arXiv subjects

Jian Su

Publications and source records attributed to Jian Su.

At least 19 recordsLinked to original sources

arXiv2026

Learning to Live: A Step Toward Lifelong Embodied Intelligence

Embodied agents are usually organized around individual tasks, yet the physical world keeps changing between tasks and during execution. We introduce LIFE (Lifelong Intelligence from Experience), a framework in which a robot observes continuously, remembers selectively, and turns experience into new capabilities. RoboBeats, a family of lightweight streaming gates, tracks conditions set by the agent and discovers new events, deciding when an observation warrants a response or a memory write. Selected observations enter Plastic Memory, a flat store of images and structured records that keeps the source evidence, so new fields and indexes can be added when later questions need them. When existing tools fall short, Explore-Distill-Mount (EDM) turns solutions found during problem solving into validated, reusable tools and skills. Under matched protocols, RoboBeats substantially improves condition tracking over its base models and, at the same writing budget, retains more useful evidence for memory question answering than uniform sampling. A case study of limited scale, replaying 24 hours of continuous video recorded by a real robot, illustrates how the integrated system builds memory from a day of experience and uses it to answer everyday questions. LIFE takes a step toward robots that become more familiar with their surroundings and more capable through experience: learning not only to complete tasks, but to live in the world.

cs.RO↗

arXiv2026

ReSight: Task-Driven Self-Improving Memory for Embodied Agents

Robots can use historical observations to answer spatiotemporal queries in familiar environments, but the presence of relevant evidence does not imply that existing tools support the operations a task requires. We introduce ReSight, a task-driven self-improving memory system for embodied agents that expands the capabilities for processing and accessing historical observations as task demands arise. ReSight retains keyframes and their spatiotemporal associations, separating observation evidence from extensible derived fields so that stored observations can be reprocessed for new needs. An execution agent and a coding agent share this memory through different interfaces: the former calls registered tools within a bounded online budget, while the latter uses a programmable workspace and a larger exploration budget to inspect evidence, write programs, and acquire models as needed. To realize capability expansion in ReSight, we develop Explore-Distill-Mount (EDM), which identifies capability gaps from execution and exploration records without task reference answers and converts exploration-derived solutions into reusable memory builders, query tools, and skills, extending memory representations and access capabilities as needed. Experiments on three public benchmarks demonstrate effective base memory access and capability expansion. With the execution model and online budget held fixed within each benchmark, EDM reduces mean position error on OC-NaVQA from 33.10 m to 24.72 m and improves high-level success on FindingDory from 38.3% to 63.8%. With the tools and skill frozen, success still improves by 28.3 percentage points on FindingDory episodes excluded from exploration and distillation. Experiments on real robot patrol memories further demonstrate reuse of these capabilities for subsequent queries and new memories, with fewer online tool calls on average.

cs.RO↗

arXiv2026

AdaRSI: On-Demand Recursive Self-Improvement for Embodied Agents

Pretrained vision-language-action models (VLAs) often fail midway through unseen long-horizon tasks. Recursive self-improvement (RSI) offers a remedy, but embodied RSI methods either refine the harness around a frozen executor and so cannot add a missing motion, or train the executor on persistent failures without asking whether a failure lies in how the executor is called or in what it can do. We ask when a self-improving embodied system should change the weights of its executor, and propose AdaRSI, an on-demand RSI framework for embodied agents. A coding agent first improves its skills and harness, then attributes persistent failures by retrying failing sub-steps from saved states, and fine-tunes only the sub-steps that remain unreliable, with demonstrations it collects itself and experience carried over from earlier tasks. On all 50 target tasks of RoboCasa365, with GR00T N1.5 as the executor and no human demonstrations, AdaRSI improves the VLA on all three suites and raises its success on unseen composite tasks from 4.6% to 13.8%. Compared with fine-tuning every sub-step, it reaches slightly higher success while using only 28% of the GPU time and 22% of the self-collected demonstrations, and fine-tuning without a harness trains more yet gains far less.

cs.CL↗

arXiv2026

Harnessing VLM in Embodied RL: From Reward Regressor to Success Detector via Exploration-Augmented Key-Observation Gating

VLMs provide promising solutions to embodied reward design. Most current methods that train VLMs as reward regressors not only require extensive annotations but are also primarily designed for finetuning pretrained policies, and perform less well in the more challenging from-scratch RL setting. This issue stems from a fundamental mismatch: regressing precise rewards is misaligned with what VLMs naturally excel at. We posit that it is more aligned with VLMs' inherent multi-modal VQA capabilities for the models to function as success detectors, and propose VLM success reward via Exploration-Augmented Key-Observation Gating (VLMeg) to well harness VLM for embodied reward design. As the core of VLMeg, the proposed exploration-augmented key-observation gating first determines whether the current RL episode is judgment-worthy based on its similarity to a generated success trajectory. Then, the key observation with the lowest confusion from the judgment-worthy episode will be selected for VLM evaluation as episodic reward. This gating mechanism not only avoids redundant calls but also enables higher detection accuracy of VLM. Considering the sparsity of judgment-worthy episodes, the advanced exploration incentives based on the generated trajectory are introduced, enabling more efficient exploration toward judgment-worthy trajectories for more VLM reward triggers and thus gaining higher RL efficiency. Extensive experiments in MetaWorld demonstrate VLMeg achieves state-of-the-art from-scratch RL, outperforming advanced VLM-based baselines on 12 challenging reward-free tasks. Furthermore, VLMeg also enables effective RL without any ground-truth reward in 3 challenging real-world experiments. Code will be open after acceptance at https://github.com/Raina915/VLMeg.

cs.LG↗

arXiv2026

HarnessVLN: Unifying Training-Free Embodied Navigation through an Agent Harness

Embodied navigation requires agents to ground instructions or object goals in spatial observations and translate plans into successful execution. As multimodal large language models (MLLMs) become increasingly capable, they offer stronger support for navigation without task-specific training; however, improved semantic reasoning alone does not ensure that proposed actions remain consistent with spatial evidence, task progress, and execution outcomes. We introduce HarnessVLN, a zero-shot, training-free framework that unifies instruction-following and object-goal navigation through a shared Agent Harness. The Harness coordinates perception, memory, and execution tools through a unified interface, validating planner proposals for evidential support, geometric feasibility, and subgoal consistency before dispatch. It jointly manages hierarchical event memory and a persistent Spatiotemporal Graph to track task progress, preserve spatial evidence, and contextualize failures. Structured execution feedback updates this shared state, guiding subsequent planning, recovery, and termination. Across R2R, RxR, HM3D-v2, and HM3D-OVON, HarnessVLN achieves success rates of 59.6%, 51.4%, 76.0%, and 59.3%, respectively, outperforming prior training-free methods. Humanoid robot deployment further demonstrates its applicability to both navigation tasks in real-world environments. The project page is available at [https://agibot-harnessvln.netlify.app/].

cs.RO↗

arXiv2026

CoCoBench: A Cooperative Coordination Benchmark for Embodied Multi-Agent Task Planning

Agent systems powered by multimodal large language models (MLLMs) have advanced rapidly in recent years, yet existing embodied-agent benchmarks still lack fine-grained diagnostics for multi-agent coordination. Most benchmarks either focus on single-agent task completion or summarize multi-agent behavior with overall task success rates, which can obscure coordination failures such as duplicated work, violations of ordering constraints, resource contention, and desynchronized handoffs. In this paper, we introduce CoCoBench, a construct-level benchmark for evaluating multi-agent embodied coordination in executable household tasks. CoCoBench contains 897 oracle-validated instances organized around four recurring coordination constructs: task allocation, sequential ordering, mutual exclusion, and handoff coordination. In addition to task success rate, CoCoBench provides construct-level scores that measure whether agents coordinate effectively. We evaluate 11 leading MLLMs across different coordination modes, observation inputs, and numbers of agents. The results show that coordination ability is highly construct-specific: strong overall performance does not imply balanced competence across different coordination types. These findings point to new directions for designing targeted model architectures and improving multi-agent coordination ability.

cs.RO↗

arXiv2026

MANSION: Multi-floor lANguage-to-3D Scene generatIOn for loNg-horizon tasks

Real-world robotic tasks are long-horizon and often span multiple floors, demanding rich spatial reasoning. However, existing embodied benchmarks are largely confined to single-floor in-house environments, failing to reflect the complexity of real-world tasks. We introduce MANSION, the first language-driven framework for generating building-scale, multi-floor 3D environments. Being aware of vertical structural constraints, MANSION generates realistic, navigable whole-building structures with diverse, human-friendly scenes, enabling the development and evaluation of cross-floor long-horizon tasks. Building on this framework, we release MansionWorld, a dataset of over 1,000 diverse buildings ranging from hospitals to offices, alongside a Task-Semantic Scene Editing Agent that customizes these environments using open-vocabulary commands to meet specific user needs. Benchmarking reveals that state-of-the-art agents degrade sharply in our settings, establishing MANSION as a critical testbed for the next generation of spatial reasoning and planning.

cs.CV↗

arXiv2026

Can LLM Safety Be Ensured by Constraining Parameter Regions?

Large language models (LLMs) are often assumed to contain ``safety regions'' -- parameter subsets whose modification directly influences safety behaviors. We conduct a systematic evaluation of four safety region identification methods spanning different parameter granularities, from individual weights to entire Transformer layers, across four families of backbone LLMs with varying sizes. Using ten safety identification datasets, we find that the identified safety regions exhibit only low to moderate overlap, as measured by IoU. The overlap drops significantly when the safety regions are further refined using utility datasets (\ie non-harmful queries). These results suggest that current techniques fail to reliably identify a stable, dataset-agnostic safety region.

cs.LG↗

arXiv2025

Deep learning-based computational method for soft matter dynamics: Deep Onsager-Machlup method

A deep learning-based computational method is proposed for soft matter dynamics -- the deep Onsager-Machlup method (DOMM). It combines the brute forces of deep neural networks (DNNs) with the fundamental physics principle -- Onsager-Machlup variational principle (OMVP). In the DOMM, the trial solution to the dynamics is constructed by DNNs that allow us to explore a rich and complex set of admissible functions. It outperforms the Ritz-type variational method where one has to impose carefully-chosen trial functions. This capability endows the DOMM with the potential to solve rather complex problems in soft matter dynamics that involve multiple physics with multiple slow variables, multiple scales, and multiple dissipative processes. Actually, the DOMM can be regarded as an extension of the deep Ritz method (DRM) developed by E and Yu that uses DNNs to solve static problems in physics. In this work, as the first step, we focus on the validation of the DOMM as a useful computational method by using it to solve several typical soft matter dynamic problems: particle diffusion in dilute solutions, and two-phase dynamics with and without hydrodynamics. The predicted results agree very well with the analytical solution or numerical solution from traditional computational methods. These results show the accuracy and convergence of DOMM and justify it as an alternative computational method for solving soft matter dynamics.

cond-mat.soft↗

arXiv2023

Battle of the Large Language Models: Dolly vs LLaMA vs Vicuna vs Guanaco vs Bard vs ChatGPT -- A Text-to-SQL Parsing Comparison

The success of ChatGPT has ignited an AI race, with researchers striving to develop new large language models (LLMs) that can match or surpass the language understanding and generation abilities of commercial ones. In recent times, a number of models have emerged, claiming performance near that of GPT-3.5 or GPT-4 through various instruction-tuning methods. As practitioners of Text-to-SQL parsing, we are grateful for their valuable contributions to open-source research. However, it is important to approach these claims with a sense of scrutiny and ascertain the actual effectiveness of these models. Therefore, we pit six popular large language models against each other, systematically evaluating their Text-to-SQL parsing capability on nine benchmark datasets with five different prompting strategies, covering both zero-shot and few-shot scenarios. Regrettably, the open-sourced models fell significantly short of the performance achieved by closed-source models like GPT-3.5, highlighting the need for further work to bridge the performance gap between these models.

cs.CL↗

arXiv2023

Dynamics of flexible fibers in confined shear flows at finite Reynolds numbers

We carry out a numerical study on the dynamics of a single non-Brownian flexible fiber in two-dimensional Couette flows at finite Reynolds numbers. We employ the bead-spring model of flexible fibers to extend the fluid particle dynamics (FPD) method that is originally developed for rigid particles in viscous liquids. We implement the extended FPD method using a multiple-relaxation-time (MRT) scheme of the lattice Boltzmann method (LBM). The numerical scheme is validated firstly by a series of benchmark simulations that involve liquid-solid coupling. The method is then used to study the dynamics of flexible fibers in Couette flows. We only consider the highly symmetric case where the fibers are placed on the symmetry center of Couette flows and we focus on the effects of the fiber stiffness, the confinement strength, and the finite Reynolds number (from 1 to 10). A diagram of the fiber shape is obtained. For fibers under weak confinement and a small Reynolds number, three distinct tumbling orbits have been identified. (1) Jeffery orbits of rigid fibers. The fibers behave like rigid rods and tumble periodically without any visible deformation. (2) S-turn orbits of slightly flexible fibers. The fiber is bent to an S-shape and is straightened again when it orients to an angle of around 45 degrees relative to the positive x direction. (3) S-coiled orbits of fairly flexible fibers. The fiber is folded to an S-shape and tumbles periodically and steadily without being straightened anymore during its rotation. Moreover, the fiber tumbling is found to be hindered by increasing either the Reynolds number or the confinement strength, or both.

cond-mat.soft↗

arXiv2022

Spontaneous Bending of Hydra Tissue Fragments Driven by Supracellular Actomyosin Bundles

Hydra tissue fragments excised freshly from Hydra body bend spontaneously to some quasi-stable shape in several minutes. We propose that the spontaneous bending is driven mechanically by supracellular actomyosin bundles inherited from parent Hydra. An active-laminated-plate model is constructed, from which we predict that the fragment shape characterized by spontaneous curvature is determined by its anisotropy in contractility and elasticity. The inward bending to endoderm side is ensured by the presence of a soft intermediate matrix (mesoglea) layer. The bending process starts diffusively from the edges and relaxes exponentially to the final quasi-stable shape. Two characteristic time scales are identified from the dissipation due to viscous drag and interlayer frictional sliding, respectively. The former is about 0.01 seconds, but the latter is much larger, about several minutes, consistent with experiments.

physics.bio-ph↗

arXiv2022

Semantic Pivoting Model for Effective Event Detection

Event Detection, which aims to identify and classify mentions of event instances from unstructured articles, is an important task in Natural Language Processing (NLP). Existing techniques for event detection only use homogeneous one-hot vectors to represent the event type classes, ignoring the fact that the semantic meaning of the types is important to the task. Such an approach is inefficient and prone to overfitting. In this paper, we propose a Semantic Pivoting Model for Effective Event Detection (SPEED), which explicitly incorporates prior information during training and captures semantically meaningful correlations between input and events. Experimental results show that our proposed model achieves state-of-the-art performance and outperforms the baselines in multiple settings without using any external resources.

cs.CL↗

arXiv2022

Variational methods and deep Ritz method for active elastic solids

Variational methods have been widely used in soft matter physics for both static and dynamic problems. These methods are mostly based on two variational principles: the variational principle of minimum free energy (MFEVP) and Onsager's variational principle (OVP). Our interests lie in the applications of these variational methods to active matter physics. In our former work [Soft Matter, 2021, 17, 3634], we have explored the applications of OVP-based variational methods for the modeling of active matter dynamics. In the present work, we explore variational (or energy) methods that are based on MFEVP for static problems in active elastic solids. We show that MFEVP can be used not only to derive equilibrium equations, but also to develop approximate solution methods, such as Ritz method, for active solid statics. Moreover, the power of Ritz-type method can be further enhanced using deep learning methods if we use deep neural networks to construct the trial solutions of the variational problems. We then apply these variational methods and the deep Ritz method to study the spontaneous bending and contraction of a thin active circular plate that is induced by internal asymmetric active contraction. The circular plate is found to be bent towards its contracting side. The study of such a simple toy system gives implications for understanding the morphogenesis of solid-like confluent cell monolayers. In addition, we introduce a so-called activogravity length to characterize the importance of gravitational forces relative to internal active contraction in driving the bending of the active plate. When the lateral plate dimension is larger than the activogravity length (about 100 micron), gravitational forces become important. Such gravitaxis behaviors at multicellular scales may play significant roles in the morphogenesis and in the up-down symmetry broken during tissue development.

cond-mat.soft↗

arXiv2019

Densely Connected Attention Propagation for Reading Comprehension

We propose DecaProp (Densely Connected Attention Propagation), a new densely connected neural architecture for reading comprehension (RC). There are two distinct characteristics of our model. Firstly, our model densely connects all pairwise layers of the network, modeling relationships between passage and query across all hierarchical levels. Secondly, the dense connectors in our network are learned via attention instead of standard residual skip-connectors. To this end, we propose novel Bidirectional Attention Connectors (BAC) for efficiently forging connections throughout the network. We conduct extensive experiments on four challenging RC benchmarks. Our proposed approach achieves state-of-the-art results on all four, outperforming existing baselines by up to $2.6\%-14.2\%$ in absolute F1 score.

cs.CL↗

arXiv2018

A time-efficient frame size adjustment based approach for RFID anti-collision

Fast and efficient identify a large number of RFID tags in the region of interest is a critical issue in various RFID applications. In this paper, a novel sub-frame-based algorithm with a time-efficient frame size adjustment strategy to reduce the time complexity for EPCglobal C1 Gen2 UHF RFID standard is proposed. By observing the slot statistics in a sub-frame, the tag quantity is estimated by the reader, which afterwards efficiently calculates an optimal frame size to fit the unread tags. Only when the expected time efficiency in the oncoming frame is higher than that in the previous frame, the reader starts the new identification round with the updated frame. Moreover, the estimation of the proposed algorithm is implemented by the look-up tables, which allows dramatically reduction in the computational complexity. Simulation results show noticeable throughput, time efficiency, and identification speed improvements of the proposed solution over the existing approaches.

cs.NI↗

arXiv2018

Reasoning with Sarcasm by Reading In-between

Sarcasm is a sophisticated speech act which commonly manifests on social communities such as Twitter and Reddit. The prevalence of sarcasm on the social web is highly disruptive to opinion mining systems due to not only its tendency of polarity flipping but also usage of figurative language. Sarcasm commonly manifests with a contrastive theme either between positive-negative sentiments or between literal-figurative scenarios. In this paper, we revisit the notion of modeling contrast in order to reason with sarcasm. More specifically, we propose an attention-based neural model that looks in-between instead of across, enabling it to explicitly model contrast and incongruity. We conduct extensive experiments on six benchmark datasets from Twitter, Reddit and the Internet Argument Corpus. Our proposed model not only achieves state-of-the-art performance on all datasets but also enjoys improved interpretability.

cs.CL↗

arXiv2017

Direct Imaging of Slow, Stored, and Stationary EIT Polaritons

Stationary and slow light effects are of great interest for quantum information applications. Using laser-cooled Rb87 atoms we have performed side imaging of our atomic ensemble under slow and stationary light conditions, which allows direct comparison with numerical models. The polaritions were generated using electromagnetically induced transparency (EIT), with stationary light generated using counter-propagating control fields. By controlling the power ratio of the two control fields we show fine control of the group velocity of the stationary light. We also compare the dynamics of stationary light using monochromatic and bichromatic control fields. Our results show negligible difference between the two situations, in contrast to previous work in EIT based systems.

quant-ph↗