arXiv subjects

SooJean Han

Publications and source records attributed to SooJean Han.

18 recordsLinked to original sources

arXiv2026

Graph-Guided Safe Diffuser: Topological Graph Guidance for Safe Diffusion Planning

Many diffusion-based planners enforce safety through inference-time guidance, but such interleaved trajectory deformations often degrade kinematic feasibility due to manifold rupture. We propose Graph-Guided Safe Diffuser (G2SD), a hierarchical framework that leverages a high-level topological graph planner to guide a low-level diffusion model. G2SD enforces safety at a structural level by abstracting the data manifold into a learned latent graph, on which high-level planning is performed. Continuous trajectories are generated by diffusion planners, which are conditioned on the graph node representations selected by the high-level planner. Theoretical analyses demonstrate conditions under which manifold rupture occurs in diffusion planners, and show that G2SD improves safety by reducing the constraint violation probability as the number of segments increases. Experiments demonstrate that G2SD substantially outperforms baselines, increasing goal-reaching rate without any collision from 40-50% to 98% in Maze2D navigation and also achieving superior task scores in locomotion.

cs.RO↗

arXiv2026

Chance-Constrained Nonlinear Covariance Control via Robust Linearization Remainder Bounds

When dealing with nonlinear systems, classical covariance steering typically propagates uncertainty via first-order linearizations, discarding higher-order Taylor remainders. This truncation causes computed statistical moments to diverge from the true physical state distribution, often leading to chance constraint violations. This paper introduces a discrete-time Sequential Convex Programming (SCP) framework that casts the deterministic one-step nonlinear numerical map as a Linear Stochastic Inclusion. The Taylor remainder is bounded within an unstructured uncertainty block over a uniform envelope. The second-moment tubes are propagated via what we refer to as a robust Stochastic Linear Matrix Inequality (S-LMI) derived from the Petersen's lemma, providing an upper bound on the expected uncentered second moment. Domain-exit risk is bounded analytically via a Markov trace inequality, and spatial chance constraints are enforced via Gauss unimodal second-moment bounds within a Difference-of-Convex program. Simulations on a state-dependent nonlinear dynamic system demonstrate constraint satisfaction.

eess.SY↗

arXiv2026

Scalable Sensor Scheduling for Continuous-Discrete Kalman Filtering via Information-Form Surrogate Dynamics

We study sensor scheduling for continuous-discrete Kalman filtering with Poisson measurement arrivals and propose an information-form deterministic surrogate for scalable offline design. Unlike the covariance-form surrogate, the sensing rates enter through sensor-specific additive information increments, eliminating mixed state-input derivatives in the transcribed nonlinear program and thereby yielding a simpler derivative structure. We further show that, together with the covariance-form surrogate, the proposed surrogate provides computable two-sided performance bounds for a given schedule under stochastic measurement arrivals. Numerical experiments demonstrate substantial computational savings, especially in many-sensor settings, while retaining comparable realized Monte Carlo performance and providing computable two-sided performance bounds for the returned schedule.

eess.SY↗

arXiv2026

Stratified Hazard Sampling: Minimal-Variance Event Scheduling for CTMC/DTMC Discrete Diffusion and Flow Models

Uniform-noise discrete diffusion and flow models (e.g., D3PM, SEDD, UDLM, DFM) generate sequences non-autoregressively by iteratively refining randomly initialized vocabulary tokens through multiple context-dependent replacements. These models are typically formulated as time-inhomogeneous CTMC/DTMC processes and sampled using independent Bernoulli change decisions at each discretization step. This induces Poisson-binomial variance in per-position jump counts that grows with the number of required edits, leading to the characteristic under-editing (residual noise) and over-editing (cascading substitutions) failure modes that degrade sample quality, especially under tight discretization budgets. In contrast, absorbing-state (mask-start) models avoid this instability by allowing each position to jump at most once. We propose Stratified Hazard Sampling (SHS), a training-free, drop-in, and hyperparameter-free inference principle for any sampler that admits a stay-vs.-replace decomposition. SHS models per-token edits as events driven by cumulative hazard (CTMC) or cumulative jump mass (DTMC) and places events by stratifying this cumulative quantity: with a single random phase per position, a token is updated whenever its accumulated hazard crosses unit-spaced thresholds. This preserves the expected number of jumps while achieving the minimum possible conditional variance among unbiased integer estimators (bounded by 1/4 for any fixed cumulative mass), without altering per-jump destination sampling and thus retaining multimodality. Experiments on uniform-noise discrete diffusion language models show that SHS consistently improves sample quality. We further show that SHS improves robustness under token-level blacklist filtering, with benefits increasing as lexical constraints grow more severe.

cs.LG↗

arXiv2025

CFG-OEC: Classifier Free Guidance with Orthogonal Error Correction

Classifier free guidance is a standard method for conditional sampling in diffusion models, but its sampling rule is not aligned with the objective used in training. This mismatch induces a structural sampling error through the interaction of conditional and unconditional prediction errors. We analyze this issue by decomposing the sampling error into a base term and a cross term determined by the alignment of the two errors. Based on this analysis we propose CFG with orthogonal error correction (CFG-OEC), a structural modification that reduces the interaction term. For practical settings where ground truth noise is not observable, we introduce a proxy computed from model predictions and a dynamic method that stabilizes correction across diffusion timesteps. Experiments in a controlled environment validate our theoretical error decomposition and proxy construction. Image generation on Stable Diffusion v1.5 and Stable Diffusion XL show that CFG-OEC improves FID and CLIP scores over CFG and CFG++ across multiple samplers and guidance regimes.

cs.LG↗

arXiv2025

SafeFlowMatcher: Safe and Fast Planning using Flow Matching with Control Barrier Functions

Generative planners based on flow matching (FM) produce high-quality paths in a single or a few ODE steps, but their sampling dynamics offer no formal safety guarantees and can yield incomplete paths near constraints. We present SafeFlowMatcher, a planning framework that couples FM with control barrier functions (CBFs) to achieve both real-time efficiency and certified safety. SafeFlowMatcher uses a two-phase (PC) integrator: (i) a prediction phase integrates the learned FM once (or a few steps) to obtain a candidate path without intervention; (ii) a correction phase refines this path with a vanishing time-scaled vector field and a CBF-based quadratic program that minimally perturbs the vector field. We prove a barrier certificate for the resulting flow system, establishing forward invariance of a robust safe set and finite-time convergence to the safe set. In addition, by enforcing safety only on the executed path, rather than all intermediate latent paths, SafeFlowMatcher avoids distributional drift and mitigates local trap problems. Moreover, SafeFlowMatcher attains faster, smoother, and safer paths than diffusion- and FM-based baselines on maze navigation, locomotion, and robot manipulation tasks. Extensive ablations corroborate the contributions of the PC integrator and the barrier certificate.

cs.RO↗

arXiv2025

Heterogeneous Predictor-based Risk-Aware Planning with Conformal Prediction in Dense, Uncertain Environments

Real-time planning among many uncertain, dynamic obstacles is challenging because predicting every agent with high fidelity is both unnecessary and computationally expensive. We present Heterogeneous Predictor-based Risk-Aware Planning (H-PRAP), a framework that allocates prediction effort to where it matters. H-PRAP introduces the Probability-based Collision Risk Index (P-CRI), a closed-form, horizon-level collision index obtained by calibrating a Gaussian surrogate with conformal prediction. P-CRI drives a router that assigns high-risk obstacles to accurate but expensive predictors and low-risk obstacles to lightweight predictors, while preserving distribution-free coverage across heterogeneous predictors through conformal prediction. The selected predictions and their conformal radii are embedded in a chance-constrained model predictive control (MPC) problem, yielding receding-horizon policies with explicit safety margins. We analyze the safety-efficiency trade-off under prediction compute budget: more portion of low-fidelity predictions reduce residual risk from dropped obstacles, but in the same time induces larger conformal radii and degrades trajectory efficiency and shrinks MPC feasibility. Extensive numerical simulations in dense, uncertain environments validate that H-PRAP attains best balance between trajectory success rate (i.e., no collisions) and the time to reach the goal (i.e., trajectory efficiency) compared to single prediction architectures.

cs.RO↗

arXiv2025

Advantages of Feedback in Distributed Data-Gathering for Accurate and Power-Efficient State-Estimation

In distributed target-tracking sensor networks, efficient data gathering methods are necessary to save communication resources and assure information accuracy. This paper proposes a Feedback (FB) distributed data-gathering method which lets the central unit feed information back to the mobile sensors; each sensor then uses it to cancel redundant transmissions and reduce communication congestion. We rigorously compare its performance, in terms of mean-squared error (MSE) and cost of power per sensor, against more conventional Non-Feedback (NF) architectures by evaluating conditions of feasibility and advantage under different architecture specifications (e.g., communication delay rate, power cost rate, maximum back-off time, sampling period, observation noise). Here, we defined the advantage as the performance gain achieved by FB over NF, while FB is said to be feasible if the advantage region is nonempty. Our theoretical analyses show that the feasibility of FB depends more on the communication power cost, while the advantage depends on the sensors' propagation delay per transmission interval; we derive concrete conditions under which these outcomes hold. Using extensive numerical simulations under a variety of settings, we confirm the accuracy of the derived conditions, and show that our theoretical results hold even for more complex scenarios where the simplifying assumptions no longer hold.

eess.SY↗

arXiv2024

A Stochastic Robust Adaptive Systems Level Approach to Stabilizing Large-Scale Uncertain Markovian Jump Linear Systems

We propose a unified framework for robustly and adaptively stabilizing large-scale networked uncertain Markovian jump linear systems (MJLS) under external disturbances and mode switches that can change the network's topology. Adaptation is achieved by using minimal information on the disturbance to identify modes that are consistent with observable data. Robust control is achieved by extending the system level synthesis (SLS) approach, which allows us to pose the problem of simultaneously stabilizing multiple plants as a two-step convex optimization procedure. Our control pipeline computes a likelihood distribution of the system's current mode, uses them as probabilistic weights during simultaneous stabilization, then updates the likelihood via Bayesian inference. Because of this "softer" probabilistic approach to robust stabilization, our control pipeline does not suffer from abrupt destabilization issues due to changes in the system's true mode, which were observed in a previous method. Separability of SLS also lets us compute localized robust controllers for each subsystem, allowing for network scalability; we use several information consensus methods so that mode estimation can also be done locally. We apply our algorithms to disturbance-rejection on two sample dynamic power grid networks, a small-scale system with 7 nodes and a large-scale grid of 25 nodes.

eess.SY↗

arXiv2024

Efficient Replay Memory Architectures in Multi-Agent Reinforcement Learning for Traffic Congestion Control

Episodic control, inspired by the role of episodic memory in the human brain, has been shown to improve the sample inefficiency of model-free reinforcement learning by reusing high-return past experiences. However, the memory growth of episodic control is undesirable in large-scale multi-agent problems such as vehicle traffic management. This paper proposes a novel replay memory architecture called Dual-Memory Integrated Learning, to augment to multi-agent reinforcement learning methods for congestion control via adaptive light signal scheduling. Our dual-memory architecture mimics two core capabilities of human decision-making. First, it relies on diverse types of memory--semantic and episodic, short-term and long-term--in order to remember high-return states that occur often in the network and filter out states that don't. Second, it employs equivalence classes to group together similar state-action pairs and that can be controlled using the same action (i.e., light signal sequence). Theoretical analyses establish memory growth bounds, and simulation experiments on several intersection networks showcase improved congestion performance (e.g., vehicle throughput) from our method.

eess.SY↗

arXiv2023

Predictive Control of Linear Discrete-Time Markovian Jump Systems by Learning Recurrent Patterns

Incorporating pattern-learning for prediction (PLP) in many discrete-time or discrete-event systems allows for computation-efficient controller design by memorizing patterns to schedule control policies based on their future occurrences. In this paper, we demonstrate the effect of PLP by designing a controller architecture for a class of linear Markovian jump systems (MJS) where the aforementioned ``patterns'' correspond to finite-length sequences of modes. In our analysis of recurrent patterns, we use martingale theory to derive closed-form solutions to quantities pertaining to the occurrence of patterns: 1) the expected minimum occurrence time of any pattern from some predefined collection, 2) the probability of a pattern being the first to occur among the collection. Our method is applicable to real-world dynamics because we make two extensions to common assumptions in prior pattern-occurrence literature. First, the distribution of the mode process is unknown, and second, the true realization of the mode process is not observable. As demonstration, we consider fault-tolerant control of a dynamic topology-switching network, and empirically compare PLP to two controllers without PLP: a baseline based on the novel System Level Synthesis (SLS) approach and a topology-robust extension of the SLS baseline. We show that PLP is able to reject disturbances as effectively as the topology-robust controller at reduced computation time and control effort. We discuss several important tradeoffs, such as the size of the pattern collection and the system scale versus the accuracy of the mode predictions, which show how different PLP implementations affect stabilization and runtime performance.

eess.SY↗

arXiv2022

OUTformation: Distributed Data-Gathering with Feedback under Unknown Environment and Communication Delay Constraints

Towards the informed design of large-scale distributed data-gathering architectures under real-world assumptions such as nonzero communication delays and unknown environment dynamics, this paper considers the effects of allowing feedback communication from the central processor to external sensors. Using simple but representative state-estimation examples, we investigate fundamental tradeoffs between the mean-squared error (MSE) of the central processor's estimate of the environment state, and the total power expenditure per sensor under more conventional architectures without feedback (INformation) versus those with broadcast feedback (OUTformation). The primary advantage of enabling feedback is that each sensor's understanding of the central processor's estimate improves, which enables each sensor to determine when and what parts of its current observations to transmit. We use theory to demonstrate conditions in which OUTformation maintains the same MSE as INformation with less power expended on average, and conditions in which OUTformation obtains less MSE than INformation at additional power cost. These performance tradeoffs are also considered under settings where environments undergo less variation, and sensors implement random backoff times to prevent transmission collisions. Our results are supported via numerical studies, which show that the properties derived in theory still hold even when some of the simplifying assumptions are removed.

eess.SY↗

arXiv2021

A Two-Part Controller Synthesis Approach for Nonlinear Stochastic Systems Perturbed by L\'evy Noise Using Renewal Theory and HJB-Based Impulse Control

We are motivated by the lack of discussion surrounding methodological control design procedures for nonlinear shot and L\'evy noise stochastic systems to propose a hierarchical controller synthesis method with two parts. The first part is a primitive pattern-learning component which recognizes specific state sequences and stores in memory the corresponding control action that needs to be taken when the sequence has occurred. The second part is a modulation control component which computes the optimal control action for a pattern when it has occurred for the first time. Throughout our presentation of both components, we provide a self-contained discussion of theoretical concepts from Poisson processes theory, renewal theory, and impulse control, all of which are necessary as background. We demonstrate application of this controller to the simplified, concrete case studies of fault-tolerance and vehicle congestion control.

eess.SY↗

arXiv2021

Trading Throughput for Freshness: Freshness-Aware Traffic Engineering and In-Network Freshness Control

In addition to traditional concerns such as throughput and latency, freshness is becoming increasingly important. To stay fresh, applications stream status updates among their components. Existing studies propose the metric age of information (AoI) to gauge the freshness and design systems to achieve low AoI. Despite active research in this area, existing results are not applicable to general wired networks for two reasons. First, they focus on wireless settings where AoI is mostly affected by interference and collision while queueing is more dominant in wired settings. Second, the legacy drop-adverse flows are not taken into account in the literature. Scheduling mixed flows with distinct performance objective is not yet addressed. In this paper, we study wired networks shared by two classes of flows, aiming for high throughput and low AoI respectively, and achieve a good trade-off between their throughput and AoI. Our approach to the problem consists of two layers: freshness-aware traffic engineering (FATE) and in-network freshness control (IFC). FATE derives sending rate/update frequency for flows via optimization, and its solution is then enforced by IFC through efficient scheduling mechanisms at each outport of in-network nodes. We also present efficient Linux implementation of IFC and demonstrate the effectiveness of FATE/IFC through extensive emulations. Our results show that it is possible to trade a little throughput (5 % lower) for much shorter AoI (49 to 71% shorter) compared to state-of-the-art traffic engineering.

cs.NI↗

arXiv2021

Incremental Nonlinear Stability Analysis of Stochastic Systems Perturbed by L\'{e}vy Noise

We present a theoretical framework for characterizing incremental stability of nonlinear stochastic systems perturbed by compound Poisson shot noise and finite-measure L\'{e}vy noise. For each noise type, we compare trajectories of the perturbed system with distinct noise sample paths against trajectories of the nominal, unperturbed system. We show that for a finite number of jumps arising from the noise process, the mean-squared error between the trajectories exponentially converge towards a bounded error ball across a finite interval of time under practical boundedness assumptions. The convergence rate for shot noise systems is the same as the exponentially-stable nominal system, but with a tradeoff between the parameters of the shot noise process and the size of the error ball. The convergence rate and the error ball for the L\'{e}vy noise system are shown to be nearly direct sums of the respective quantities for the shot and white noise systems separately, a result which is analogous to the L\'{e}vy-Khintchine theorem. We demonstrate our results using several numerical case studies.

eess.SY↗

arXiv2021

FaSTrack: a Modular Framework for Real-Time Motion Planning and Guaranteed Safe Tracking

Real-time, guaranteed safe trajectory planning is vital for navigation in unknown environments. However, real-time navigation algorithms typically sacrifice robustness for computation speed. Alternatively, provably safe trajectory planning tends to be too computationally intensive for real-time replanning. We propose FaSTrack, Fast and Safe Tracking, a framework that achieves both real-time replanning and guaranteed safety. In this framework, real-time computation is achieved by allowing any trajectory planner to use a simplified \textit{planning model} of the system. The plan is tracked by the system, represented by a more realistic, higher-dimensional \textit{tracking model}. We precompute the tracking error bound (TEB) due to mismatch between the two models and due to external disturbances. We also obtain the corresponding tracking controller used to stay within the TEB. The precomputation does not require prior knowledge of the environment. We demonstrate FaSTrack using Hamilton-Jacobi reachability for precomputation and three different real-time trajectory planners with three different tracking-planning model pairs.

cs.RO↗

arXiv2020

System Level Synthesis via Dynamic Programming

System Level Synthesis (SLS) parametrization facilitates controller synthesis for large, complex, and distributed systems by incorporating system level constraints (SLCs) into a convex SLS problem and mapping its solution to stable controller design. Solving the SLS problem at scale efficiently is challenging, and current attempts take advantage of special system or controller structures to speed up the computation in parallel. However, those methods do not generalize as they rely on the specific system/controller properties. We argue that it is possible to solve general SLS problems more efficiently by exploiting the structure of SLS constraints. In particular, we derive dynamic programming (DP) algorithms to solve SLS problems. In addition to the plain SLS without any SLCs, we extend DP to tackle infinite horizon SLS approximation and entrywise linear constraints, which form a superclass of the locality constraints. Comparing to convex program solver and naive analytical derivation, DP solves SLS 4 to 12 times faster and scales with little computation overhead. We also quantize the cost of synthesizing a controller that stabilizes the system in a finite horizon through simulations.

eess.SY↗

arXiv2017

FaSTrack: a Modular Framework for Fast and Guaranteed Safe Motion Planning

Fast and safe navigation of dynamical systems through a priori unknown cluttered environments is vital to many applications of autonomous systems. However, trajectory planning for autonomous systems is computationally intensive, often requiring simplified dynamics that sacrifice safety and dynamic feasibility in order to plan efficiently. Conversely, safe trajectories can be computed using more sophisticated dynamic models, but this is typically too slow to be used for real-time planning. We propose a new algorithm FaSTrack: Fast and Safe Tracking for High Dimensional systems. A path or trajectory planner using simplified dynamics to plan quickly can be incorporated into the FaSTrack framework, which provides a safety controller for the vehicle along with a guaranteed tracking error bound. This bound captures all possible deviations due to high dimensional dynamics and external disturbances. Note that FaSTrack is modular and can be used with most current path or trajectory planners. We demonstrate this framework using a 10D nonlinear quadrotor model tracking a 3D path obtained from an RRT planner.

cs.RO↗