已审阅并标注「强化学习」的研究与资产。
条目依据已审阅标签归类，同一主题不表示资源之间存在依赖关系。日期保留来源精度，注释保留原语言。
Introduces asynchronous distributed reinforcement learning with centralized updates and decentralized interaction collection for device-control agents.
Combines offline initialization and online reinforcement learning in Android environments, with automated evaluation and curriculum-based sampling.
Combines supervised and reinforcement learning for MiniWoB web tasks and investigates whether agents understand HTML structure or memorize target elements.
Introduces GLAINTEL, a Flan-T5-based product-search web agent studied under unsupervised learning, demonstration-based training, and unsupervised domain adaptation. It compares ways to combine demonstrations with reinforcement learning.
Builds a web agent through simplified HTML observations, human-and-AI browsing data, curriculum training, and reinforcement-learning refinement.
Trains a screenshot-based instruction-grounding model to predict target coordinates, using reinforcement learning to improve spatial decoding.
Combines reinforcement learning with behavioral priors from human interactions to learn computer control through keyboard and mouse actions.
Presents an Android environment for training and evaluating reinforcement-learning agents through touchscreen interaction.
Uses workflows extracted from demonstrations to constrain reinforcement-learning exploration for interactive web tasks.
Introduces World of Bits, a web-task platform combining human demonstrations, keyboard and mouse actions, and cached HTTP traffic for repeatable task environments. It evaluates agents trained through behavioral cloning and reinforcement learning.
Over 100 web-interaction environments extending OpenAI's MiniWoB, controlled from Python through a Gymnasium API that drives the browser with Selenium.
A browser-hosted mobile-app simulator with programmable state, parameterized tasks and deterministic state-checking judges for parallel GUI-agent evaluation and online reinforcement learning.
Desktop grounding prompts selected from GroundCUA and formatted for click-based reinforcement learning with target-box rewards.
Training and evaluation Parquet bundles accompanying GUI-R1, including GUI grounding and mobile/web action-prediction data.
Mobile-agent supervision from emulator and simulator trajectories, plus executable tasks with starting states and success checks for reinforcement learning.
UI-AGILE training annotations and screenshots together with converted GUI-grounding and AndroidControl evaluation files.
A public example subset of the UI-S1 mobile training data, packaged as JSONL annotations and Parquet images.
An eight-billion-parameter MiniCPM-V-based model that consumes smartphone screenshots and predicts compact actions for Chinese and English Android app tasks.
UI-TARS-1.5-7B-derived checkpoint trained with ARPO on OSWorld desktop tasks.
Mobile GUI agent research release with technical report, benchmark data, and evaluation tooling from vivo AI Lab.
Larger EvoCUA computer-use agent checkpoint (20260105 release) with about 33B BF16 parameters in the Qwen3-VL architecture, tagged for OSWorld and for English and Chinese.
EvoCUA computer-use agent checkpoint (20260105 release) with about 8.8B BF16 parameters in the Qwen3-VL architecture, tagged for OSWorld and for English and Chinese.
GUI-grounding training code that modifies reasoning templates, reward functions, and the GRPO objective, with a linked model release.
A Qwen2.5-VL-3B checkpoint released for R1-style visual grounding experiments, predicting interface target locations from screenshots and instructions.
A Qwen2.5-VL grounding checkpoint trained with Gaussian point and coverage rewards whose granularity adapts to interface-element scale.
GUI action-model training and evaluation code with verifiable rewards, plus linked reinforcement-learning data and checkpoints.
A single Hub release containing GUI-R1 3B and 7B action-model checkpoints trained with reinforcement learning across interface platforms.
A GUI-grounding checkpoint trained with adaptive exploration rewards and multiple candidate answers to improve instruction-to-element alignment.
3B GUI agent trained in two stages: distilling spatial reasoning from teacher models, then reinforcement learning that rewards accurate sub-goals and trains error recovery.
A Qwen2.5-VL checkpoint trained with the Actor2Reasoner procedure to combine interface grounding with planning and recovery reasoning.
A GLM-4.1V-based mobile action checkpoint released by the MobileRL project for Android interaction.
A released LLaVA checkpoint for visual browser interaction from the PAE autonomous task-proposal and policy-learning system.
GUI agent trained by RL with a continuous reward, a simple-thinking reward and cropping-based resampling on about 9k samples, plus a plug-in decomposed grounding-and-selection inference method.
The released UI-AGILE model for GUI grounding with reward shaping, crop-based resampling and decomposed selection during inference.
Rule-based reinforcement learning for GUI action prediction, releasing UI-R1-3B, trained from Qwen2.5-VL-3B, the efficiency-oriented UI-R1-E-3B, and training data.
A Qwen2.5-VL checkpoint trained with reinforcement learning for GUI action prediction and evaluated on mobile, desktop and web grounding tasks.
A mobile GUI checkpoint trained with semi-online reinforcement learning that uses offline trajectories to improve multi-turn action prediction.
A WebRL browser-policy checkpoint for performing actions in the five website environments of WebArena.
The outcome-supervision checkpoint used by WebRL to score browser trajectories during policy training.
Android device-control platform with screenshots, accessibility telemetry, input actions, and an optional reinforcement-learning task interface.
Replay-based policy-optimization framework for GUI agents, with distributed OSWorld rollouts and screenshot/action histories.
Offline device-control training code that learns action values and uses them to select promising actions, with linked trajectories and checkpoints.
Distributed online reinforcement-learning framework for Android agents, with environment setup and trajectory collection tools.
Code for screenshot grounding through supervised training, teacher-guided distillation, reinforcement learning, and benchmark evaluation.
GUI-grounding research code for spatial voting across sampled predictions and test-time policy optimization using region-consistency rewards.
SDK for RL environments and evals spanning coding, browser, computer-use and robotics agents, whose environments expose SSH, MCP, CDP or VNC capabilities behind a minimal start-and-grade protocol.
Android GUI-agent adaptation framework combining app exploration, generated curricula and rollout feedback with hint-conditioned step-level GRPO; the documented public training path targets Qwen3-VL-8B.
Online agentic RL framework for mobile GUI agents built around the difficulty-adaptive ADAGRPO algorithm, with Dockerized AndroidWorld and AndroidLab evaluation and a MobileRL-9B checkpoint.
A browser-agent training system that proposes tasks, collects visual interaction trajectories, and uses automated evaluation to drive policy improvement.
Computer-use training framework with VeriGen task synthesis, adaptive rollout sampling, and visual-context segmentation.
An archived Gym-compatible client that exposes screen observations and keyboard/mouse actions over VNC for remote application and browser environments, including the historical World of Bits integration.
Training code that learns a value environment model from offline Android interaction data and uses its frozen value estimates to guide GUI policy optimization.
Environment setup, multi-turn reinforcement-learning and evaluation tooling for browser agents interacting with WebArena.
Training pipeline that evolves web tasks from prior attempts and uses an outcome reward model to refine browser policies.
GUI-grounding training and evaluation code with same-family distillation, GRPO recipes and point-in-box scoring, using a Windows-style web environment.
Research code that learns browser-control policies by constraining exploration with workflows induced from demonstrations, with a Selenium-based MiniWoB environment and experiment harness.
第 181–240 条，共 242 条资源。
通过主题导出清单获取完整记录和证据。克隆目录后，可将主题记录直接交给标准 RAG 与引用导出工具。
