TRON introduces a generator–verifier framework for visual reasoning RL, where Python-based environments continuously generate new images, questions, and exactly verifiable rewards on demand. The current suite contains 520 environments covering spatial, mathematical, diagram, pattern/logic, and counting reasoning tasks.</p>\n","updatedAt":"2026-06-03T05:33:48.889Z","author":{"_id":"6534835bdea545ecda914773","avatarUrl":"/avatars/7d6e0fc4e348f1ef37793a2147406181.svg","fullname":"Tianze Yang","name":"ytz009","type":"user","isPro":true,"isHf":false,"isHfAdmin":false,"isMod":false,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.8159855008125305},"editors":["ytz009"],"editorAvatarUrls":["/avatars/7d6e0fc4e348f1ef37793a2147406181.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.01599","authors":[{"_id":"6a1fbabce292c1c78ecb1554","name":"Tianze Yang","hidden":false},{"_id":"6a1fbabce292c1c78ecb1555","name":"Yucheng Shi","hidden":false},{"_id":"6a1fbabce292c1c78ecb1556","name":"Ruitong Sun","hidden":false},{"_id":"6a1fbabce292c1c78ecb1557","name":"Jingyuan Huang","hidden":false},{"_id":"6a1fbabce292c1c78ecb1558","name":"Ninghao Liu","hidden":false},{"_id":"6a1fbabce292c1c78ecb1559","name":"Jin Sun","hidden":false}],"publishedAt":"2026-06-01T02:52:49.000Z","submittedOnDailyAt":"2026-06-03T00:00:00.000Z","title":"TRON: Targeted Rule-Verifiable Online Environments for Visual Reasoning RL","submittedOnDailyBy":{"_id":"6534835bdea545ecda914773","avatarUrl":"/avatars/7d6e0fc4e348f1ef37793a2147406181.svg","isPro":true,"fullname":"Tianze Yang","user":"ytz009","type":"user","name":"ytz009"},"summary":"Reinforcement learning (RL) for visual reasoning needs scalable, verifiable, and controllable training signals. Existing visual RL post-training trains on static curated datasets, with fixed image-question-answer samples bounded by their collection budget. In this work, we introduce TRON (Targeted, Rule-verifiable Online eNvironments), an online environment substrate: a training rollout is generated on demand by a controllable generator-verifier program that samples a fresh latent visual state, renders an image, asks a question, and exactly verifies the answer. A single run can therefore draw an unbounded stream of fresh instances at the difficulty level required by the current curriculum. The current TRON suite contains 520 environments organized into five ability buckets (spatial, mathematical, diagram, pattern/logic, and counting); the same substrate supports both a single full model trained on all buckets and per-bucket ability-specialist models, with no additional data collection. We also introduce a substrate analysis covering generation reliability, instance and level diversity, cross-environment near-duplicates, and base-model pass rate by difficulty level. RL post-training with METHOD consistently improves performance on ten external multimodal reasoning benchmarks across Qwen3-VL-4B, Qwen2.5-VL-7B, and MiMo-VL-7B-SFT.","upvotes":10,"discussionId":"6a1fbabde292c1c78ecb155a","projectPage":"https://tron-rl.github.io/","githubRepo":"https://github.com/YangTianze009/TRON","githubRepoAddedBy":"user","ai_summary":"TRON enables scalable and controllable reinforcement learning for visual reasoning through an online environment substrate that generates unlimited diverse training instances with verifiable answers.","ai_keywords":["reinforcement learning","visual reasoning","online environment substrate","controllable generator-verifier program","latent visual state","curriculum learning","ability buckets","multimodal reasoning benchmarks"],"ai_summary_model":"Qwen/Qwen2.5-Coder-32B-Instruct","githubStars":1,"organization":{"_id":"64cf499559503263d9da21df","name":"UniversityofGeorgia","fullname":"University of Georgia","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/64cf47db9cff7382036bf387/LuMBYKv7Iv4kfpJzaJvvE.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"64beb6b6140491ca9f803ebf","avatarUrl":"/avatars/0daa2e813a13668b8b708cd8c12763d9.svg","isPro":false,"fullname":"Yucheng SHi","user":"YuchengShi","type":"user"},{"_id":"65a54be67ec6af0f95d1cc45","avatarUrl":"/avatars/8edf1809fbcf744fc7cea19b83d30fa9.svg","isPro":false,"fullname":"Hao Zhen","user":"wreotben","type":"user"},{"_id":"66926e96f658d6ee95ef6d9d","avatarUrl":"/avatars/cb523e7ebaa5e80bb9cb33715ec1e6e1.svg","isPro":false,"fullname":"Junyao Yang","user":"TberiusJunyao","type":"user"},{"_id":"6534835bdea545ecda914773","avatarUrl":"/avatars/7d6e0fc4e348f1ef37793a2147406181.svg","isPro":true,"fullname":"Tianze Yang","user":"ytz009","type":"user"},{"_id":"68266b5261ed4d89177c3612","avatarUrl":"/avatars/e9d214c78916e3cfa554e545fd831446.svg","isPro":false,"fullname":"Kishan Panaganti","user":"kishanpb","type":"user"},{"_id":"67baa6c519e9dba50ece56b1","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/F3ZmRr7nCzfSVbpS4SY1E.png","isPro":false,"fullname":"Ninghao Liu","user":"NeoLiu43","type":"user"},{"_id":"6951c555b519522f565dfd0c","avatarUrl":"/avatars/9028d619483f359639ae7bfe4769da45.svg","isPro":false,"fullname":"ZhongzhiLi","user":"Zhongzhi1228","type":"user"},{"_id":"68e491dc0b7b68550d08acc9","avatarUrl":"/avatars/6c096137b58c727f1fe89a8ea21b4a51.svg","isPro":false,"fullname":"Huang Jingyuan","user":"JingyuanHuang","type":"user"},{"_id":"620783f24e28382272337ba4","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/620783f24e28382272337ba4/zkUveQPNiDfYjgGhuFErj.jpeg","isPro":false,"fullname":"GuoLiangTang","user":"Tommy930","type":"user"},{"_id":"63fac64d6b75d93aa13616e0","avatarUrl":"/avatars/573be0f4fe4a206700aa972629e79abf.svg","isPro":false,"fullname":"Jiaxi Li","user":"plusn","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"64cf499559503263d9da21df","name":"UniversityofGeorgia","fullname":"University of Georgia","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/64cf47db9cff7382036bf387/LuMBYKv7Iv4kfpJzaJvvE.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.01599.md"}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2606.01599 in a model README.md to link it from this page.

Cite arxiv.org/abs/2606.01599 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2606.01599 in a Space README.md to link it from this page.

Add this paper to a collection to link it from this page.