Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.08432

EMHO: EMbodied Agent Harness Optimization via Experience Traces

Hyun Jung Lee·Jungtaek Kim·Jongwon Jeong·Tae-Eui Kam·Donghyun Kim·Yong Jae Lee
View original source ↗Read source document ↗Export metadata ↓

Abstract

Improving embodied agents often focuses on optimizing the underlying model through training, while the surrounding agent harness that controls planning, context, and tool use is typically engineered. We ask whether this harness can instead improve itself directly from experience traces under sparse environmental feedback. We propose EMbodied Agent Harness Optimization (EMHO), a self-evolving framework that keeps the embodied model frozen and iteratively revises its harness by analyzing execution trajectories and prior harness history. EMHO optimizes beyond skills or recovery prompts, modifying how the agent monitors progress, uses vision tools, grounds observations, and responds to failures. To support multiple subtasks with a single harness, we introduce EMHO-Merge, which addresses trade-offs in jointly optimizing a single shared harness across subtasks by using episode-level gains and losses to guide evidence-supported refinement of when and how revised behaviors are applied. We evaluate EMHO on EmbodiedBench across navigation and manipulation tasks, and EMHO consistently improves task success for both Qwen 9B and 27B models. Qualitative analysis shows that EMHO goes beyond recovering from failures and unproductive actions to reshape how the embodied agent interprets and interacts with its environment.

Explore related subjects

cs.AI

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Hyun Jung Lee, Jungtaek Kim, Jongwon Jeong, Tae-Eui Kam, Donghyun Kim, Yong Jae Lee. 2026-10-06. EMHO: EMbodied Agent Harness Optimization via Experience Traces. https://arxiv.org/abs/2610.08432

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

FREIDA: A Framework for developing quantitative agent based models based on qualitative expert knowledge

Agent Based Models (ABMs) often deal with systems where there is a lack of quantitative data or where quantitative data alone may be insufficient to fully capture the complexities of real-world systems. Expert knowledge and qualitative insights, such as those obtained through interviews, ethnographic research, historical accounts, or participatory workshops, are critical in constructing realistic behavioral rules, interactions, and decision-making processes within these models. However, there is a scarcity of systematic approaches that are able to incorporate both qualitative and quantitative data across the entire modeling cycle. To address this, we propose FREIDA, a systematic mixed-methods framework to develop, train, and validate ABMs, particularly in data-sparse contexts. The main technical innovation introduced within this framework is the extraction of what we call Expected System Behaviors (ESBs) from qualitative data, which are testable statements evaluated through model simulations. Divided into Calibration Statements (CS) for model calibration and Validation Statements (VS) for model validation, ESBs underpin a rigorous evaluation mechanism on the same footing as quantitative data. By structuring qualitative insights as explicit model constraints, FREIDA creates a transparent foundation for applying established modelling practices such as Sensitivity Analysis (SA) and Uncertainty Quantification (UQ), allowing modellers to assess parameter influence, model robustness, and remaining uncertainties in a systematic manner. Through this, qualitative insights can inform not only model specification but also parameterization, validation, and continuous improvement of model reliability and fitness for purpose, addressing a long-standing challenge in agent-based modeling. We illustrate the application of FREIDA through a case study of criminal cocaine networks in the Netherlands.

cs.AI↗
arXiv2026

Towards Explainable Conversational AI for Early Diagnosis with Large Language Models

Healthcare systems around the world are grappling with issues such as inefficient diagnostics, rising costs, and limited access to specialists. These challenges often contribute to delays in treatment and poorer health outcomes. Most existing AI and deep learning based health assessment systems offer limited interactivity and transparency, reducing their usefulness for user-centered health support. This research introduces a conversational chatbot powered by a Large Language Model (LLM), using GPT-4o, Retrieval-Augmented Generation, and explainable AI techniques. The chatbot engages users in a dynamic conversation to extract and normalize symptoms while identifying and ranking potential health conditions through similarity matching and adaptive questioning. Using Chain-of-Thought prompting, the system also provides more transparent explanations of its reasoning process. When evaluated against traditional machine learning models, including Naive Bayes, Logistic Regression, SVM, Random Forest, and KNN using both TF-IDF and CountVectorizer feature extraction, the proposed LLM-based system achieved a Top-1 accuracy of 90% and a Top-3 accuracy of 100%. The system was additionally evaluated through a cross-sectional expert evaluation involving 17 physicians across all 14 conditions, with the results indicating generally favorable assessments of conversational quality, early diagnostic plausibility, and safety-related criteria. These findings demonstrate the potential of explainable conversational AI as a health and well-being support tool for early symptom assessment. However, the proposed system is not intended for clinical diagnosis or clinical decision-making, and further validation would be required before any use in healthcare practice.

cs.AI↗
arXiv2026

Persuasion Propagation: Does Persuasion Change AI Agent Behavior?

AI agents combine normal conversation with autonomous task execution, allowing earlier context to shape how later tasks are executed. Yet studying this possibility is challenging because agent behavior is noisy and costly to reproduce, and observed behavioral changes can be difficult to separate from generic context sensitivity. We first ask whether task-irrelevant persuasion changes how agents act, a phenomenon we call \emph{persuasion propagation}, and whether persuasion susceptibility explains variation in these effects. We show that persuasive interaction leaves a measurable, task-dependent behavioral footprint on downstream agent execution. In web research, it increases execution time by 56\% and domains visited by 16.3\% relative to a baseline, while coding shows faster execution with larger revisions. However, agents that are more susceptible to persuasive conversational context do not reliably predict the magnitude of the downstream behavioral shifts. This motivates behavior-level evaluation of persuasion in AI agents.

cs.AI↗