Meta tags:
Headings (most frequently used words):
agentic, of, machine, design, compositional, machines, construction, the, challenging, task, reasoning, gallery, tasks, performance, leaderboard, llm, agent, generated, results, from, rl, finetuned, llms, bibtex,
Text of the page (most frequently used words):
the (15), design (10), #machine (9), max (8), mean (8), compositional (7), agentic (6), std (6), 14b (5), #machines (4), and (4), liu (4), qwen2 (4), instruct (4), score (4), llms (4), gemini (4), pro (4), task (4), besiegefield (3), arxiv (3), catapult (3), models (3), with (3), agent (3), single (3), 2025 (2), author (2), zhang (2), wenqian (2), weiyang (2), zhen (2), cold (2), start (2), validity (2), ratio (2), car (2), from (2), generated (2), llama (2), scout (2), 17b (2), 16e (2), kimi (2), 0711 (2), preview (2), deepseek (2), claude (2), opus (2), 20250514 (2), doubao (2), seed (2), 250615 (2), qwen3 (2), coder (2), 480b (2), a35b (2), openai (2), hierarchical (2), iterative (2), editing (2), performance (2), moving (2), boulders (2), throwing (2), stones (2), reasoning (2), construction (2), mechanism (2), chinese (2), university (2), hong (2), kong (2), article, zhang2025besiegefield, title, journal, preprint, 2510, 14980, year, bibtex, reinforcement, learning, verifiable, rewards, rlvr, improves, qwen, model, results, finetuned, systems, different, llm, leaderboard, picking, objects, bottom, well, curved, track, delivering, through, ring, far, away, rough, terrains, gallery, tasks, example, cot, inspector, agents, blue, text, highlights, moderate, capability, spatial, imagined, physical, simulation, click, zoom, demonstrate, motion, human, designed, trebuchet, powerful, medieval, type, each, component, works, close, coordination, enabling, launch, projectile, much, farther, than, those, built, also, show, how, omitting, even, part, can, cause, entire, fail, highlighting, inherent, difficulty, 您的浏览器不支持视频播放, challenging, illustrated, our, environment, figure, shows, high, level, sketch, workflow, along, resulting, their, simulated, objective, create, that, throws, long, distances, equal, advising, corresponding, huggingface, demo, code, coldstart, dataset, paper, shenzhen, playground, for,
Text of the page (random words):
besiegefield playground for compositional machine design agentic design of compositional machines wenqian zhang 1 weiyang liu 2 zhen liu 1 1 the chinese university of hong kong shenzhen 2 the chinese university of hong kong arxiv paper coldstart dataset code huggingface demo equal advising corresponding author the task of compositional machine design is illustrated in our besiegefield environment the figure shows a high level sketch of the agentic workflow w gemini pro 2 5 along with the resulting machines and their simulated performance the design objective is to create a machine that throws boulders long distances the challenging task of compositional machine design 您的浏览器不支持视频播放 we demonstrate the motion mechanism of a human designed trebuchet a powerful medieval type of catapult each component works in close coordination enabling the machine to launch the projectile much farther than those built by llms we also show how omitting even a single part can cause the entire mechanism to fail highlighting the inherent difficulty of compositional machine design agentic machine construction agentic machine construction reasoning click to zoom example cot of inspector agents w gemini 2 5 pro blue text highlights the moderate capability of llms in spatial reasoning and imagined physical simulation gallery of tasks moving on rough terrains throwing stones far away throwing stones through a ring delivering boulders moving on a curved track picking objects from the bottom of a well performance leaderboard catapult task car task models single agent iterative editing hierarchical design mean max std mean max std mean max std gemini 2 5 pro 2 30 9 00 3 86 4 67 21 95 8 68 9 83 18 19 8 35 openai o3 2 87 5 22 1 96 9 14 14 01 3 71 2 00 11 11 3 98 qwen3 coder 480b a35b 1 75 9 24 3 17 5 10 12 02 5 54 3 90 6 52 2 54 doubao seed 1 6 250615 3 18 8 20 2 99 4 82 9 10 3 41 1 73 4 76 2 39 claude opus 4 20250514 1 19 4 82 2 21 1 18 4 91 2 18 2 27 9 32 4 22 deepseek v3 3 50 4 86 2 17 3 07 5 24 2 55 2 41 4 93 2 58 kimi k2 0711 preview 2 57 9 05 3 72 2 82 11 39 5 23 5 39 12 02 5 16 llama 4 scout 17b 16e 3 18 5 64 1 95 1 28 5 94 2 41 3 59 11 83 4 15 models single agent iterative editing hierarchical design mean max std mean max std mean max std gemini 2 5 pro 33 96 40 85 6 73 34 34 41 66 13 96 29 96 41 52 7 78 openai o3 15 28 32 08 8 97 14 34 35 08 11 79 28 39 36 18 11 01 qwen3 coder 480b a35b 8 87 11 50 4 46 15 24 28 95 13 12 12 59 34 05 10 78 doubao seed 1 6 250615 3 51 9 40 4 85 8 11 10 04 3 58 18 75 26 02 4 38 claude opus 4 20250514 9 83 12 98 1 28 8 07 28 04 12 48 14 56 38 67 20 69 deepseek v3 9 06 10 53 3 68 8 23 18 84 7 12 17 92 31 94 12 85 kimi k2 0711 preview 1 75 8 09 2 80 14 36 28 34 9 47 1 94 14 99 5 48 llama 4 scout 17b 16e 0 02 0 03 0 01 3 04 12 76 5 23 1 55 2 00 0 32 llm agent generated machines generated by agentic systems with different llms results from rl finetuned llms reinforcement learning with verifiable rewards rlvr improves qwen 2 5 14b model in compositional machine design models catapult car validity ratio mean score max score validity ratio mean score max score qwen2 5 14b instruct 11 50 0 06 2 41 46 50 4 97 19 10 qwen2 5 14b instruct cold start 9 50 0 11 5 54 40 50 4 67 20 23 qwen2 5 14b instruct rl 12 50 0 13 5 92 41 50 3 72 24 08 qwen2 5 14b instruct cold start rl 11 50 0 14 7 14 42 50 5 05 45 72 bibtex article zhang2025besiegefield title agentic design of compositional machines author zhang wenqian and liu weiyang and liu zhen journal arxiv preprint arxiv 2510 14980 year 2025 besiegefield 2025
|