This paper proposes the LAMO framework to address the challenge of deploying lightweight MLLM-powered autonomous GUI agents on resource-constrained devices. LAMO enhances lightweight MLLMs with GUI-specific knowledge and task scalability through multi-role orchestration.
The paper introduces the "Turing Test on Screen" to model anti-detection for autonomous GUI agents, emphasizing humanization for their survival in digital ecosystems. It establishes the Agent Humanization Benchmark (AHB) and proposes methods to improve agents' imitability and utility, utilizing a new dataset of mobile touch dynamics.
RiskWebWorld is presented as the first highly realistic interactive benchmark for evaluating GUI agents in e-commerce risk management, addressing their underexplored effectiveness in high-stakes investigative domains. It features 1,513 tasks from production risk-control pipelines and a Gymnasium-compliant infrastructure for scalable evaluation, revealing a dramatic capability gap across diverse models.
UI-Oceanus é uma estrutura que escala agentes GUI generalistas, focando em dominar a física da interação através de feedback ambiental em vez de imitar trajetórias. O sistema utiliza exploração autônoma e predição de dinâmicas futuras para construir um modelo de mundo interno robusto, superando limitações de dados e supervisão.