arXiv subjects

Mintae Kim

Publications and source records attributed to Mintae Kim.

13 recordsLinked to original sources

arXiv2026

Do Current Retrievers Cover All the Evidence? A Controlled Study of Conjunctive Cross-Page Retrieval

Finding a long document relevant to a multi-part request is not the same as establishing that it contains every requested piece of evidence. We study this gap for conjunctive document retrieval, where two or three explicit conditions must be supported on different pages of one document. We use n-Clue as a controlled measurement instrument: 1{,}000 queries over 2{,}021 documents pair all-condition golds with naturally occurring documents that satisfy only a subset, and a complete-first success requires a top-10 gold to precede every released subset qrel. Across 70 configurations, condition-wise decomposition improves two dense backbones by 6.8--7.3 points and lexical--visual fusion adds 8.7, while four generic rerankers all reduce Gold-NDCG; these directions replicate on a four-source stress set. Scaling one dense family from 0.6B to 8B changes complete-first success by 0.0 points. The strongest displayed hybrid illustrates the resulting gap: it finds a gold for 81.1\% of queries but succeeds complete-first on only 35.8\%, and the gap persists across condition count, target length, candidate density, query rendering, and the four-source stress set. Finally, page-aware visual systems surface stored support for every condition on only 5.1--5.3\% of queries. These results identify condition coverage, rather than gold discovery alone, as the central bottleneck.

cs.CV↗

arXiv2026

Trajectory-Regularized Stochastic Optimal Control via KL Divergence

We introduce trajectory-regularized stochastic optimal control (TRSOC), which augments standard stochastic optimal control (SOC) with a Kullback--Leibler (KL) divergence between controlled and reference trajectory distributions. Using Girsanov's theorem, the trajectory KL reduces to a quadratic drift mismatch penalty, yielding a modified running cost that preserves the dynamic programming (DP) structure. We derive the corresponding Hamilton--Jacobi--Bellman (HJB) equation and characterize the optimal policy. In the linear-quadratic (LQ) setting, the formulation admits a closed-form solution with an augmented control cost. Experiments show that the regularization parameter induces a trade-off between performance-driven and reference-preserving behavior, including cases with reference dynamics learned from offline data.

eess.SY↗

arXiv2026

Robust Adversarial Policy Optimization Under Dynamics Uncertainty

Reinforcement learning (RL) policies often fail under dynamics that differ from training, a gap not fully addressed by domain randomization or existing adversarial RL methods. Distributionally robust RL provides a formal remedy but still relies on surrogate adversaries to approximate intractable primal problems, leaving blind spots that potentially cause instability and over-conservatism. We propose a dual formulation that directly exposes the robustness-performance trade-off. At the trajectory level, a temperature parameter from the dual problem is approximated with an adversarial network, yielding efficient and stable worst-case rollouts within a divergence bound. At the model level, we employ Boltzmann reweighting over dynamics ensembles, focusing on more adverse environments to the current policy rather than uniform sampling. The two components act independently and complement each other: trajectory-level steering ensures robust rollouts, while model-level sampling provides policy-sensitive coverage of adverse dynamics. The resulting framework, robust adversarial policy optimization (RAPO) outperforms robust RL baselines, improving resilience to uncertainty and generalization to out-of-distribution dynamics while maintaining dual tractability.

cs.LG↗

arXiv2026

WOMBET: World Model-Based Experience Transfer for Robust and Sample-efficient Reinforcement Learning

Reinforcement learning (RL) in robotics is often limited by the cost and risk of data collection, motivating experience transfer from a source task to a target task. Offline-to-online RL leverages prior data but typically assumes a given fixed dataset and does not address how to generate reliable data for transfer. We propose World Model-Based Experience Transfer (WOMBET), a framework that jointly generates and utilizes prior data. WOMBET learns a world model in the source task and generates offline data via uncertainty-penalized planning, followed by filtering trajectories with high return and low epistemic uncertainty. It then performs online fine-tuning in the target task using adaptive sampling between offline and online data, enabling a stable transition from prior-driven initialization to task-specific adaptation. We show that the uncertainty-penalized objective provides a lower bound on the true return and derive a finite-sample error decomposition capturing distribution mismatch and approximation error. Empirically, WOMBET improves sample efficiency and final performance over strong baselines on continuous control benchmarks, demonstrating the benefit of jointly optimizing data generation and transfer.

cs.LG↗

arXiv2026

ReinPool: Reinforcement Learning Pooling Multi-Vector Embeddings for Retrieval System

Multi-vector embedding models have emerged as a powerful paradigm for document retrieval, preserving fine-grained visual and textual details through token-level representations. However, this expressiveness comes at a staggering cost: storing embeddings for every token inflates index sizes by over $1000\times$ compared to single-vector approaches, severely limiting scalability. We introduce \textbf{ReinPool}, a reinforcement learning framework that learns to dynamically filter and pool multi-vector embeddings into compact, retrieval-optimized representations. By training with an inverse retrieval objective and NDCG-based rewards, ReinPool identifies and retains only the most discriminative vectors without requiring manual importance annotations. On the Vidore V2 benchmark across three vision-language embedding models, ReinPool compresses multi-vector representations by $746$--$1249\times$ into single vectors while recovering 76--81\% of full multi-vector retrieval performance. Compared to static mean pooling baselines, ReinPool achieves 22--33\% absolute NDCG@3 improvement, demonstrating that learned selection significantly outperforms heuristic aggregation.

cs.IR↗

arXiv2026

Finite Memory Belief Approximation for Optimal Control in Partially Observable Markov Decision Processes

We study finite memory belief approximation for partially observable (PO) stochastic optimal control (SOC) problems. While belief states are sufficient for SOC in partially observable Markov decision processes (POMDPs), they are generally infinite-dimensional and impractical. We interpret truncated input-output (IO) histories as inducing a belief approximation and develop a metric-based theory that directly relates information loss to control performance. Using the Wasserstein metric, we derive policy-conditional performance bounds that quantify value degradation induced by finite memory along typical closed-loop trajectories. Our analysis proceeds via a fixed-policy comparison: we evaluate two cost functionals under the same closed-loop execution and isolate the effect of replacing the true belief by its finite memory approximation inside the belief-level cost. For linear quadratic Gaussian (LQG) systems, we provide closed-form belief mismatch evaluation and empirically validate the predicted mechanism, demonstrating that belief mismatch decays approximately exponentially with memory length and that the induced performance mismatch scales accordingly. Together, these results provide a metric-aware characterization of what finite memory belief approximation can and cannot achieve in PO settings.

eess.SY↗

arXiv2025

RoVerFly: Robust and Versatile Implicit Hybrid Control of Quadrotor-Payload Systems

Designing robust controllers for precise trajectory tracking with quadrotors is challenging due to nonlinear dynamics and underactuation, and becomes harder with flexible cable-suspended payloads that add degrees of freedom and hybrid dynamics. Classical model-based methods offer stability guarantees but require extensive tuning and often fail to adapt when the configuration changes-when a payload is added or removed, or when its mass or cable length varies. We present RoVerFly, a unified learning-based control framework where a single reinforcement learning (RL) policy functions as an implicit hybrid controller, managing complex dynamics without explicit mode detection or controller switching. Trained with task and domain randomization, the controller is resilient to disturbances and varying dynamics. It achieves strong zero-shot generalization across payload settings-including no payload as well as varying mass and cable length-without re-tuning, while retaining the interpretability and structure of a feedback tracking controller. Code and supplementary materials are available at https://github.com/mintaeshkim/roverfly.

cs.RO↗

arXiv2025

Estimation of Aerodynamics Forces in Dynamic Morphing Wing Flight

Accurate estimation of aerodynamic forces is essential for advancing the control, modeling, and design of flapping-wing aerial robots with dynamic morphing capabilities. In this paper, we investigate two distinct methodologies for force estimation on Aerobat, a bio-inspired flapping-wing platform designed to emulate the inertial and aerodynamic behaviors observed in bat flight. Our goal is to quantify aerodynamic force contributions during tethered flight, a crucial step toward closed-loop flight control. The first method is a physics-based observer derived from Hamiltonian mechanics that leverages the concept of conjugate momentum to infer external aerodynamic forces acting on the robot. This observer builds on the system's reduced-order dynamic model and utilizes real-time sensor data to estimate forces without requiring training data. The second method employs a neural network-based regression model, specifically a multi-layer perceptron (MLP), to learn a mapping from joint kinematics, flapping frequency, and environmental parameters to aerodynamic force outputs. We evaluate both estimators using a 6-axis load cell in a high-frequency data acquisition setup that enables fine-grained force measurements during periodic wingbeats. The conjugate momentum observer and the regression model demonstrate strong agreement across three force components (Fx, Fy, Fz).

cs.RO↗

arXiv2025

Annotation-Free Reinforcement Learning Query Rewriting via Verifiable Search Reward

Optimizing queries for Retrieval-Augmented Generation (RAG) systems poses a significant challenge, particularly across diverse modal indices. We introduce RL-QR, a novel annotation-free reinforcement learning framework for query rewriting that eliminates the need for costly human-annotated data. By leveraging verifiable search rewards derived from index-aligned synthetic queries, RL-QR overcomes human-annotation dependencies, extending its applicability to various modalities and index domains. Experimental results demonstrate the framework's robustness, achieving substantial retrieval performance gains of up to 3.9$\times$ on lexical retrievers and 3.5$\times$ on semantic retrievers on the MTEB VIDORE V2 benchmark for unstructured visual documents, along with consistent 5\% to 10\% improvements on MS MARCO v2.1 and internal industrial datasets.

cs.CV↗

arXiv2025

ixi-GEN: Efficient Industrial sLLMs through Domain Adaptive Continual Pretraining

The emergence of open-source large language models (LLMs) has expanded opportunities for enterprise applications; however, many organizations still lack the infrastructure to deploy and maintain large-scale models. As a result, small LLMs (sLLMs) have become a practical alternative despite inherent performance limitations. While Domain Adaptive Continual Pretraining (DACP) has been explored for domain adaptation, its utility in commercial settings remains under-examined. In this study, we validate the effectiveness of a DACP-based recipe across diverse foundation models and service domains, producing DACP-applied sLLMs (ixi-GEN). Through extensive experiments and real-world evaluations, we demonstrate that ixi-GEN models achieve substantial gains in target-domain performance while preserving general capabilities, offering a cost-efficient and scalable solution for enterprise-level deployment.

cs.CL↗

arXiv2024

Learning-based Trajectory Tracking for Bird-inspired Flapping-Wing Robots

Bird-sized flapping-wing robots offer significant potential for agile flight in complex environments, but achieving agile and robust trajectory tracking remains a challenge due to the complex aerodynamics and highly nonlinear dynamics inherent in flapping-wing flight. In this work, a learning-based control approach is introduced to unlock the versatility and adaptiveness of flapping-wing flight. We propose a model-free reinforcement learning (RL)-based framework for a high degree-of-freedom (DoF) bird-inspired flapping-wing robot that allows for multimodal flight and agile trajectory tracking. Stability analysis was performed on the closed-loop system comprising of the flapping-wing system and the RL policy. Additionally, simulation results demonstrate that the RL-based controller can successfully learn complex wing trajectory patterns, achieve stable flight, switch between flight modes spontaneously, and track different trajectories under various aerodynamic conditions.

cs.RO↗

arXiv2024

Large Language Models for Knowledge-Free Network Management: Feasibility Study and Opportunities

Traditional network management algorithms have relied on prior knowledge of system models and networking scenarios. In practice, a universal optimization framework is desirable where a sole optimization module can be readily applied to arbitrary network management tasks without any knowledge of the system. To this end, knowledge-free optimization techniques are necessary whose operations are independent of scenario-specific information including objective functions, system parameters, and network setups. The major challenge of this paradigm-shifting approach is the requirement of a hyper-intelligent black-box optimizer that can establish efficient decision-making policies using its internal reasoning capabilities. This article presents a novel knowledge-free network management paradigm with the power of foundation models called large language models (LLMs). Trained on vast amounts of datasets, LLMs can understand important contexts from input prompts containing minimal system information, thereby offering remarkable inference performance even for entirely new tasks. Pretrained LLMs can be potentially leveraged as foundation models for versatile network optimization. By eliminating the dependency on prior knowledge, LLMs can be seamlessly applied for various network management tasks. The viability of this approach is demonstrated for resource management problems using GPT-3.5-Turbo. Numerical results validate that knowledge-free LLM optimizers are able to achieve comparable performance to existing knowledge-based optimization algorithms.

cs.NI↗

arXiv2024

Cooperative Multi-Agent Deep Reinforcement Learning Methods for UAV-aided Mobile Edge Computing Networks

This paper presents a cooperative multi-agent deep reinforcement learning (MADRL) approach for unmmaned aerial vehicle (UAV)-aided mobile edge computing (MEC) networks. An UAV with computing capability can provide task offlaoding services to ground internet-of-things devices (IDs). With partial observation of the entire network state, the UAV and the IDs individually determine their MEC strategies, i.e., UAV trajectory, resource allocation, and task offloading policy. This requires joint optimization of decision-making process and coordination strategies among the UAV and the IDs. To address this difficulty, the proposed cooperative MADRL approach computes two types of action variables, namely message action and solution action, each of which is generated by dedicated actor neural networks (NNs). As a result, each agent can automatically encapsulate its coordination messages to enhance the MEC performance in the decentralized manner. The proposed actor structure is designed based on graph attention networks such that operations are possible regardless of the number of IDs. A scalable training algorithm is also proposed to train a group of NNs for arbitrary network configurations. Numerical results demonstrate the superiority of the proposed cooperative MADRL approach over conventional methods.

cs.IT↗