The rapid iteration of autonomous vehicle (AV) deployments leads to increasing needs for building realistic and scalable multi-agent traffic simulators for efficient evaluation. Recent advances in this area focus on closed-loop simulators that enable generating diverse and interactive scenarios. This paper introduces Neural Interactive Agents (NIVA), a probabilistic framework for multi-agent simulation driven by a hierarchical Bayesian model that enables closed-loop, observation-conditioned simulation through autoregressive sampling from a latent, finite mixture of Gaussian distributions. We demonstrate how NIVA unifies preexisting sequence-to-sequence trajectory prediction models and emerging closed-loop simulation models trained on Next-token Prediction (NTP) from a Bayesian inference perspective. Experiments on the Waymo Open Motion Dataset demonstrate that NIVA attains competitive performance compared to the existing method while providing embellishing control over intentions and driving styles.

Multimodal large language models (MLLMs) have shown significant potential in enhancing driving scene understanding and visual question answering (VQA) through advanced logical reasoning capabilities. These tasks support driving action generation and explanation, especially in end-to-end autonomous driving applications. However, deploying these models poses a significant challenge due to their substantial parameter sizes and computational demands, which often exceed onboard computational limits. A key limitation stems from the large number of visual tokens needed to capture detailed, long-context visual information, resulting in increased latency and memory use. To address this, we propose Video Token Sparsification (VTS), a novel approach that leverages redundancy in consecutive video frames to reduce visual tokens while preserving critical information. VTS employs a lightweight CNN-based model to identify key frames and prune less informative tokens, mitigating hallucinations and boosting inference throughput without performance loss. Comprehensive experiments on the LingoQA and DRAMA benchmarks show that VTS achieves up to a 33% improvement in inference throughput and a 28% reduction in memory usage compared to baselines, maintaining comparable performance.

We explore the advantages of using deep learning-based intent sharing for platooning of connected automated vehicles (CAVs). Unlike traditional platooning algorithms that rely on status-sharing - exchanging current position, speed, and acceleration-our approach focuses on intent-sharing, where CAVs share predicted future trajectories with other CAVs. We introduce a deep learning model to generate the intent for each CAV and integrate it into a receding horizon control framework. Our approach aims to minimize spacing errors with the leading vehicle while improving energy efficiency and maintaining string stability. Through microscopic simulations using real-world highway data, we demonstrate that our intent messages significantly enhance energy efficiency compared to conventional status and intent-based platooning algorithms. Moreover, we show that this improvement is particularly pronounced when reducing the frequency of intent message transmission.

Modern cars monitor their surroundings and record video to deter intruders, but current surveillance systems operate independently without communication. Connected vehicles can share detected features of suspicious individuals, improving tracking and alerting approaching drivers before they park in vulnerable spots. This paper explores this use case, where connected vehicles exchange features over the network upon detecting suspicious individuals. Real-world data analysis shows that connected vehicle surveillance improves detection accuracy by 38%.

This study investigates the integration of intent-sharing within a multi-agent platooning system using a reinforcement learning-based control approach. The method enables vehicles to exchange both their current state information and predictions of future speed and acceleration, thereby enhancing decision-making and coordination. By incorporating intent-sharing, the system effectively reduces spacing errors and maintains optimal inter-vehicle spacing, which results in smoother acceleration and deceleration profiles; critical factors for passenger comfort, safety, and platoon string stability. Experimental data confirm that intent-sharing significantly improves string stability, safety, comfort, and gap-keeping accuracy in connected vehicle platooning.

In this paper we investigate the benefits of intent-sharing for platooning of connected automated vehicles (CAVs). Intent sharing is a new class of cooperation for CAVs, where each CAV share their planned future trajectory with other CAVs. In contrast, most existing platooning algorithms, are based on status-sharing cooperation, where CAVs only exchange their current position, speed, and acceleration. We present a receding horizon control framework for CAVs that utilizes intent messages, and minimizes spacing error with respect to the leading vehicle, while considering comfort and string stability. We compare our intent-sharing approach to a baseline CACC algorithm through microscopic simulations based on real-world highway data. Our simulations reveal significant improvements in energy efficiency and road capacity compared to the baseline CACC algorithm.

In this paper, we prove state-of-the-art Bayesian regret bounds for Thompson Sampling in reinforcement learning in a multitude of settings. We present a refined analysis of the information ratio, and show an upper bound of order $\widetilde{O}(H\sqrt{d_{l_1}T})$ in the time inhomogeneous reinforcement learning problem where $H$ is the episode length and $d_{l_1}$ is the Kolmogorov $l_1-$dimension of the space of environments. We then find concrete bounds of $d_{l_1}$ in a variety of settings, such as tabular, linear and finite mixtures, and discuss how our results improve the state-of-the-art.

We study two model selection settings in stochastic linear bandits (LB). In the first setting, which we refer to as feature selection, the expected reward of the LB problem is in the linear span of at least one of $M$ feature maps (models). In the second setting, the reward parameter of the LB problem is arbitrarily selected from $M$ models represented as (possibly) overlapping balls in $\mathbb R^d$. However, the agent only has access to misspecified models, i.e., estimates of the centers and radii of the balls. We refer to this setting as parameter selection. For each setting, we develop and analyze a computationally efficient algorithm that is based on a reduction from bandits to full-information problems. This allows us to obtain regret bounds that are not worse (up to a $\sqrt{\log M}$ factor) than the case where the true model is known. This is the best reported dependence on the number of models $M$ in these settings. Finally, we empirically show the effectiveness of our algorithms using synthetic and real-world experiments.

In this work we investigate meta-learning (or learning-to-learn) approaches in multi-task linear stochastic bandit problems that can originate from multiple environments. Inspired by the work of [1] on meta-learning in a sequence of linear bandit problems whose parameters are sampled from a single distribution (i.e., a single environment), here we consider the feasibility of meta-learning when task parameters are drawn from a mixture distribution instead. For this problem, we propose a regularized version of the OFUL algorithm that, when trained on tasks with labeled environments, achieves low regret on a new task without requiring knowledge of the environment from which the new task originates. Specifically, our regret bound for the new algorithm captures the effect of environment misclassification and highlights the benefits over learning each task separately or meta-learning without recognition of the distinct mixture components.

We study how adding unknown linear safety constraints affects the performance of Thompson Sampling in the linear stochastic bandit problem. The additional constraints must be met at each round in spite of uncertainty about the environment requiring that the learner acts conservatively in choosing her actions. In this setting, we propose Safe-LTS, the first safe Thompson Sampling based algorithm, and we prove that it achieves no-regret learning. We obtain regrets that have the same dependence on the total number of rounds (modulo logarithmic factors) as Safe-UCB, a recently proposed safe algorithm that uses the upper confidence bound principle. Finally, we provide numerical simulations that demonstrate the efficacy of our algorithm.

We study stage-wise conservative linear stochastic bandits: an instance of bandit optimization, which accounts for (unknown) safety constraints that appear in applications such as online advertising and medical trials. At each stage, the learner must choose actions that not only maximize cumulative reward across the entire time horizon, but further satisfy a linear baseline constraint that takes the form of a lower bound on the instantaneous reward. For this problem, we present two novel algorithms, stage-wise conservative linear Thompson Sampling (SCLTS) and stage-wise conservative linear UCB (SCLUCB), that respect the baseline constraints and enjoy probabilistic regret bounds of order $\mathcal{O}(\sqrt{T} \log^{3/2}T)$ and $\mathcal{O}(\sqrt{T} \log T)$, respectively. Notably, the proposed algorithms can be adjusted with only minor modifications to tackle different problem variations, such as, constraints with bandit-feedback, or an unknown sequence of baseline rewards. We discuss these and other improvements over the state-of-the art. For instance, compared to existing solutions, we show that SCLTS plays the (non-optimal) baseline action at most $\mathcal{O}(\log{T})$ times (compared to $\mathcal{O}(\sqrt{T})$). Finally, we make connections to another studied form of safety-constraints that takes the form of an upper bound on the instantaneous reward. While this incurs additional complexity to the learning process as the optimal action is not guaranteed to belong to the safe-set at each round, we show that SCLUCB can properly adjust in this setting via a simple modification.