arXiv subjects

Dimitar Filev

Publications and source records attributed to Dimitar Filev.

At least 19 recordsLinked to original sources

arXiv2026

Toward Unified Multimodal Representation Learning for Autonomous Driving

Contrastive Language-Image Pre-training (CLIP) has shown impressive performance in aligning visual and textual representations. Recent studies have extended this paradigm to 3D vision to improve scene understanding for autonomous driving. A common strategy is to employ pairwise cosine similarity between modalities to guide the training of a 3D encoder. However, considering the similarity between individual modality pairs rather than all modalities jointly fails to ensure consistent and unified alignment across the entire multimodal space. In this paper, we propose a Contrastive Tensor Pre-training (CTP) framework that simultaneously aligns multiple modalities in a unified embedding space to enhance end-to-end autonomous driving. Compared with pairwise cosine similarity alignment, our method extends the 2D similarity matrix into a multimodal similarity tensor. Furthermore, we introduce a tensor loss to enable joint contrastive learning across all modalities. For experimental validation of our framework, we construct a text-image-point cloud triplet dataset derived from existing autonomous driving datasets. The results show that our proposed unified multimodal alignment framework achieves favorable performance for both scenarios: (i) aligning a 3D encoder with pretrained CLIP encoders, and (ii) pretraining all encoders from scratch.

cs.CV↗

arXiv2026

NaviDriveVLM: Decoupling High-Level Reasoning and Motion Planning for Autonomous Driving

Vision-language models (VLMs) have emerged as a promising direction for end-to-end autonomous driving (AD) by jointly modeling visual observations, driving context, and language-based reasoning. However, existing VLM-based systems face a trade-off between high-level reasoning and motion planning: large models offer strong semantic understanding but are costly to adapt for precise control, whereas small VLM models can be fine-tuned efficiently but often exhibit weaker reasoning. We propose NaviDriveVLM, a decoupled framework that separates reasoning from action generation using a large-scale Navigator and a lightweight trainable Driver. This design preserves reasoning ability, reduces training cost, and provides an explicit interpretable intermediate representation for downstream planning. Experiments on the nuScenes benchmark show that NaviDriveVLM outperforms large VLM baselines in end-to-end motion planning.

cs.RO↗

arXiv2025

Taming Spontaneous Stop-and-Go Traffic Waves: A Computational Mechanism Design Perspective

It is well known that stop-and-go waves can be generated spontaneously in traffic even without bottlenecks. Can such undesirable traffic patterns, induced by intrinsic human driving behaviors, be tamed effectively and inexpensively? Taking advantage of emerging connectivity and autonomy technologies, we envision a simple yet realistic traffic control system to achieve this goal. To prove the concept, we design such a system to suppress these waves while maximizing traffic throughput in the Tadaki setting: a circular road with varying number of vehicles. We first introduce our driver behavior model and demonstrate how our calibrated human driving agents can closely reproduce the observed human driving patterns in the original Tadaki experiment. We then propose a simple control system mediated via connected automated vehicles (CAV) whose ideal speed parameter is treated as a system-level control variable adapted to the local vehicle density of the traffic. The objective of the control system is set up as a tradeoff: maximizing throughput while minimizing traffic oscillation. Following computational mechanism design, we search for the optimal control policy as a function of vehicle density and the tradeoff attitude parameter. This can be done by letting all vehicles play a simulated game of CAV-modulated traffic under such a control system. Our simulation results show that the improvements in traffic efficiency and smoothness are substantial. Finally, we envision how such a traffic control system can be realized in an environment with smart vehicles connected to a smart infrastructure or via a scheme of variable speed advisory.

eess.SY↗

arXiv2025

Control of a commercially available vehicle by a tetraplegic human using a brain-computer interface

Brain-computer interfaces (BCIs) read neural signals directly from the brain to infer motor planning and execution. However, the implementation of this technology has been largely limited to laboratory settings, with few real-world applications. We developed a BCI system to drive a vehicle in both simulated and real-world environments. We demonstrate that an individual with tetraplegia, implanted with intracortical BCI electrodes in the posterior parietal cortex (PPC) and the hand knob region of the motor cortex (MC), reacts at least as fast and precisely as motor intact participants. This BCI participant, living in California, could also remotely drive a Ford Mustang Mach-E vehicle in Michigan. Our teledriving tasks relied on cursor movement control for speed and steering in a closed urban test facility and through a predefined obstacle course. These two tasks serve as a proof-of-concept that takes into account the safety and feasibility of BCI-controlled driving. The final BCI system added click control for full-stop braking and thus enabled bimanual cursor-and-click control for simulated town driving with the same proficiency level as the motor intact control group through a virtual town with traffic. This first-of-its-kind implantable BCI application not only highlights the versatility and innovative potentials of BCIs but also illuminates the promising future for the development of life-changing solutions to improve independent mobility for those who suffer catastrophic neurological injury.

eess.SY↗

arXiv2025

Hierarchical Game-Based Multi-Agent Decision-Making for Autonomous Vehicles

This paper develops a game-theoretic decision-making framework for autonomous driving in multi-agent scenarios. A novel hierarchical game-based decision framework is developed for the ego vehicle. This framework features an interaction graph, which characterizes the interaction relationships between the ego and its surrounding traffic agents (including AVs, human driven vehicles, pedestrians, and bicycles, and others), and enables the ego to smartly select a limited number of agents as its game players. Compared to the standard multi-player games, where all surrounding agents are considered as game players, the hierarchical game significantly reduces the computational complexity. In addition, compared to pairwise games, the most popular approach in the literature, the hierarchical game promises more efficient decisions for the ego (in terms of less unnecessary waiting and yielding). To further reduce the computational cost, we then propose an improved hierarchical game, which decomposes the hierarchical game into a set of sub-games. Decision safety and efficiency are analyzed in both hierarchical games. Comprehensive simulation studies are conducted to verify the effectiveness of the proposed frameworks, with an intersection-crossing scenario as a case study.

eess.SY↗

arXiv2025

A Survey of Reinforcement Learning-Based Motion Planning for Autonomous Driving: Lessons Learned from a Driving Task Perspective

Reinforcement learning (RL), with its ability to explore and optimize policies in complex, dynamic decision-making tasks, has emerged as a promising approach to addressing motion planning (MoP) challenges in autonomous driving (AD). Despite rapid advancements in RL and AD, a systematic description and interpretation of the RL design process tailored to diverse driving tasks remains underdeveloped. This survey provides a comprehensive review of RL-based MoP for AD, focusing on lessons from task-specific perspectives. We first outline the fundamentals of RL methodologies, and then survey their applications in MoP, analyzing scenario-specific features and task requirements to shed light on their influence on RL design choices. Building on this analysis, we summarize key design experiences, extract insights from various driving task applications, and provide guidance for future implementations. Additionally, we examine the frontier challenges in RL-based MoP, review recent efforts to addresse these challenges, and propose strategies for overcoming unresolved issues.

cs.LG↗

arXiv2025

Learning Autonomy: Off-Road Navigation Enhanced by Human Input

In the area of autonomous driving, navigating off-road terrains presents a unique set of challenges, from unpredictable surfaces like grass and dirt to unexpected obstacles such as bushes and puddles. In this work, we present a novel learning-based local planner that addresses these challenges by directly capturing human driving nuances from real-world demonstrations using only a monocular camera. The key features of our planner are its ability to navigate in challenging off-road environments with various terrain types and its fast learning capabilities. By utilizing minimal human demonstration data (5-10 mins), it quickly learns to navigate in a wide array of off-road conditions. The local planner significantly reduces the real world data required to learn human driving preferences. This allows the planner to apply learned behaviors to real-world scenarios without the need for manual fine-tuning, demonstrating quick adjustment and adaptability in off-road autonomous driving technology.

cs.RO↗

arXiv2023

Minimum-Time Trajectory Optimization With Data-Based Models: A Linear Programming Approach

In this paper, we develop a computationally-efficient approach to minimum-time trajectory optimization using input-output data-based models, to produce an end-to-end data-to-control solution to time-optimal planning/control of dynamic systems and hence facilitate their autonomous operation. The approach integrates a non-parametric data-based model for trajectory prediction and a continuous optimization formulation based on an exponential weighting scheme for minimum-time trajectory planning. The optimization problem in its final form is a linear program and is easy to solve. We validate the approach and illustrate its application with a spacecraft relative motion planning problem.

eess.SY↗

arXiv2023

Game Projection and Robustness for Game-Theoretic Autonomous Driving

Game-theoretic approaches are envisioned to bring human-like reasoning skills and decision-making processes for autonomous vehicles (AVs). However, challenges including game complexity and incomplete information still remain to be addressed before they can be sufficiently practical for real-world use. Game complexity refers to the difficulties of solving a multi-player game, which include solution existence, algorithm convergence, and scalability. To address these difficulties, a potential game based framework was developed in our recent work. However, conditions on cost function design need to be enforced to make the game a potential game. This paper relaxes the conditions and makes the potential game approach applicable to more general scenarios, even including the ones that cannot be molded as a potential game. Incomplete information refers to the ego vehicle's lack of knowledge of other traffic agents' cost functions. Cost function deviations between the ego vehicle estimated/learned other agents' cost functions and their actual ones are often inevitable. This motivates us to study the robustness of a game-theoretic solution. This paper defines the robustness margin of a game solution as the maximum magnitude of cost function deviations that can be accommodated in a game without changing the optimality of the game solution. With this definition, closed-form robustness margins are derived. Numerical studies using highway lane-changing scenarios are reported.

eess.SY↗

arXiv2023

A Comparison between Markov Chain and Koopman Operator Based Data-Driven Modeling of Dynamical Systems

Markov chain-based modeling and Koopman operator-based modeling are two popular frameworks for data-driven modeling of dynamical systems. They share notable similarities from a computational and practitioner's perspective, especially for modeling autonomous systems. The first part of this paper aims to elucidate these similarities. For modeling systems with control inputs, the models produced by the two approaches differ. The second part of this paper introduces these models and their corresponding control design methods. We illustrate the two approaches and compare them in terms of model accuracy and computational efficiency for both autonomous and controlled systems in numerical examples.

eess.SY↗

arXiv2023

Communication-Efficient Decentralized Multi-Agent Reinforcement Learning for Cooperative Adaptive Cruise Control

Connected and autonomous vehicles (CAVs) promise next-gen transportation systems with enhanced safety, energy efficiency, and sustainability. One typical control strategy for CAVs is the so-called cooperative adaptive cruise control (CACC) where vehicles drive in platoons and cooperate to achieve safe and efficient transportation. In this study, we formulate CACC as a multi-agent reinforcement learning (MARL) problem. Diverging from existing MARL methods that use centralized training and decentralized execution which require not only a centralized communication mechanism but also dense inter-agent communication during training and online adaptation, we propose a fully decentralized MARL framework for enhanced efficiency and scalability. In addition, a quantization-based communication scheme is proposed to reduce the communication overhead without significantly degrading the control performance. This is achieved by employing randomized rounding numbers to quantize each piece of communicated information and only communicating non-zero components after quantization. Extensive experimentation in two distinct CACC settings reveals that the proposed MARL framework consistently achieves superior performance over several contemporary benchmarks in terms of both communication efficiency and control efficacy. In the appendix, we show that our proposed framework's applicability extends beyond CACC, showing promise for broader intelligent transportation systems with intricate action and state spaces.

eess.SY↗

arXiv2023

Targeted collapse regularized autoencoder for anomaly detection: black hole at the center

Autoencoders have been extensively used in the development of recent anomaly detection techniques. The premise of their application is based on the notion that after training the autoencoder on normal training data, anomalous inputs will exhibit a significant reconstruction error. Consequently, this enables a clear differentiation between normal and anomalous samples. In practice, however, it is observed that autoencoders can generalize beyond the normal class and achieve a small reconstruction error on some of the anomalous samples. To improve the performance, various techniques propose additional components and more sophisticated training procedures. In this work, we propose a remarkably straightforward alternative: instead of adding neural network components, involved computations, and cumbersome training, we complement the reconstruction loss with a computationally light term that regulates the norm of representations in the latent space. The simplicity of our approach minimizes the requirement for hyperparameter tuning and customization for new applications which, paired with its permissive data modality constraint, enhances the potential for successful adoption across a broad range of applications. We test the method on various visual and tabular benchmarks and demonstrate that the technique matches and frequently outperforms more complex alternatives. We further demonstrate that implementing this idea in the context of state-of-the-art methods can further improve their performance. We also provide a theoretical analysis and numerical simulations that help demonstrate the underlying process that unfolds during training and how it helps with anomaly detection. This mitigates the black-box nature of autoencoder-based anomaly detection algorithms and offers an avenue for further investigation of advantages, fail cases, and potential new directions.

cs.LG↗

arXiv2023

KARNet: Kalman Filter Augmented Recurrent Neural Network for Learning World Models in Autonomous Driving Tasks

Autonomous driving has received a great deal of attention in the automotive industry and is often seen as the future of transportation. The development of autonomous driving technology has been greatly accelerated by the growth of end-to-end machine learning techniques that have been successfully used for perception, planning, and control tasks. An important aspect of autonomous driving planning is knowing how the environment evolves in the immediate future and taking appropriate actions. An autonomous driving system should effectively use the information collected from the various sensors to form an abstract representation of the world to maintain situational awareness. For this purpose, deep learning models can be used to learn compact latent representations from a stream of incoming data. However, most deep learning models are trained end-to-end and do not incorporate any prior knowledge (e.g., from physics) of the vehicle in the architecture. In this direction, many works have explored physics-infused neural network (PINN) architectures to infuse physics models during training. Inspired by this observation, we present a Kalman filter augmented recurrent neural network architecture to learn the latent representation of the traffic flow using front camera images only. We demonstrate the efficacy of the proposed model in both imitation and reinforcement learning settings using both simulated and real-world datasets. The results show that incorporating an explicit model of the vehicle (states estimated using Kalman filtering) in the end-to-end learning significantly increases performance.

cs.LG↗

arXiv2023

Experience-Based Evolutionary Algorithms for Expensive Optimization

Optimization algorithms are very different from human optimizers. A human being would gain more experiences through problem-solving, which helps her/him in solving a new unseen problem. Yet an optimization algorithm never gains any experiences by solving more problems. In recent years, efforts have been made towards endowing optimization algorithms with some abilities of experience learning, which is regarded as experience-based optimization. In this paper, we argue that hard optimization problems could be tackled efficiently by making better use of experiences gained in related problems. We demonstrate our ideas in the context of expensive optimization, where we aim to find a near-optimal solution to an expensive optimization problem with as few fitness evaluations as possible. To achieve this, we propose an experience-based surrogate-assisted evolutionary algorithm (SAEA) framework to enhance the optimization efficiency of expensive problems, where experiences are gained across related expensive tasks via a novel meta-learning method. These experiences serve as the task-independent parameters of a deep kernel learning surrogate, then the solutions sampled from the target task are used to adapt task-specific parameters for the surrogate. With the help of experience learning, competitive regression-based surrogates can be initialized using only 1$d$ solutions from the target task ($d$ is the dimension of the decision space). Our experimental results on expensive multi-objective and constrained optimization problems demonstrate that experiences gained from related tasks are beneficial for the saving of evaluation budgets on the target problem.

cs.NE↗

arXiv2022

Safe Control and Learning Using Generalized Action Governor

This paper introduces the Generalized Action Governor (AG), a supervisory scheme that augments a nominal closed-loop system with the capability to enforce state and input constraints through online action adjustment. We develop a generalized AG theory for discrete-time systems under bounded uncertainties, and relax the usual requirement of positive invariance to returnability of a safe set. Based on the theory, we present tailored AG design procedures for linear systems and for discrete systems with finite state and action spaces. We further study safe online learning enabled by the AG and present two safe learning strategies, namely safe Q-learning and safe data-driven Koopman operator-based control, both integrated with the AG to guarantee constraint satisfaction during learning. Numerical results illustrate the proposed methods.

eess.SY↗

arXiv2022

Safe and Human-Like Autonomous Driving: A Predictor-Corrector Potential Game Approach

This paper proposes a novel decision-making framework for autonomous vehicles (AVs), called predictor-corrector potential game (PCPG), composed of a Predictor and a Corrector. To enable human-like reasoning and characterize agent interactions, a receding-horizon multi-player game is formulated. To address the challenges caused by the complexity in solving a multi-player game and by the requirement of real-time operation, a potential game (PG) based decision-making framework is developed. In the PG Predictor, the agent cost functions are heuristically predefined. We acknowledge that the behaviors of other traffic agents, e.g., human-driven vehicles and pedestrians, may not necessarily be consistent with the predefined cost functions. To address this issue, a best response-based PG Corrector is designed. In the Corrector, the action deviation between the ego vehicle prediction and the surrounding agent actual behaviors are measured and are fed back to the ego vehicle decision-making, to correct the prediction errors caused by the inaccurate predefined cost functions and to improve the ego vehicle strategies. Distinguished from most existing game-theoretic approaches, this PCPG 1) deals with multi-player games and guarantees the existence of a pure-strategy Nash equilibrium (PSNE), convergence of the PSNE seeking algorithm, and global optimality of the derived PSNE when multiple PSNE exist; 2) is computationally scalable in a multi-agent scenario; 3) guarantees the ego vehicle safety under certain conditions; and 4) approximates the actual PSNE of the system despite the unknown cost functions of others. Comparative studies between the PG, the PCPG, and the control barrier function (CBF) based approaches are conducted in diverse traffic scenarios, including oncoming traffic scenario and multi-vehicle intersection-crossing scenario.

eess.SY↗

arXiv2022

Robust Action Governor for Uncertain Piecewise Affine Systems with Non-convex Constraints and Safe Reinforcement Learning

The action governor is an add-on scheme to a nominal control loop that monitors and adjusts the control actions to enforce safety specifications expressed as pointwise-in-time state and control constraints. In this paper, we introduce the Robust Action Governor (RAG) for systems the dynamics of which can be represented using discrete-time Piecewise Affine (PWA) models with both parametric and additive uncertainties and subject to non-convex constraints. We develop the theoretical properties and computational approaches for the RAG. After that, we introduce the use of the RAG for realizing safe Reinforcement Learning (RL), i.e., ensuring all-time constraint satisfaction during online RL exploration-and-exploitation process. This development enables safe real-time evolution of the control policy and adaptation to changes in the operating environment and system parameters (due to aging, damage, etc.). We illustrate the effectiveness of the RAG in constraint enforcement and safe RL using the RAG by considering their applications to a soft-landing problem of a mass-spring-damper system.

eess.SY↗

arXiv2022

Robust AI Driving Strategy for Autonomous Vehicles

There has been significant progress in sensing, perception, and localization for automated driving, However, due to the wide spectrum of traffic/road structure scenarios and the long tail distribution of human driver behavior, it has remained an open challenge for an intelligent vehicle to always know how to make and execute the best decision on road given available sensing / perception / localization information. In this chapter, we talk about how artificial intelligence and more specifically, reinforcement learning, can take advantage of operational knowledge and safety reflex to make strategical and tactical decisions. We discuss some challenging problems related to the robustness of reinforcement learning solutions and their implications to the practical design of driving strategies for autonomous vehicles. We focus on automated driving on highway and the integration of reinforcement learning, vehicle motion control, and control barrier function, leading to a robust AI driving strategy that can learn and adapt safely.

cs.RO↗