• Home
  • Search
  • Web Agent Agentic Reinforcement Learning Decision Model Under Multi-Cost and Failure Risk Constraints
  • https://doi.org/10.20944/preprints202602.0095.v1Copy DOI Icon

Web Agent Agentic Reinforcement Learning Decision Model Under Multi-Cost and Failure Risk Constraints

Show More
  • Abstract
  • Literature Map
  • Similar Papers
Abstract

Intelligent agent interactions in real-world web environments are commonly constrained by request budgets,time delays, anti-crawling restrictions, and operational failure risks. Strategies solely optimizing task successrates often exhibit unusable phenomena such as "high success but high cost" or "low risk but conservativefailure."This paper proposes a constrained Agentic reinforcement learning model for Web Agents, unifyingpage access, search requests, and external API calls into a unified long-term decision-making framework withassociated costs. It simultaneously incorporates cost budget constraints and tail risk control into theoptimization objective: constructing a multidimensional cost vector comprising cumulative request count,total latency, and failure penalties to achieve budget compliance via Lagrange dual updates;while employing aCVaR risk term to suppress excessive exploration of high-failure-probability paths, thereby achievingadaptive balance among "completion rate, cost, and risk."Experiments were conducted across 30–70site/page templates and 800–1,500 end-to-end web tasks (including information extraction, pricecomparison, form submission, and cross-page navigation). Interaction sequences spanned 20–120 steps withtool scales of 30–200. Performance was benchmarked against unconstrained RL, budget-constrained RL, andrule-based/scripted web agents, quantifying task completion rates, cost-per-success, failure rates, and policystability.scripted web agents. We quantified task completion rates, cost-per-success, failure rates, and policystability.Results demonstrate that at equivalent completion rates, our method reduces C-PS by 22%–31% andlowers failure rates by 18%–26% under high failure penalties. Under fixed budgets, task completion ratesincrease by 10%–16%, highlighting the necessity and effectiveness of constraint modeling for practical WebAgent deployment.

Similar Papers
  • Research Article
  • Citations9

WebPilot: A Versatile and Autonomous Multi-Agent System for Web Task Execution with Strategic Exploration

  • Apr 11, 2025
  • Proceedings of the AAAI Conference on Artificial Intelligence
  • Yao Zhang +5
  • Research Article
  • Citations71

If it hasn’t failed, does it work? On ‘the worst we can expect’ from observational trial results, with reference to airway management devices

  • May 07, 2012
  • Anaesthesia
  • J J Pandit
  • Research Article
  • Citations5

Confidence limits on the failure rate and a rapid projection nomogram for the lognormal distribution

  • Jan 01, 1984
  • Microelectronics Reliability
  • A.S Jordan
  • Conference Article
  • Citations12

Methods and Costs to Achieve Ultra Reliable Life Support

  • Jul 15, 2012
  • Harry Jones
  • Conference Article
  • Citations2

Reliability Prediction Using Additive Weibull Model

  • Nov 05, 2024
  • SAE technical papers on CD-ROM/SAE technical paper series
  • Dasharath Baliram Koulage +2
  • Research Article
  • Citations8

On the suitability of insourced Additive Manufacturing for spare parts management

  • Jan 01, 2022
  • IFAC-PapersOnLine
  • F Lolli +3
  • Research Article
  • Citations2

HR-SC—an academic-developed machine learning framework to classify HRD-positive ovarian cancer patients and predict sensitivity to olaparib

  • May 19, 2025
  • ESMO Open
  • L Beltrame +27
  • Discussion
  • Citations1

Medical vs surgical therapy in preventing visual field loss

  • Feb 01, 2003
  • Ophthalmology
  • Lenworth N Johnson
  • Research Article

Timing of intervention: Assessing early vs late endoscopic and surgical interventions in acute cholangitis

  • Jan 16, 2026
  • World Journal of Gastrointestinal Endoscopy
  • Silpa Choday +2
  • Research Article
  • Citations7

Extension of the Duane Plotting Technique

  • Jun 01, 1985
  • IEEE Transactions on Reliability
  • D Stewart Peck
  • Research Article
  • Citations198

A Multihop Task Offloading Decision Model in MEC-Enabled Internet of Vehicles

  • Feb 15, 2023
  • IEEE Internet of Things Journal
  • Chen Chen +4
  • Abstract
  • Citations1

PD10-06 THE TENSION FREE VAGINAL TAPE VERSUS THE PUBOVAGINAL SLINGS FOR WOMEN WITH NEUROGENIC STRESS URINARY INCONTINENCE

  • Mar 31, 2015
  • The Journal of Urology
  • Ahmed El-Azab
  • Research Article
  • Citations40

An Investigation into the Use of a Single Component Self-etching Primer Adhesive System for Orthodontic Bonding: a randomized controlled clinical trial

  • Mar 01, 2006
  • Journal of Orthodontics
  • K House +2
  • Research Article

DESENVOLVIMENTO DE UMA FERRAMENTA EM PYTHON PARA ANÁLISE DE DADOS DE TEMPO DE FALHA COM APLICAÇÕES EM ENGENHARIA DE MANUTENÇÃO

  • Jul 15, 2025
  • ARACÊ
  • João Vitor Paiva Luz +1
  • Research Article
  • Citations3

Impact of Simulation-Based Surgical Training in Laparoscopy on Satisfaction Level and Proficiency in Surgical Skills.

  • Feb 01, 2025
  • Advances in medical education and practice
  • Fareeda Mukhtar +4
Cactus Communications logo

Copyright 2026 Cactus Communications. All rights reserved.