples, while softmax is for reinforcing features about speakers. For simplification, we term our method Multitasking Triplet Generative Adversarial Networks (MTGAN). Experiment on short utterances demonstrates that MTGAN reduces the verification equal error rate (EER) by 67% (relatively) and 32% (relatively) over conventional i-vector method and state-of-the-art triplet loss method respectively. This effectively indicates that MTGAN outperforms triplet methods in the aspect of expressing the high-level feature of speaker information.

Wenhao Ding, Shuaijun Li, Guilin Zhang et al.

In this paper, we present an accurate approach to estimate vehicles' pose and shape from off-board multiview images. The images are taken by monocular cameras and have small overlaps. We utilize state-of-the-art convolutional neural networks (CNNs) to extract vehicles' semantic keypoints and introduce a Cross Projection Optimization (CPO) method to estimate the 3D pose. During the iterative CPO process, an adaptive shape adjustment method named Hierarchical Wireframe Constraint (HWC) is implemented to estimate the shape. Our approach is evaluated under both simulated and real-world scenes for performance verification. It's shown that our algorithm outperforms other existing monocular and stereo methods for vehicles' pose and shape estimation. This approach provides a new and robust solution for off-board visual vehicle localization and tracking, which can be applied to massive surveillance camera networks for intelligent transportation.