arXiv subjects

Haishan Liu

Publications and source records attributed to Haishan Liu.

16 recordsLinked to original sources

arXiv2026

RoboSemanticBench: Diagnosing Semantic Grounding in Action Prediction for VLA Models

Vision-language-action (VLA) models are built on the premise that semantic understanding from pretrained language or vision-language backbones should guide robot action prediction. Yet robot fine-tuning is optimized as imitation over task-specific action distributions, and many evaluations can be solved through visual or instruction-action shortcuts. We introduce RoboSemanticBench (RSB), an embodied benchmark for diagnosing semantic grounding in action prediction: whether post-trained VLA models can use complex instruction semantics to select and manipulate the correct physical target. In each episode, a robot receives a multiple-choice math or general-knowledge question, observes candidate answer blocks, and must grasp the block corresponding to the correct answer. RSB covers controlled arithmetic, grade-school mathematical understanding, and commonsense or factual understanding under four-choice and ten-choice suites. Across representative VLA models, we find that many policies learn to grasp candidate blocks but select the semantically correct block at near-random or below-random rates after controlling for grasp success, revealing a persistent gap between backbone-level semantic competence and action prediction.

cs.RO↗

arXiv2026

IntentVLA: Short-Horizon Intent Modeling for Aliased Robot Manipulation

Robot imitation data are often multimodal: similar visual-language observations may be followed by different action chunks because human demonstrators act with different short-horizon intents, task phases, or recent context. Existing frame-conditioned VLA policies infer each chunk from the current observation and instruction alone, so under partial observability they may resample different intents across adjacent replanning steps, leading to inter-chunk conflict and unstable execution. We introduce IntentVLA, a history-conditioned VLA framework that encodes recent visual observations into a compact short-horizon intent representation and uses it to condition chunk generation. We further introduce AliasBench, a 12-task ambiguity-aware benchmark on RoboTwin2 with matched training data and evaluation environments that isolate short-horizon observation aliasing. Across AliasBench, SimplerEnv, LIBERO, and RoboCasa, IntentVLA improves rollout stability and outperforms strong VLA baselines

cs.RO↗

arXiv2026

PhysBrain 1.0 Technical Report

Vision-language-action models have advanced rapidly, but robot trajectories alone provide limited coverage for learning broad physical understanding. PhysBrain 1.0 studies a complementary route: converting large-scale human egocentric video into structured physical commonsense supervision before robot adaptation. Our data engine extracts scene elements, spatial dynamics, action execution, and depth-aware relations, then turns them into question-answer supervision for training PhysBrain VLMs. The resulting physical priors are further transferred to VLA policies through a capability-preserving and language-sensitive adaptation design. Across multimodal QA benchmarks and embodied control benchmarks, including ERQA, PhysBench, SimplerEnv-WidowX, LIBERO, and RoboCasa, PhysBrain 1.0 achieves SOTA results and shows especially strong out-of-domain performance on SimplerEnv. These results suggest that scaling physical commonsense from human interaction video can provide an effective bridge from multimodal understanding to robot action.

cs.RO↗

arXiv2026

FrameSkip: Learning from Fewer but More Informative Frames in VLA Training

Vision-Language-Action (VLA) policies are commonly trained from dense robot demonstration trajectories, often collected through teleoperation, by sampling every recorded frame as if it provided equally useful supervision. We argue that this convention creates a temporal supervision imbalance: long low-change segments dominate the training stream, while manipulation-critical transitions such as alignment, contact, grasping, and release appear only sparsely. We introduce FrameSkip, a data-layer frame selection framework that scores trajectory frames using action variation, visual-action coherence, task-progress priors, and gripper-transition preservation, then remaps training samples toward high-importance frames under a target retention ratio. Because FrameSkip operates only in the dataloader, it leaves the VLA architecture, action head, training objective, and inference procedure unchanged. Across RoboCasa-GR1, SimplerEnv, and LIBERO, FrameSkip improves the success-retention trade-off over full-frame training and simpler frame selection variants, achieving a macro-average success rate of 76.15% across the three benchmarks compared with 66.50% for full-frame training while using a compressed trajectory view that retains 20% of unique frames in the main setting.

cs.RO↗

arXiv2026

3D-Mix for VLA: A Plug-and-Play Module for Integrating VGGT-based 3D Information into Vision-Language-Action Models

Vision-Language-Action (VLA) models leverage Multimodal Large Language Models (MLLMs) for robotic control, but recent studies reveal that MLLMs exhibit limited spatial intelligence due to training predominantly on 2D data, resulting in inadequate 3D perception for manipulation tasks. While recent approaches incorporate specialized 3D vision models such as VGGT to enhance spatial understanding, they employ diverse integration mechanisms without systematic investigation, leaving the optimal fusion strategy unclear. We conduct a comprehensive pilot study comparing nine VGGT integration schemes on standardized benchmarks and find that semantic-conditioned gated fusion, which adaptively balances 2D semantic and 3D geometric features based on task context, achieved the strongest performance among all nine evaluated fusion schemes in our pilot study. We present 3D-Mix, a plug-and-play module that integrates into diverse VLA architectures (GR00T-style and $\pi$-style) without modifying existing MLLM or action expert components. Experiments across six MLLM series (nine model variants, 2B--8B parameters) on SIMPLER and LIBERO show that 3D-Mix delivers consistent performance gains, averaging +7.0% on the out-of-domain (OOD) SIMPLER benchmark across all nine GR00T-style variants, establishing a principled approach for enhancing spatial intelligence in VLA systems.

cs.RO↗

arXiv2024

Resolving the Valence of Iron Oxides by Resonant Photoemission Spectroscopy

Precisely determining the oxidation states of metal cations within variable-valence transition metal oxides remains a significant challenge, yet it is crucial for understanding and predicting the properties of these technologically important materials. Iron oxides, in particular, exhibit a remarkable diversity of electronic structures due to the variable valence states of iron (Fe2+ and Fe3+), however, quantitative analysis using conventional X-ray photoelectron spectroscopy (XPS) is challenging because of significant overlapping of the Fe2p spectra among different oxidation states. In this study, we leverage the intriguing case of Pt supported FeO2 phase of monolayer thickness (ML) as a model system and employ Resonant Photoemission Spectroscopy (ResPES) to directly quantify the cation valence states and compositional ratios in this complex Fe oxide. Our results reveal that this ultrathin FeO2 film (Pt-O-Fe-O), contrary to the +3 valence predicted by density functional theory (DFT), consists of an equal mixture of Fe2+ and Fe3+ cations, yielding an average valence of +2.5. Structurally, FeO2 is likely derived from the Fe3O4 sublattice, featuring an octahedral Fe layer (50% Fe3+ and 50% Fe2+) bonded to upper and lower oxygen layers.

cond-mat.mtrl-sci↗

arXiv2019

Measuring Long-term Impact of Ads on LinkedIn Feed

Organic updates (from a member's network) and sponsored updates (or ads, from advertisers) together form the newsfeed on LinkedIn. The newsfeed, the default homepage for members, attracts them to engage, brings them value and helps LinkedIn grow. Engagement and Revenue on feed are two critical, yet often conflicting objectives. Hence, it is important to design a good Revenue-Engagement Tradeoff (RENT) mechanism to blend ads in the feed. In this paper, we design experiments to understand how members' behavior evolve over time given different ads experiences. These experiences vary on ads density, while the quality of ads (ensured by relevance models) is held constant. Our experiments have been conducted on randomized member buckets and we use two experimental designs to measure the short term and long term effects of the various treatments. Based on the first three months' data, we observe that the long term impact is at a much smaller scale than the short term impact in our application. Furthermore, we observe different member cohorts (based on user activity level) adapt and react differently over time.

cs.SI↗

arXiv2012

Identities in Nonlinear Realizations of Supersymmetry

In this paper, we emphasize that a UV SUSY-breaking theory can be realized either linearly or nonlinearly. Both realizations form the dual descriptions of the UV SUSY-breaking theory. Guided by this observation, we find subtle identities involving the Goldstino field and matter fields in the standard nonlinear realization from trivial ones in the linear realization. Rather complicated integrands in the standard nonlinear realization are identified as total-divergences. Especially, identities only involving the Goldstino field reveal the self-consistency of the Grassmann algebra. As an application of these identities, we prove that the nonlinear Kahler potential without or with gauge interactions is unique, if the corresponding linear one is fixed. Our identities pick out the total-divergence terms and guarantee this uniqueness.

hep-th↗

arXiv2012

f(R) Gravities, Killing Spinor Equations, "BPS" Domain Walls and Cosmology

We derive the condition on f(R) gravities that admit Killing spinor equations and construct explicit such examples. The Killing spinor equations can be used to reduce the fourth-order differential equations of motion to the first order for both the domain wall and FLRW cosmological solutions. We obtain exact "BPS" domain walls that describe the smooth Randall-Sundrum II, AdS wormholes and the RG flow from IR to UV. We also obtain exact smooth cosmological solutions that describe the evolution from an inflationary starting point with a larger cosmological constant to an ever-expanding universe with a smaller cosmological constant. In addition, We find exact smooth solutions of pre-big bang models, bouncing or crunching universes. An important feature is that the scalar curvature R of all these metrics is varying rather than a constant. Another intriguing feature is that there are two different f(R) gravities that give rise to the same "BPS" solution. We also study linearized f(R) gravities in (A)dS vacua.

hep-th↗

arXiv2012

f(R) Theories of Supergravities and Pseudo-supergravities

We present f(R) theories of ten-dimensional supergravities, including the fermionic sector up to the quadratic order in fermion fields. They are obtained by performing the conformal scaling on the usual supergravities to the f(R) frame in which the dilaton becomes an auxiliary field and can be integrated out. The f(R) frame coincides with that of M-theory, D2-branes or NS-NS 5-branes. We study various BPS p-brane solutions and their near-horizon AdS \times sphere geometries in the context of the f(R) theories. We find that new solutions emerge with global structures that do not exist in the corresponding solutions of the original supergravity description. In lower dimensions, We construct the f(R) theory of N=2, D=5 gauged supergravity with a vector multiplet, and that for the four-dimensional U(1)^4 gauged theory with three vector fields set equal. We find that some previously-known BPS singular "superstars" become wormholes in the f(R) theories. We also construct a large class of f(R) (gauged) pseudo-supergravities. In addition we show that the breathing mode in the Kaluza-Klein reduction of Gauss-Bonnet gravity on S^1 is an auxiliary field and can be integrated out.

hep-th↗

arXiv2011

Leading-Order Actions of Goldstino Fields

This paper starts with a self-contained discussion of the so-called Akulov-Volkov action S_AV, which is traditionally taken to be the leading-order action of Goldstino field. Explicit expressions for S_AV and its chiral version S_AV^ch are presented. We then turn to the issue on how these actions are related to the leading-order action S_NL proposed in the newly proposed constrained superfield formalism. We show that S_NL may yield S_AV/S_AV^ch or a totally different action S_KS, depending on how the auxiliary field in the former is integrated out. However, S_KS and S_AV/S_AV^ch always yield the same S-matrix elements, as one would have expected from general considerations in quantum field theory.

hep-th↗

arXiv2011

Gauged Kaluza-Klein AdS Pseudo-supergravity

We obtain the pseudo-supergravity extension of the D-dimensional Kaluza-Klein theory, which is the circle reduction of pure gravity in D+1 dimensions. The fermionic partners are pseudo-gravitino and pseudo-dilatino. The full Lagrangian is invariant under the pseudo-supersymmetric transformation, up to quadratic order in fermion fields. We find that the theory possesses a U(1) global symmetry that can be gauged so that all the fermions are charged under the Kaluza-Klein vector. The gauging process generates a scalar potential that has a maximum, leading to the AdS vacuum. Whist the highest dimension for gauged AdS supergravity is seven, our gauged AdS pseudo-supergravities can exist in arbitrary dimensions.

hep-th↗

arXiv2011

Killing Spinors for the Bosonic String and the Kaluza-Klein Theory with Scalar Potentials

The paper consists mainly of two parts. In the first part, we obtain well-defined Killing spinor equations for the low-energy effective action of the bosonic string with the conformal anomaly term. We show that the conformal anomaly term is the only scalar potential that one can add into the action that is consistent with the Killing spinor equations. In the second part, we demonstrate that the Kaluza-Klein theory can be gauged so that the Killing spinors are charged under the Kaluza-Klein vector. This gauging process generates a scalar potential with a maximum that gives rise to an AdS spacetime. We also construct solutions of these theories.

hep-th↗

arXiv2011

On Black Hole Stability in Critical Gravities

We consider extended cosmological gravities with Ricci tensor and scalar squared terms in diverse dimensions. These theories admit solutions of Einstein metrics, including the Schwarzschild-Tangherlini AdS black holes, whose mass and entropy vanish at the critical point. We perform linearized analysis around the black holes and show that in general the spectrum consists of the usual spin-2 massless and ghost massive modes. We demonstrate that there is no exponentially-growing tachyon mode in the black holes. At the critical point, the massless spin-2 modes have zero energy whilst the massive spin-2 modes are replaced by the log modes. There always exist certain linear combination of massless and log modes that has negative energy. Thus the stability of the black holes requires that the log modes to be truncated out by the boundary condition.

hep-th↗

arXiv2011

Critical Points of D-Dimensional Extended Gravities

We study the parameter space of D-dimensional cosmological Einstein gravity together with quadratic curvature terms. In D>4 there are in general two distinct (anti)-de Sitter vacua. We show that for appropriate choice of the parameters there exists a critical point for one of the vacua, for which there are only massless tensor, but neither massive tensor nor scalar, gravitons. At criticality, the linearized excitations have vanishing energy (as do black hole solutions). A further restriction of the parameters gives a one-parameter cosmological Einstein plus Weyl^2 model with a unique vacuum, whose Λis determined.

hep-th↗

arXiv2010

Thermodynamical Metrics and Black Hole Phase Transitions

An important phase transition in black hole thermodynamics is associated with the divergence of the specific heat with fixed charge and angular momenta, yet one can demonstrate that neither Ruppeiner's entropy metric nor Weinhold's energy metric reveals this phase transition. In this paper, we introduce a new thermodynamical metric based on the Hessian matrix of several free energy. We demonstrate, by studying various charged and rotating black holes, that the divergence of the specific heat corresponds to the curvature singularity of this new metric. We further investigate metrics on all thermodynamical potentials generated by Legendre transformations and study correspondences between curvature singularities and phase transition signals. We show in general that for a system with n-pairs of intensive/extensive variables, all thermodynamical potential metrics can be embedded into a flat (n,n)-dimensional space. We also generalize the Ruppeiner metrics and they are all conformal to the metrics constructed from the relevant thermodynamical potentials.

hep-th↗