Tensor method scales exact MDP optimization linearly for infrastructure
Kronecker-factored transitions cut complexity from exponential to linear while keeping global dynamic programming solutions intact.
cs.SY
Systems and Control
cs.SY is an alias for eess.SY. This section includes theoretical and experimental research covering all facets of automatic control systems. The section is focused on methods of control system analysis and design using tools of modeling, simulation and optimization. Specific areas of research include nonlinear, distributed, adaptive, stochastic and robust control in addition to hybrid and discrete event systems. Application areas include automotive and aerospace control systems, network control, biological systems, multiagent and cooperative control, robotics, reinforcement learning, sensor networks, control of cyber-physical and energy-related systems, and control of computing systems.
sort pith recommended most recent
Smaller models match large LLM teachers in car motion planning
On-policy distillation beats RL baseline and nears full accuracy on nuScenes despite 5x size reduction.
· “On-Policy Distillation of Language Models for Autonomous Vehicle Motion Planning”
Control duality reduces all EnKF variants to hyperparameter choices
A single framework classifies stochastic and deterministic ensemble Kalman filters and supplies a route to new designs.
· “A Unified Control Theory Derivation of Discrete-Time Linear Ensemble Kalman Filters”
Crossing counts estimate ARMA poles via autocorrelation
Only event tallies from the series need storage to recover poles for assessing control loop behavior.
· “A New Approach for ARMA Pole Estimation Using Higher-Order Crossings”
Simulations replicate 23-Hz oscillations from Texas electronic load
EMT models match fault recorder data when load reaches 320 MW, validating analysis of grid disturbances
· “Replicating Real-World 23-Hz Oscillations Caused by Large Electronic Loads”
Bayesian method locates radioactive sources with movement penalty
The strategy models radiation with a heteroscedastic Gaussian process and achieves sublinear regret in simulations while limiting robot tr
· “Radioactive Source Seeking using Bayesian Optimisation with Movement Penalty”
Active learning builds thermal twins with one-fifth the simulations
Surrogate models match full accuracy on heat exchanger flow and transfer rates using far fewer trajectories than random sampling
· “Physics-based Digital Twins for Integrated Thermal Energy Systems Using Active Learning”
Two-layer architecture links freight sensing to signal priority decisions
Review shows how detection errors and timing uncertainties shape reliable priority requests at urban intersections
· “From Sensing to Decision: A Generic Architecture for Freight Signal Priority Systems”
This paper reviews recent advancements in mm-wave oscillators below 100 GHz and…
A literature review summarizing design approaches, performance metrics, challenges, and emerging techniques for mm-wave and sub-THz/THz…
· “Recent Advances in mm-Wave and Sub-THz/THz Oscillators for FutureG Technologies”
Inverse design yields 3D-printed elastomers at exact transparency and hardness targets
Response surface models and desirability optimisation replace trial-and-error mixing so custom tactile sensor materials are obtained on the
Criterion gives general settings for GFM inverter distance protection
It maintains better dependability for nearby faults than quadrilateral relays while allowing general settings across inverter types andfault
Learns maximal state-control invariant sets from failures without dynamics
FAIL iteratively recovers predecessor half-spaces by regression on one-step failing pairs and converges monotonically for deterministic LTI
· “Failure-Aware Iterative Learning of State-Control Invariant Sets”
Agents trade safety margins to protect priority neighbors
Collaborative barrier functions let agents in coupled systems assist higher-priority peers while keeping the whole group safe through local,
· “Collaborative Altruistic Safety in Coupled Multi-Agent Systems”
TRASE-NODEs learn dynamics from less data by tracking sensitivity
Augmenting neural ODEs with state and sensitivity dynamics captures control effects and cuts prediction errors on oscillator and power-elect
Closed-form law found for time-optimal flexible moves
Maneuver time splits into rigid-body minimum plus a bounded synchronization cost, so synthesis becomes a single scalar inversion.
EV with rooftop PV and V2G saves up to EUR 2410 yearly
Aging-aware MPC optimizes bidirectional flows for arbitrage and self-consumption, beating unidirectional charging with only 1.27 percent额外电池
Attack-trained ranking finds power-grid sensors worth protecting
In IEEE 14-bus simulations, top-ranked nodes worsen frequency and net-power deviations about 50% more than random picks.
· “Adversarially-Informed Node Criticality Identification in Power Grid Measurements”
Three-phase AC is an engineering optimum
A balanced two-phase system already gives smooth power, so the real advantage is three-wire symmetry and standardization.
80.91 mAP: a deformable head lifts single-stage 3D detection
Plug it into SECOND, VoxelNet, or TANet and each gains about one AP point on KITTI cars
· “TADP: Task-Aware Deformable Prediction for Single-Stage 3D Object Detection”
AI agent designs cell-edge power control at 99.5% of best solver
Unattended, it rewrote architecture, loss and features to hit 600x cheaper inference.
· “Agentic Autoresearch for Cell-Edge Power Control: Radically Redefining the Researcher's Role”
Stealthy GPS spoofing redirects drones to a target within 0.5 m
Tests land drones on target within 0.5 m while two anomaly detectors stay quiet.
Unsupervised model splits drivers into three traffic regimes
A density-matrix representation of each driver also reproduces the fundamental diagram and hysteresis on I-24 highway data.
A bridge sampler draws Boltzmann samples on curved manifolds
It trains an intrinsic tangent control, so particles never leave spheres, Stiefel manifolds, or robot loop-closure spaces.
· “Hard-Constrained Sampling on Embedded Riemannian Manifolds via Adjoint Schr\"{o}dinger Bridges”
LLM-written radio policies are 94% accurate on unseen missions
Zero-shot prompting already recovers the mandatory quality thresholds; extra examples only fine-tune the weights.
One model dispatches seven grids at 1.23% cost gap
Jointly trained from IEEE 57 to a real 4,492-bus Korean grid, it stays feasible under outages and topology changes.
· “UNION: A Unified AC-OPF Framework for Topology-Varying Real-Time Grid Operation”
Data-driven RTO models pass validation yet pick phantom optima
Worst-case profit loss tops 30% for hybrid and 50% for black-box models despite perfect fits.
Wind shifts can push the Texas grid past high-voltage limits
A uniform-budget search ranks scenarios by hidden vulnerability; wind cuts in the Panhandle top the list.
· “Critical Weather Scenario Screening Using Weather-to-Voltage (W2V) Predictive Modeling”
Two-stage pipeline lifts line-fault accuracy from 31% to 96%
Detector-classifier split plus zero-sequence features beats a federated baseline without GPUs
Random collisions alone can trigger orbital-debris runaway
A Feller-diffusion limit shows a debris cascade can start at the threshold even when the deterministic drift is zero.
· “Stochastic Dynamics of Low Earth Orbit Near Full Capacity”
Fair sharing of grid exports costs an extra 3.6 MWh in curtailment
On a 33-bus feeder, fairness raises curtailment from 2.11 to 5.72 MWh while holding the max cumulative ratio to 11.2%.
Graph attention layer redesign cuts attack-detection latency by 8x
In HVAC graphs up to 1536 nodes, latency drops from 1.37 ms to 0.146 ms while F1 stays near 0.98.
· “Graph-Native Attention Acceleration for Attack Detection in Cyber-Physical Systems”
Noisy data alone can build invariant safety sets
One SDP chain yields feedback gains and invariant ellipsoids for state and output feedback, no model needed
· “Data-Driven Synthesis of Robust Positively Invariant Sets: From State Feedback to Output Feedback”
Neural-net fuel blending cuts cost 19 percent
Closed-loop simulation shows a model predictive controller keeping power, pressure, and methane number in spec.
· “Artificial intelligence-based predictive fuel blending control for flare gas mitigation”
Delay fixes a hard range for plug-in DER unit counts
Stability and no-reverse-power constraints set the connectable unit count; beyond a worst-case delay, none are admissible.
One derivation generates all four flight-control inversion laws
A unified equation yields model-based, sensory, incremental, and hybrid NDI laws and shows how to choose between them.
· “Dynamic Inversion: An Incrementally Evolving Methodology for Flight Control Design”
Derivative-free policy search reaches global H-infinity optimality
A convex-lifting proximal method needs only O~(d epsilon^-3) function evaluations and keeps iterates feasible.
Autonomy-aware 4D flight permits cut reserved airspace by 25%
A MILP sizes U-space volumes from 95% trajectory envelopes instead of fixed buffers, freeing airspace for more drones.
· “Leveraging UAV Autonomy for Minimum 4D Flight Authorization Volumes”
Blended heading rule tracks moving sources with lower error
Loop-based strategy combines EKF estimate with Fisher-information gain to beat estimate-only and info-only turns.
Zero safety violations on a 7-DOF arm via certified symbolic control
Certifying every command lets the robot hold, replan, or stop rather than act uncertified.
· “CSymPlan: Certified Symbolic Planning and Control for High-DOF Manipulators”
An undergraduate team built and documented an open
The NeuralNexus Arm is a $1,512 open-source 6-DOF stepper-driven manipulator with a mixed onboard and external driver architecture…
Generative model designs cooling channels that cut GPU heat 33.6%
Selected channel also cuts temperature spread 52.5% and pressure drop 72.8%, with CFD checks within about 9%.
This paper shows that a grid operator can detect when multiple AI data centers are…
Correlated AI data-center load episodes can be detected from grid measurements alone, with detection confidence growing as the square root…
· “Real-Time Edge-based Detection of Correlated AI Data-Center Load Episodes”
The paper characterizes feedback Nash equilibria for a new class of multi-player…
For linear-cost positive-dynamics games, feedback Nash equilibria are characterized by coupled vector ODEs or algebraic equations, giving…
This paper develops a stability analysis framework for multi-electrolyzer hydrogen plants
A three-port admittance model and plant-level aggregation show that balanced power allocation improves stability margins of…
This paper builds a framework for deciding when a coarse
A framework using structured stochastic liftings, delay windows, and state-space enrichment to detect and repair non-closure in probability…
· “Identifying Probability Localization Dynamics via Structured Stochastic Liftings”
Active learning grows grid-stability regions 20-fold
A Gaussian-process search adds certified area beyond Lyapunov limits with at most 220 EMT runs per system.
Drill-down cuts geospatial anomaly scans by 99.7%
Start coarse, zoom only where anomalies appear: 4,800 cells inspected instead of 2 million.
Dynamic semantic routes hit 97.4% accuracy for 5G intent extraction
Maps operator language to JSON-like schemas, moving 5G management toward zero-touch automation.
· “Advanced LLM-Enhanced Intent-Based 5G Network Management using Dynamic Semantic Routes”
Numerator scaling restores safety control with the safe set intact
The reciprocal barrier certifies the full obstacle-free interior for force-controlled nonholonomic robots, not a shrunken subset.
· “Scaling-Based Reciprocal Control Barrier Functions for Nonholonomic Mobile Robots”
Calibrated friction dynamics put simulated torque policies on a real robot
Trajectory matching plus domain randomization lets a TQC policy reach targets on a Franka Panda without real-world training.
Uncertainty-aware airline planning cuts realised costs 4.1%
A network that prices remaining-life risk beats deterministic and real airline schedules on 500 real-data test runs.
Finite excitation now guarantees exponential parameter convergence
Orthogonalized memory sets the decay rate by user gain, not excitation strength.
· “Robust Model Reference Adaptive Control with Combined Adaptation under Finite Excitation Condition”
Null-steering jammers deliver both security and wireless power in HAPS
New formulas predict joint rate-energy coverage and secrecy from ρ, D, and jammer power.
· “Secure Wireless Information Transfer and Energy Harvesting in HAPS-Based Network”
Glidepath angle, not late deceleration, saves approach fuel
Matched-angle tests on four airframes show deceleration architecture adds only 0.3-3.9% at the 3.5-degree final.
· “Optimal TRACON Descent Procedures under Wind Uncertainty and Fuel Savings Factors”
Open phase reads as a huge negative impedance
Two meters compute a connection impedance; an open phase flips it negative and much larger than line impedance.
· “A Novel Open Phase Detection Method with Synchronized Phasors for Distribution Systems”
A 51-step rendezvous control law fits on a CubeSat computer
Offline-verified lookup control fits 10.5 MB, evaluates below 300 ns, and provably settles to one fixed gain.
· “Certifiable Explicit Model Predictive Control for Spacecraft Rendezvous under Bounded Disturbances”
Ride-share pricing and matching solved 1,000x faster
A gradient-guided algorithm also lifts simulated revenue by 5.2% and service rate by 8.2% in two-city tests.
Predictive converter control: 88% less compute
Precomputed boundary maps pick the optimal switch state instantly, matching classical FCS-MPC on test hardware.
· “A Simple and Extremely Efficient Predictive Control for Power Converters”
Stein-equation ADI now accepts shifts anywhere in the complex plane
Generalized LRCF-ADI enables frequency-limited Gramians and data-driven balanced truncation from samples alone.
· “A New Generalized Low-Rank Cholesky Factor ADI Algorithm for Large-Scale Stein Equations”
Grid-aware PSO hits the discrete optimum for PWM harmonics
It solves directly on the controller's timer grid, cutting solve time while keeping deployed harmonics accurate.
Event-driven simulator runs power-electronics grid 11.6x faster
Advancing converter modules only at switching events keeps errors under 0.2 percent while slashing computation.
· “Event-Driven Simulation of Power Electronics Rich Grid Models”
One template unifies H-infinity and attack worst-case metrics
Worst-case attack impact becomes an exact SDP for one-constraint systems and an exact LP for positive systems.
· “Antagonistic Control: Foundations, Scalability and Nonlinearity”
Simulink arc model reproduces DC arc V-I curve from 10-1500 A
Coupled thermal-fluid-magnetic simulation has voltage stabilize within milliseconds across the range.
· “A 2D Axisymmetric Multi-Domain DC Arc Model for Simulink Implementation”
Chaskey-12 and ML-KEM fit 7 of 8 sampled-value profiles
Compiler-optimized Chaskey-12 plus ML-KEM key setup meets sampled-value timing on all but the 96-kHz profile.
· “A Lightweight and Post-Quantum Secure Framework for IEC 61869-9 Sampled Value Communication”
Multi-step training recovers sparse equations from noisy time series
Chaotic systems: short trajectories diverge, but long-run statistics match the true attractor.
· “Symbolic Neural ODEs: Learning interpretable models from time-series data”
Opinion-guided strategies coordinate with any rational agent
Two identical agents settle on one admissible joint behavior from their interaction alone.
· “Opinion-Guided Layered Strategies for Decentralized Coordination”
Two per-converter indices certify stability in mixed GFM–GFL grids
Stability certificates for large converter-dominated systems no longer need detailed GFM control models.
· “Impacts of Heterogeneous Grid-Forming Devices on Power System Dynamics Quantified by DW Shells”
Tiny AI add-on cuts motor transient error by 68%
An on-chip side-channel spots torque swings in under 200 microseconds and trims regulation error without changing the main loop.
One pretrained model steadies eight microgrids
Prompt-Decision Transformer uses few-shot expert trajectories as prompts, adapting to new grids with minimal tuning.
Lane recovery cuts travel time 13.7% in correlated outages
But under severe random detector loss the gain reverses, so the benefit is regime-dependent.
Warehouse digout planners self-tune
An extremum-seeking loop tunes digout weights online, removing manual sweeps and adapting to station outages.
· “Model-Free Adaptive Parameter Tuning for Efficient Multi-Robot Warehouse Operations”
Condition-guided adaptation lifts fault-diagnosis accuracy to 0.918
A frozen source model's residual response picks reliable unlabeled samples, lifting accuracy above the best baseline on gearbox and motor…
Resonant data-center loads cut safe margin nearly fivefold
Periodic demand makes loadability a frequency-dependent fold, so static screening misses the real limit.
Recursive Bayesian filter cuts lane-change prediction flips by a third
Temporal smoothing plus calibration keeps intent estimates accurate while making them steadier and better calibrated.
One RL agent keeps the grid thermally safe and damps oscillations
Trained on random load swings, it holds N-1 limits while lifting damping and critical clearing times.
· “Multi-Objective Deep Reinforcement Learning for Secure and Stable Power System Operation”
0.9991 monitor coverage is the key to fail-operational drone swarms
Catching 99.91 percent of flight-critical faults meets the one-in-ten-million per hour crash target.
· “A Safety-Driven Architectural Framework for Fail-Operational Drone Swarms in Critical Missions”
Adding price lags worsens ERCOT spike forecasts
Normal hours improve, but spike-hour MAE rises 5.8% and physical drivers stay hidden.
DRL-MPC split cuts freeway control computation 30x
Slow MPC picks route splits, a fast learned policy meters ramps, holding queue limits even when the model is wrong.
Bayesian warm-start cuts asteroid-hop iteration count by a third
One Bézier control point, tuned online with no training data, beats the straight-line guess on all 20 Eros site transfers.
Temporal-logic conditioning lifts robot spec success by up to 41 points
A single vision-language-action policy follows natural-language tasks and unseen formal safety requirements without retraining.
· “Logic-VLA: A Temporal Logic Conditioned Vision-Language-Action Model”
