arXiv subjects

Zhaoxuan Zhu

Publications and source records attributed to Zhaoxuan Zhu.

12 recordsLinked to original sources

arXiv2026

Eluna: An Agentic LLM System for Automating Warehouse Operations with Reasoning and Task Execution

Warehouse operations are governed by Standard Operating Procedures (SOPs) that encode complex, multi-system decision logic, which must be executed reliably under strict time constraints, yet LLM agents lack mechanisms to enforce procedural compliance and degrade under the context overload full SOP specifications introduce. We present Eluna, a production-deployed agentic system for reliable SOP execution. Eluna is a graph-guided, multi-agent framework that encodes SOPs as directed acyclic graphs with progressive disclosure and delegates independent tasks to parallel sub-agents, each with persistent code execution and live data access. To meet production latency and accuracy needs, we use asymmetric episodic distillation where a strong teacher is improved through episodic error memories, then a smaller student is fine-tuned on the corrected trajectories with memory stripped, internalizing corrections without inference-time overhead. On a 13-task benchmark and two production applications, our fine-tuned models match or exceed their teacher, beat all larger off-the-shelf baselines, and reach 94% expert agreement on the ticket processing application.

cs.LG↗

arXiv2026

Thermodynamics and Tomonaga-Luttinger liquid behavior of the quantum 1D hard rod model

The one-dimensional hard rod model describes impenetrable bosons with finite diameter, extending the Lieb-Liniger model to systems with excluded volume interactions. Here, we investigate the thermodynamics of quantum HRs using Yang-Yang theory, path integral quantum Monte-Carlo calculations, and Luttinger liquid theory. We first discuss the behavior of characteristic thermodynamic quantities, exhibiting deviations to the Lieb-Liniger model for sufficiently high densities, with excellent agreement between analytical and numerical results. We then show that the hard rod model exhibits Tomonaga-Luttinger liquid behavior across a wide range of parameters, at zero and finite temperature, as unveiled by correlation functions. The Tomonaga-Luttinger parameter and thermal length can be extracted by fitting correlation functions to Tomonaga-Luttinger liquid theory, hence demonstrating a robust method for thermometry. This work provides a comprehensive study of strongly correlated hard rod systems at finite temperatures, with applications to quantum wires, spin chains, and ultracold atoms.

cond-mat.str-el↗

arXiv2024

Effective tight-binding models in optical moiré potentials

A twist between two systems offers the possibility to drastically change the underlying physical properties. To that end, we study the bandstructure of twisted moiré potentials in detail. At sets of commensurate twisting angles, the low energy single-particle spectrum of a twisted moiré potential will form into distinct bands and gaps. To a first approximation, energy bands can be qualitatively modelled by harmonic states, localised in different potential minima. The bands are intrinsically linked to the number of distinct minima and size of the moiré unit cell, with smaller cells producing larger gaps and vice versa. For shallower potential depths, degeneracies between harmonic states are lifted by virtue of anharmonic confinement and coupling between states. Depending on the exact geometry of potential minima, bands can then be classified by $4$ unique forms of tight-binding models. We find excellent agreement between the continuous spectrum and fitting to our tight-binding models, allowing for accurate tunnelling rates and onsite energies to be extracted. Our results are directly relevant to the bosonic, many-body problem, and thus provide further understanding on the relative stability of quantum phases both in theory and experiments. In particular, the prominence of gaps can be mapped to strongly correlated insulating phases. Furthermore, tunnelling rates of different bands serve as thresholds on temperature in which a phase can be either a normal fluid or superfluid.

cond-mat.quant-gas↗

arXiv2024

Angle of Arrival Estimation with Transformer: A Sparse and Gridless Method with Zero-Shot Capability

Automotive Multiple-Input Multiple-Output (MIMO) radars have gained significant traction in Advanced Driver Assistance Systems (ADAS) and Autonomous Vehicles (AV) due to their cost-effectiveness, resilience to challenging operating conditions, and extended detection range. To fully leverage the advantages of MIMO radars, it is crucial to develop an Angle of Arrival (AOA) algorithm that delivers high performance with reasonable computational workload. This work introduces AAETR (Angle of Arrival Estimation with TRansformer) for high performance gridless AOA estimation. Comprehensive evaluations across various signal-to-noise ratios (SNRs) and multi-target scenarios demonstrate AAETR's superior performance compared to super resolution AOA algorithms such as Iterative Adaptive Approach (IAA). The proposed architecture features efficient, scalable, sparse and gridless angle-finding capability, overcoming the issues of high computational cost and straddling loss in SNR associated with grid-based IAA. AAETR requires fewer tunable hyper-parameters and is end-to-end trainable in a deep learning radar perception pipeline. When trained on large-scale simulated datasets then evaluated on real dataset, AAETR exhibits remarkable zero-shot sim-to-real transferability and emergent sidelobe suppression capability. This highlights the effectiveness of the proposed approach and its potential as a drop-in module in practical systems.

cs.CV↗

arXiv2024

Weak Superfluidity in Twisted Optical Potentials

A controlled twist between different underlying lattices allows one to interpolate, under a unified framework, across ordered and (quasi-)disordered matter while drastically changing quantum transport properties. Here, we use quantum Monte Carlo simulations to determine the unique phase diagrams of strongly-correlated ultracold bosons in twisted optical potentials. We show that at commensurate twisting angles, spectral gaps govern the formation of insulating patterns, separated by thin superfluid domains. The latter form weak superfluids, which are very sensitive to thermal fluctuations, but can be stabilized under appropriate parameter control. In contrast, slightly changing the twisting angle to a incommensurate value destroys most spectral gaps, leaving behind a prominent Bose glass phase. Our results are directly applicable to current generation experiments that quantum simulate moiré physics.

cond-mat.quant-gas↗

arXiv2024

Localization and spectral structure in two-dimensional quasicrystal potentials

Quasicrystals, a fascinating class of materials with long-range but nonperiodic order, have revolutionized our understanding of solid-state physics due to their unique properties at the crossroads of long-range-ordered and disordered systems. Since their discovery, they continue to spark broad interest for their structural and electronic properties. The quantum simulation of quasicrystals in synthetic quantum matter systems offers a unique playground to investigate these systems with unprecedented control parameters. Here, we investigate the localization properties and spectral structure of quantum particles in 2D quasicrystalline optical potentials. While states are generally localized at low energy and extended at high energy, we find alternating localized and critical states at intermediate energies. Moreover, we identify a complex succession of gaps in the energy spectrum. We show that the most prominent gap arises from strongly localized ring states, with the gap width determined by the energy splitting between states with different quantized winding numbers. In addition, we find that these gaps are stable for quasicrystals with different rotational symmetries and potential depths, provided that other localized states do not enter the gap generated by the ring states. Our findings shed light on the unique properties of quantum quasicrystals and have implications for their many-body counterparts.

cond-mat.quant-gas↗

arXiv2023

Eco-Driving Control of Connected and Automated Vehicles using Neural Network based Rollout

Connected and autonomous vehicles have the potential to minimize energy consumption by optimizing the vehicle velocity and powertrain dynamics with Vehicle-to-Everything info en route. Existing deterministic and stochastic methods created to solve the eco-driving problem generally suffer from high computational and memory requirements, which makes online implementation challenging. This work proposes a hierarchical multi-horizon optimization framework implemented via a neural network. The neural network learns a full-route value function to account for the variability in route information and is then used to approximate the terminal cost in a receding horizon optimization. Simulations over real-world routes demonstrate that the proposed approach achieves comparable performance to a stochastic optimization solution obtained via reinforcement learning, while requiring no sophisticated training paradigm and negligible on-board memory.

cs.LG↗

arXiv2023

A Deep Reinforcement Learning Framework for Eco-driving in Connected and Automated Hybrid Electric Vehicles

Connected and Automated Vehicles (CAVs), in particular those with multiple power sources, have the potential to significantly reduce fuel consumption and travel time in real-world driving conditions. In particular, the Eco-driving problem seeks to design optimal speed and power usage profiles based upon look-ahead information from connectivity and advanced mapping features, to minimize the fuel consumption over a given itinerary. In this work, the Eco-driving problem is formulated as a Partially Observable Markov Decision Process (POMDP), which is then solved with a state-of-art Deep Reinforcement Learning (DRL) Actor Critic algorithm, Proximal Policy Optimization. An Eco-driving simulation environment is developed for training and evaluation purposes. To benchmark the performance of the DRL controller, a baseline controller representing the human driver, a trajectory optimization algorithm and the wait-and-see deterministic optimal solution are presented. With a minimal onboard computational requirement and a comparable travel time, the DRL controller reduces the fuel consumption by more than 17% compared against the baseline controller by modulating the vehicle velocity over the route and performing energy-efficient approach and departure at signalized intersections, over-performing the more computationally demanding trajectory optimization method

eess.SY↗

arXiv2023

Thermodynamic Phase Diagram of Two-Dimensional Bosons in a Quasicrystal Potential

Quantum simulation of quasicrystals in synthetic bosonic matter now paves the way to the exploration of these intriguing systems in wide parameter ranges. Yet thermal fluctuations in such systems compete with quantum coherence, and significantly affect the zero-temperature quantum phases. Here we determine the thermodynamic phase diagram of interacting bosons in a two-dimensional, homogeneous quasicrystal potential. Our results are found using quantum Monte Carlo simulations. Finite-size scaling is carefully considered and the quantum phases are systematically distinguished from thermal phases. In particular, we demonstrate stabilization of a genuine Bose glass phase against the normal fluid in sizable parameter ranges. Our results for strong interactions are interpreted using a fermionization picture and experimental relevance is discussed.

cond-mat.quant-gas↗

arXiv2022

Safe Model-based Off-policy Reinforcement Learning for Eco-Driving in Connected and Automated Hybrid Electric Vehicles

Connected and Automated Hybrid Electric Vehicles have the potential to reduce fuel consumption and travel time in real-world driving conditions. The eco-driving problem seeks to design optimal speed and power usage profiles based upon look-ahead information from connectivity and advanced mapping features. Recently, Deep Reinforcement Learning (DRL) has been applied to the eco-driving problem. While the previous studies synthesize simulators and model-free DRL to reduce online computation, this work proposes a Safe Off-policy Model-Based Reinforcement Learning algorithm for the eco-driving problem. The advantages over the existing literature are three-fold. First, the combination of off-policy learning and the use of a physics-based model improves the sample efficiency. Second, the training does not require any extrinsic rewarding mechanism for constraint satisfaction. Third, the feasibility of trajectory is guaranteed by using a safe set approximated by deep generative models. The performance of the proposed method is benchmarked against a baseline controller representing human drivers, a previously designed model-free DRL strategy, and the wait-and-see optimal solution. In simulation, the proposed algorithm leads to a policy with a higher average speed and a better fuel economy compared to the model-free agent. Compared to the baseline controller, the learned strategy reduces the fuel consumption by more than 21\% while keeping the average speed comparable.

cs.LG↗

arXiv2021

A GPU Implementation of a Look-Ahead Optimal Controller for Eco-Driving Based on Dynamic Programming

Predictive energy management of Connected and Automated Vehicles (CAVs), in particular those with multiple power sources, has the potential to significantly improve energy savings in real-world driving conditions. In particular, the eco-driving problem seeks to design optimal speed and power usage profiles based upon available information from connectivity and advanced mapping features to minimize the fuel consumption between two designated locations. In this work, the eco-driving problem is formulated as a three-state receding horizon optimal control problem and solved via Dynamic Programming (DP). The optimal solution, in terms of vehicle speed and battery State of Charge (SoC) trajectories, allows a connected and automated hybrid electric vehicle to intelligently pass the signalized intersections and minimize fuel consumption over a prescribed route. To enable real-time implementation, a parallel architecture of DP is proposed for an NVIDIA GPU with CUDA programming. Simulation results indicate that the proposed optimal controller delivers more than 15% fuel economy benefits compared to a baseline control strategy and that the solver time can be reduced by more than 90% by the parallel implementation when compared to a serial implementation.

eess.SY↗

arXiv2017

The Spatial Shape of Avalanches

In disordered elastic systems, driven by displacing a parabolic confining potential adiabatically slowly, all advance of the system is in bursts, termed avalanches. Avalanches have a finite extension in time, which is much smaller than the waiting-time between them. Avalanches also have a finite extension $\ell$ in space, i.e. only a part of the interface of size $\ell$ moves during an avalanche. Here we study their spatial shape $\left< S(x)\right>_{\ell}$ given $\ell$, as well as its fluctuations encoded in the second cumulant $\left< S^{2}(x)\right>_{\ell}^{\rm c}$. We establish scaling relations governing the behavior close to the boundary. We then give analytic results for the Brownian force model, in which the microscopic disorder for each degree of freedom is a random walk. Finally, we confirm these results with numerical simulations. To do this properly we elucidate the influence of discretization effects, which also confirms the assumptions entering into the scaling ansatz. This allows us to reach the scaling limit already for avalanches of moderate size. We find excellent agreement for the universal shape, its fluctuations, including all amplitudes.

cond-mat.dis-nn↗