MTDrive: Multi-turn Interactive Reinforcement Learning for Autonomous Driving

Trajectory planning is a core task in autonomous driving, requiring the prediction of safe and comfortable paths across diverse scenarios. Integrating Multi-modal Large Language Models (MLLMs) with Reinforcement Learning (RL) has shown promise in addressing "long-tail" scenarios. However, existing methods are constrained to single-turn reasoning, limiting their ability to handle complex tasks requiring iterative refinement. To overcome this limitation, we present MTDrive, a multi-turn framework that enables MLLMs to iteratively refine trajectories based on environmental feedback. MTDrive introduces Multi-Turn Group Relative Policy Optimization (mtGRPO), which mitigates reward sparsity by computing relative advantages across turns. We further construct an interactive trajectory understanding dataset from closed-loop simulation to support multi-turn training. Experiments on the NAVSIM benchmark demonstrate superior performance compared to existing methods, validating the effectiveness of our multi-turn reasoning paradigm. Additionally, we implement system-level optimizations to reduce data transfer overhead caused by high-resolution images and multi-turn sequences, achieving 2.5x training throughput. Our data, models, and code will be made available soon.

Solving Inverse Problems with Flow-based Models via Model Predictive Control

Flow-based generative models provide strong unconditional priors for inverse problems, but guiding their dynamics for conditional generation remains challenging. Recent work casts training-free conditional generation in flow models as an optimal control problem; however, solving the resulting trajectory optimisation is computationally and memory intensive, requiring differentiation through the flow dynamics or adjoint solves. We propose MPC-Flow, a model predictive control framework that formulates inverse problem solving with flow-based generative models as a sequence of control sub-problems, enabling practical optimal control-based guidance at inference time. We provide theoretical guarantees linking MPC-Flow to the underlying optimal control objective and show how different algorithmic choices yield a spectrum of guidance algorithms, including regimes that avoid backpropagation through the generative model trajectory. We evaluate MPC-Flow on benchmark image restoration tasks, spanning linear and non-linear settings such as in-painting, deblurring, and super-resolution, and demonstrate strong performance and scalability to massive state-of-the-art architectures via training-free guidance of FLUX.2 (32B) in a quantised setting on consumer hardware.

Drive-JEPA: Video JEPA Meets Multimodal Trajectory Distillation for End-to-End Driving

End-to-end autonomous driving increasingly leverages self-supervised video pretraining to learn transferable planning representations. However, pretraining video world models for scene understanding has so far brought only limited improvements. This limitation is compounded by the inherent ambiguity of driving: each scene typically provides only a single human trajectory, making it difficult to learn multimodal behaviors. In this work, we propose Drive-JEPA, a framework that integrates Video Joint-Embedding Predictive Architecture (V-JEPA) with multimodal trajectory distillation for end-to-end driving. First, we adapt V-JEPA for end-to-end driving, pretraining a ViT encoder on large-scale driving videos to produce predictive representations aligned with trajectory planning. Second, we introduce a proposal-centric planner that distills diverse simulator-generated trajectories alongside human trajectories, with a momentum-aware selection mechanism to promote stable and safe behavior. When evaluated on NAVSIM, the V-JEPA representation combined with a simple transformer-based decoder outperforms prior methods by 3 PDMS in the perception-free setting. The complete Drive-JEPA framework achieves 93.3 PDMS on v1 and 87.8 EPDMS on v2, setting a new state-of-the-art.

Li-ViP3D++: Query-Gated Deformable Camera-LiDAR Fusion for End-to-End Perception and Trajectory Prediction

End-to-end perception and trajectory prediction from raw sensor data is one of the key capabilities for autonomous driving. Modular pipelines restrict information flow and can amplify upstream errors. Recent query-based, fully differentiable perception-and-prediction (PnP) models mitigate these issues, yet the complementarity of cameras and LiDAR in the query-space has not been sufficiently explored. Models often rely on fusion schemes that introduce heuristic alignment and discrete selection steps which prevent full utilization of available information and can introduce unwanted bias. We propose Li-ViP3D++, a query-based multimodal PnP framework that introduces Query-Gated Deformable Fusion (QGDF) to integrate multi-view RGB and LiDAR in query space. QGDF (i) aggregates image evidence via masked attention across cameras and feature levels, (ii) extracts LiDAR context through fully differentiable BEV sampling with learned per-query offsets, and (iii) applies query-conditioned gating to adaptively weight visual and geometric cues per agent. The resulting architecture jointly optimizes detection, tracking, and multi-hypothesis trajectory forecasting in a single end-to-end model. On nuScenes, Li-ViP3D++ improves end-to-end behavior and detection quality, achieving higher EPA (0.335) and mAP (0.502) while substantially reducing false positives (FP ratio 0.147), and it is faster than the prior Li-ViP3D variant (139.82 ms vs. 145.91 ms). These results indicate that query-space, fully differentiable camera-LiDAR fusion can increase robustness of end-to-end PnP without sacrificing deployability.

Unsupervised Anomaly Detection in Multi-Agent Trajectory Prediction via Transformer-Based Models

Identifying safety-critical scenarios is essential for autonomous driving, but the rarity of such events makes supervised labeling impractical. Traditional rule-based metrics like Time-to-Collision are too simplistic to capture complex interaction risks, and existing methods lack a systematic way to verify whether statistical anomalies truly reflect physical danger. To address this gap, we propose an unsupervised anomaly detection framework based on a multi-agent Transformer that models normal driving and measures deviations through prediction residuals. A dual evaluation scheme has been proposed to assess both detection stability and physical alignment: Stability is measured using standard ranking metrics in which Kendall Rank Correlation Coefficient captures rank agreement and Jaccard index captures the consistency of the top-K selected items; Physical alignment is assessed through correlations with established Surrogate Safety Measures (SSM). Experiments on the NGSIM dataset demonstrate our framework's effectiveness: We show that the maximum residual aggregator achieves the highest physical alignment while maintaining stability. Furthermore, our framework identifies 388 unique anomalies missed by Time-to-Collision and statistical baselines, capturing subtle multi-agent risks like reactive braking under lateral drift. The detected anomalies are further clustered into four interpretable risk types, offering actionable insights for simulation and testing.

Context Structure Reshapes the Representational Geometry of Language Models

Large Language Models (LLMs) have been shown to organize the representations of input sequences into straighter neural trajectories in their deep layers, which has been hypothesized to facilitate next-token prediction via linear extrapolation. Language models can also adapt to diverse tasks and learn new structure in context, and recent work has shown that this in-context learning (ICL) can be reflected in representational changes. Here we bring these two lines of research together to explore whether representation straightening occurs \emph{within} a context during ICL. We measure representational straightening in Gemma 2 models across a diverse set of in-context tasks, and uncover a dichotomy in how LLMs' representations change in context. In continual prediction settings (e.g., natural language, grid world traversal tasks) we observe that increasing context increases the straightness of neural sequence trajectories, which is correlated with improvement in model prediction. Conversely, in structured prediction settings (e.g., few-shot tasks), straightening is inconsistent -- it is only present in phases of the task with explicit structure (e.g., repeating a template), but vanishes elsewhere. These results suggest that ICL is not a monolithic process. Instead, we propose that LLMs function like a Swiss Army knife: depending on task structure, the LLM dynamically selects between strategies, only some of which yield representational straightening.

Disturbance-Aware Flight Control of Robotic Gliding Blimp via Moving Mass Actuation

Robotic blimps, as lighter-than-air (LTA) aerial systems, offer long endurance and inherently safe operation but remain highly susceptible to wind disturbances. Building on recent advances in moving mass actuation, this paper addresses the lack of disturbance-aware control frameworks for LTA platforms by explicitly modeling and compensating for wind-induced effects. A moving horizon estimator (MHE) infers real-time wind perturbations and provides these estimates to a model predictive controller (MPC), enabling robust trajectory and heading regulation under varying wind conditions. The proposed approach leverages a two-degree-of-freedom (2-DoF) moving-mass mechanism to generate both inertial and aerodynamic moments for attitude and heading control, thereby enhancing flight stability in disturbance-prone environments. Extensive flight experiments under headwind and crosswind conditions show that the integrated MHE-MPC framework significantly outperforms baseline PID control, demonstrating its effectiveness for disturbance-aware LTA flight.

Don't double it: Efficient Agent Prediction in Occlusions

Occluded traffic agents pose a significant challenge for autonomous vehicles, as hidden pedestrians or vehicles can appear unexpectedly, yet this problem remains understudied. Existing learning-based methods, while capable of inferring the presence of hidden agents, often produce redundant occupancy predictions where a single agent is identified multiple times. This issue complicates downstream planning and increases computational load. To address this, we introduce MatchInformer, a novel transformer-based approach that builds on the state-of-the-art SceneInformer architecture. Our method improves upon prior work by integrating Hungarian Matching, a state-of-the-art object matching algorithm from object detection, into the training process to enforce a one-to-one correspondence between predictions and ground truth, thereby reducing redundancy. We further refine trajectory forecasts by decoupling an agent's heading from its motion, a strategy that improves the accuracy and interpretability of predicted paths. To better handle class imbalances, we propose using the Matthews Correlation Coefficient (MCC) to evaluate occupancy predictions. By considering all entries in the confusion matrix, MCC provides a robust measure even in sparse or imbalanced scenarios. Experiments on the Waymo Open Motion Dataset demonstrate that our approach improves reasoning about occluded regions and produces more accurate trajectory forecasts than prior methods.

Do Pathology Foundation Models Encode Disease Progression? A Pseudotime Analysis of Visual Representations

Vision foundation models trained on discretely sampled images achieve strong performance on classification benchmarks, yet whether their representations encode the continuous processes underlying their training data remains unclear. This question is especially pertinent in computational pathology, where we posit that models whose latent representations implicitly capture continuous disease progression may better reflect underlying biology, support more robust generalization, and enable quantitative analyses of features associated with disease transitions. Using diffusion pseudotime, a method developed to infer developmental trajectories from single-cell transcriptomics, we probe whether foundation models organize disease states along coherent progression directions in representation space. Across four cancer progressions and six models, we find that all pathology-specific models recover trajectory orderings significantly exceeding null baselines, with vision-only models achieving the highest fidelities $(τ> 0.78$ on CRC-Serrated). Model rankings by trajectory fidelity on reference diseases strongly predict few-shot classification performance on held-out diseases ($ρ= 0.92$), and exploratory analysis shows cell-type composition varies smoothly along inferred trajectories in patterns consistent with known stromal remodeling. Together, these results demonstrate that vision foundation models can implicitly learn to represent continuous processes from independent static observations, and that trajectory fidelity provides a complementary measure of representation quality beyond downstream performance. While demonstrated in pathology, this framework could be applied to other domains where continuous processes are observed through static snapshots.

The Patient is not a Moving Document: A World Model Training Paradigm for Longitudinal EHR

Large language models (LLMs) trained with next-word-prediction have achieved success as clinical foundation models. Representations from these language backbones yield strong linear probe performance across biomedical tasks, suggesting that patient semantics emerge from next-token prediction at scale. However, this paradigm treats patients as a document to be summarized rather than a dynamical system to be simulated; a patient's trajectory emerges from their state evolving under interventions and time, requiring models that simulate dynamics rather than predict tokens. To address this, we introduce SMB-Structure, a world model for structured EHR that grounds a joint-embedding prediction architecture (JEPA) with next-token prediction (SFT). SFT grounds our model to reconstruct future patient states in token space, while JEPA predicts those futures in latent space from the initial patient representation alone, forcing trajectory dynamics to be encoded before the next state is observed. We validate across two large-scale cohorts: Memorial Sloan Kettering (23,319 oncology patients; 323,000+ patient-years) and INSPECT (19,402 pulmonary embolism patients). Using a linear probe evaluated at multiple points along the disease trajectory, we demonstrate that our training paradigm learns embeddings that capture disease dynamics not recoverable by autoregressive baselines, enabling SMB-Structure to achieve competitive performance on complex tasks characterized by high patient heterogeneity. Model weights are available at https://huggingface.co/standardmodelbio/SMB-v1-1.7B-Structure.