{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/arpo-end-to-end-policy-optimization-for-gui","title":"ARPO:End-to-End Policy Optimization for GUI Agents with Experience Replay","arxiv_id":"2505.16282","date":"2025-05-22","proceeding":null,"authors":["Fanbin Lu","Zhisheng Zhong","Shu Liu","Chi-Wing Fu","Jiaya Jia"],"abstract":"Training large language models (LLMs) as interactive agents for controlling graphical user interfaces (GUIs) presents a unique challenge to optimize long-horizon action sequences with multimodal feedback from complex environments. While recent works have advanced multi-turn reinforcement learning (RL) for reasoning and tool-using capabilities in LLMs, their application to GUI-based agents remains relatively underexplored due to the difficulty of sparse rewards, delayed feedback, and high rollout costs. In this paper, we investigate end-to-end policy optimization for vision-language-based GUI agents with the aim of improving performance on complex, long-horizon computer tasks. We propose Agentic Replay Policy Optimization (ARPO), an end-to-end RL approach that augments Group Relative Policy Optimization (GRPO) with a replay buffer to reuse the successful experience across training iterations. To further stabilize the training process, we propose a task selection strategy that filters tasks based on baseline agent performance, allowing the agent to focus on learning from informative interactions. Additionally, we compare ARPO with offline preference optimization approaches, highlighting the advantages of policy-based methods in GUI environments. Experiments on the OSWorld benchmark demonstrate that ARPO achieves competitive results, establishing a new performance baseline for LLM-based GUI agents trained via reinforcement learning. Our findings underscore the effectiveness of reinforcement learning for training multi-turn, vision-language GUI agents capable of managing complex real-world UI interactions. Codes and models:https://github.com/dvlab-research/ARPO.git.","url_abs":"https://arxiv.org/abs/2505.16282v1","url_pdf":"https://arxiv.org/pdf/2505.16282v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"arpo-end-to-end-policy-optimization-for-gui","repo_url":"https://github.com/dvlab-research/arpo","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"},{"task_slug":"reinforcement-learning-2","task_name":"reinforcement-learning"}],"methods":[{"method_slug":"focus","method_name":"Focus"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":"https://syntology.ai/paper/2505.16282","atlas_url":"https://app.syntology.ai/?focus=2505.16282","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2505.16282"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-25T09:33:49+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/dvlab-research/arpo","reach":{"status":"ok","spdx":"Apache-2.0"}}],"summary":{"ran":6,"ran_honours":1,"unverified":9},"by_repo_kind":{"official":{"samples":16,"ran":7,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"07fd593c156d2586","entry":"escape_single_quotes","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/trainer/gui_agent.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/trainer/gui_agent.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"07fd593c156d2586"}},{"code_sha256_prefix":"33a3cccef610855c","entry":"get_kl_controller","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/trainer/core_algos.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/trainer/core_algos.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"33a3cccef610855c"}},{"code_sha256_prefix":"c242cf8145d74b19","entry":"parse_action","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/trainer/gui_agent.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/trainer/gui_agent.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"c242cf8145d74b19"}},{"code_sha256_prefix":"4049cf08f994464e","entry":"prepare_fa2_from_position_ids","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/models/transformers/flash_attention_utils.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/models/transformers/flash_attention_utils.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"4049cf08f994464e"}},{"code_sha256_prefix":"4b07ccef58be043a","entry":"reduce_metrics","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/trainer/metrics.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/trainer/metrics.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"4b07ccef58be043a"}},{"code_sha256_prefix":"e252767324188623","entry":"round_by_factor","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/trainer/gui_agent.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/trainer/gui_agent.py","link_basis":"harvester_set","language":"python","status":"ran_honours","verification_level":1,"contract_check":"HONOURS","metamorphic_tier":null,"behaviour_fingerprint":true,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"e252767324188623"}},{"code_sha256_prefix":"a91c18341292d50e","entry":"unpad_dataproto","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/protocol.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/protocol.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"a91c18341292d50e"}},{"code_sha256_prefix":"69132eba50f047ff","entry":"collate_fn","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/utils/dataset.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/utils/dataset.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"69132eba50f047ff"}},{"code_sha256_prefix":"d3ee7b3d8f0cf2fa","entry":"compute_gae_advantage_return","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/trainer/core_algos.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/trainer/core_algos.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"d3ee7b3d8f0cf2fa"}},{"code_sha256_prefix":"b7dacadfa17e35d2","entry":"compute_grpo_outcome_advantage","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/trainer/core_algos.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/trainer/core_algos.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"b7dacadfa17e35d2"}},{"code_sha256_prefix":"d71e909038b965a1","entry":"flash_attention_forward","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/models/transformers/flash_attention_utils.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/models/transformers/flash_attention_utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"d71e909038b965a1"}},{"code_sha256_prefix":"8c950c2610637eaa","entry":"get_rope_index","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/models/transformers/qwen2_vl.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/models/transformers/qwen2_vl.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"8c950c2610637eaa"}},{"code_sha256_prefix":"3d40f4d5f0572fce","entry":"pad_dataproto_to_divisor","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/protocol.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/protocol.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"3d40f4d5f0572fce"}},{"code_sha256_prefix":"e6be35b8d742e7d6","entry":"qwen2_vl_attn_forward","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/models/transformers/qwen2_vl.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/models/transformers/qwen2_vl.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"e6be35b8d742e7d6"}},{"code_sha256_prefix":"62d02a7c36f65d6d","entry":"qwen_2_mixed_modality_forward","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/models/transformers/qwen2_vl.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/models/transformers/qwen2_vl.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"62d02a7c36f65d6d"}},{"code_sha256_prefix":"ad658a1d6fe829ba","entry":"union_tensor_dict","repo":"dvlab-research/arpo","repo_kind":"official","path":"verl/protocol.py","file_url":"https://github.com/dvlab-research/arpo/blob/HEAD/verl/protocol.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"ad658a1d6fe829ba"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}