arXiv subjects

Harim Song

Publications and source records attributed to Harim Song.

3 recordsLinked to original sources

arXiv2026

Binding Paraphrased Instructions to Encountered Objects in sLLM Agents under Partial Observability

Reliable deployment of LLM agents requires grounding diverse user expressions in their environments. Although RL-finetuned smaller large language model (sLLM) agents rival larger prompted LLMs on embodied tasks, their robustness to instruction paraphrases remains unclear. We introduce ALFWorld-Para to evaluate this robustness under controlled paraphrases, revealing 21-52% declines in success rates (SRs), predominantly driven by object synonyms. Despite retaining knowledge of synonym referents, agents often fail to take the corresponding objects even after observing them, exposing a gap between referential knowledge and object binding. Partial observability further complicates binding, as referents may remain unseen or have already been encountered under different labels. To address this challenge, we propose BINDIntent, a training-free framework that probabilistically binds referring expressions to observed labels using semantic proximity and estimated persistence against unseen candidates. Without additional training or LLM calls, BINDIntent improves synonym-paraphrase success of RL-finetuned sLLMs by 17.5-49.1 percentage points (pp) with only 0.4 ms of CPU overhead per step. It also raises GiGPO-7B's overall SR from 61.8% to 77.4%, surpassing a prompted 24B LLM. Code and benchmark are available at https://github.com/cau-hai-lab/ALFWorld-Para.

cs.LG↗

arXiv2026

Quantitative control and recording of materials-synthesis processes using an automated experimentation platform

Data-driven materials development requires the collection of large amounts of high-quality materials data. Full autonomy of materials experiments is anticipated, but its technical hurdles are high and its adoption remains limited. In this study, we constructed a simple, easy-to-deploy automated experimentation platform that focuses not on full autonomy but on the reliable automation and quantitative recording of experimental processes. Specifically, commercially available instruments such as robot arms, electric pipettes, web cameras, and an electronic balance are combined, components such as fixtures are fabricated with a 3D printer, and the instruments are operated by control code generated by an AI agent based on a large language model. As a demonstration, we applied the platform to a two-solution mixing experimental system and synthesized ZIF-8, a metal-organic framework. A white suspension phase was observed in the product, and X-ray diffraction measurements confirmed that it was ZIF-8. We also found that its particle size distribution depends strongly on the solution dispensing speed of the electric pipette, which is a parameter that is difficult to control or record in manual operation. This dependence was reproduced in repeated runs, confirming the repeatability of the automated synthesis. This result is a good example showing that the control and recording of process parameters that are rarely quantified in manual work can govern the quality of materials data. All control code, CAD models, and documentation are made publicly available to encourage the spread of laboratory-scale automation of experiments.

cond-mat.mtrl-sci↗

arXiv2026

Propagation-mediated amplification of \{11\={2}0\}-biased inversion domain boundary alignment in polarity-mixed GaN lateral overgrowth

GaN polarity inversion and the associated inversion domain boundaries (IDBs) are frequently observed during lateral overgrowth and are often discussed in terms of the small energetic spread among competing IDB structures predicted by first-principles calculations. In circular mask openings, \(\{11\bar{2}0\}\)-aligned IDBs have previously been explained by geometric closure of a single-polarity hexagonal domain at the circular boundary. Here we examine an experimentally distinct regime in which opposite-polarity domains already coexist within the opening before the later development of long, straight IDB traces. In this mixed-polarity regime, the final trace orientation cannot be attributed solely to the macroscopic circular boundary. Nevertheless, plan-view SEM line-trace statistics show that IDB orientations remain biased toward the \(\{11\bar{2}0\}\) family. To quantify how this bias develops during propagation, we perform distance-resolved, length-weighted orientation analysis in concentric annular regions defined from the opening center. The resulting metrics show that \(\{11\bar{2}0\}\)-biased alignment is progressively amplified with propagation distance, while the orientation distribution becomes narrower, indicating systematic sharpening of the preferred alignment state. We further apply the same ring-resolved statistical operators to minimal two-domain propagation simulations in a circular opening and find that a propagation-mediated anisotropy reproduces the observed radial amplification under fixed circular geometry. Together, these results establish a quantitative phenomenology of \(\{11\bar{2}0\}\)-biased IDB alignment in polarity-mixed GaN lateral overgrowth on patterned sapphire and indicate that, although mask-boundary-imposed selection may describe single-polarity closure cases, the present mixed-polarity regime is better explained by propagation-mediated amplification.

cond-mat.mtrl-sci↗