{"id":1155011,"date":"2025-11-06T18:18:46","date_gmt":"2025-11-07T02:18:46","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-blog-post&p=1155011"},"modified":"2025-12-11T20:25:13","modified_gmt":"2025-12-12T04:25:13","slug":"phi-ground","status":"publish","type":"msr-blog-post","link":"https:\/\/www.microsoft.com\/en-us\/research\/articles\/phi-ground\/","title":{"rendered":"Phi-Ground\u6a21\u578b\uff1a\u8ba9AI\u5b66\u4f1a\u201c\u770b\u5c4f\u5e55\u201d"},"content":{"rendered":"\n

\u7f16\u8005\u6309\uff1a\u968f\u7740\u591a\u6a21\u6001\u548c\u63a8\u7406\u6a21\u578b\u7684\u5feb\u901f\u53d1\u5c55\uff0c\u80fd\u591f\u81ea\u4e3b\u7406\u89e3\u5e76\u64cd\u4f5c\u8ba1\u7b97\u673a\u754c\u9762\u7684\u667a\u80fd\u4f53\uff08Computer Use Agent, CUA\uff09\u6b63\u9010\u6e10\u6210\u4e3a\u73b0\u5b9e\u3002\u5176\u4e2d\uff0c\u56fe\u5f62\u754c\u9762\u5b9a\u4f4d\uff08GUI Grounding\uff09\u662f\u5b9e\u73b0\u8fd9\u4e00\u80fd\u529b\u7684\u6838\u5fc3\u73af\u8282\uff0c\u5b83\u51b3\u5b9a\u4e86\u667a\u80fd\u4f53\u80fd\u5426\u51c6\u786e\u5730\u5b8c\u6210\u70b9\u51fb\u3001\u8f93\u5165\u7b49\u5177\u4f53\u64cd\u4f5c\u3002\u7136\u800c\uff0c\u73b0\u6709\u6a21\u578b\u5728\u5173\u952e\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u7684\u51c6\u786e\u7387\u4ecd\u8f83\u4f4e\uff0c\u8ddd\u79bb\u5b9e\u9645\u5e94\u7528\u5c1a\u6709\u5dee\u8ddd\u3002\u5bf9\u6b64\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u8fd1\u671f\u53d1\u5e03\u4e86\u6280\u672f\u62a5\u544a\u7cfb\u7edf\u5206\u6790\u4e86 GUI Grounding \u6a21\u578b\u7684\u8bad\u7ec3\u8fc7\u7a0b\uff0c\u5e76\u63d0\u51fa\u4e86 Phi-Ground \u7cfb\u5217\u6a21\u578b\u3002\u5728\u4e94\u4e2a\u4e3b\u6d41\u57fa\u51c6\u4e0a\uff0cPhi-Ground \u6a21\u578b\u90fd\u53d6\u5f97\u4e86\u540c\u7c7b\u89c4\u6a21\u6a21\u578b\u7684\u6700\u4f73\u8868\u73b0\uff0c\u4e3a\u63a8\u52a8\u53ef\u7528\u667a\u80fd\u4f53\u7684\u53d1\u5c55\u63d0\u4f9b\u4e86\u575a\u5b9e\u7684\u57fa\u7840\u3002<\/p>\n\n\n\n


\n\n\n\n

\u5728\u5982\u4eca\u7684\u65e5\u5e38\u64cd\u4f5c\u4ee5\u53ca\u6570\u5b57\u5316\u529e\u516c\u7684\u8fc7\u7a0b\u4e2d\uff0c\u4eba\u4eec\u5e38\u5e38\u5e0c\u671b\u8ba1\u7b97\u673a\u4e0d\u4ec5\u4ec5\u662f\u88ab\u52a8\u6267\u884c\u547d\u4ee4\uff0c\u800c\u80fd\u50cf\u4eba\u7c7b\u52a9\u624b\u4e00\u6837\u4e3b\u52a8\u534f\u52a9\u6211\u4eec\u5b8c\u6210\u4efb\u52a1\u3002\u5728\u591a\u6a21\u6001 AI \u548c\u667a\u80fd\u4f53\u6280\u672f\u4e0d\u65ad\u589e\u5f3a\u7684\u80cc\u666f\u4e0b\uff0c\u8ba1\u7b97\u673a\u4f7f\u7528\u667a\u80fd\u4f53\uff08Computer Use Agent\uff0cCUA\uff09\u5e94\u8fd0\u800c\u751f\uff0c\u8ba9\u8fd9\u4e00\u613f\u666f\u9010\u6b65\u6210\u4e3a\u73b0\u5b9e\u3002<\/p>\n\n\n\n

\u5982\u679c\u8bf4\u5728\u73b0\u5b9e\u4e16\u754c\u4e2d\uff0c\u6211\u4eec\u671f\u5f85\u4eba\u5f62\u673a\u5668\u4eba\u80fd\u5e2e\u52a9\u6211\u4eec\u642c\u8fd0\u91cd\u7269\uff0c\u90a3\u4e48\u865a\u62df\u4e16\u754c\u4e2d\uff0c\u57fa\u4e8e\u56fe\u5f62\u754c\u9762\uff08GUI\uff09\u7684 CUA \u5219\u662f\u6211\u4eec\u5177\u6709\u201c\u52a8\u624b\u80fd\u529b\u201d\u7684\u6570\u5b57\u5206\u8eab\u3002\u8fd9\u4e9b CUA \u53ef\u4ee5\u76f4\u63a5\u4e0e\u64cd\u4f5c\u7cfb\u7edf\u4ea4\u4e92\uff0c\u4e0d\u4f9d\u8d56\u7279\u5b9a\u5e94\u7528\u7684 API\uff0c\u540c\u65f6\u8fd8\u4fbf\u4e8e\u4eba\u7c7b\u76d1\u7763\u3002\u8fd9\u5c31\u662f\u4eba\u5f62\u673a\u5668\u4eba\u548c CUA\uff0c\u76f8\u6bd4\u4e8e\u4f20\u7edf\u673a\u68b0\u81c2\u548c\u7f51\u9875\/\u5e94\u7528\u667a\u80fd\u4f53\u5177\u6709\u66f4\u5e7f\u9614\u4e0a\u9650\u7684\u539f\u56e0\u3002\u800c\u56fe\u5f62\u754c\u9762\u5b9a\u4f4d\uff08GUI Grounding\uff09\u5982\u540c CUA \u7684\u624b\u8db3\uff0c\u627f\u62c5\u7740\u76f4\u63a5\u4e0e\u73b0\u5b9e\u4ea4\u4e92\u3001\u5f71\u54cd\u7cfb\u7edf\u6210\u8d25\u7684\u8d23\u4efb\uff0c\u51b3\u5b9a\u4e86\u667a\u80fd\u4f53\u201c\u770b\u5f97\u61c2\u201d\u548c\u201c\u70b9\u5f97\u51c6\u201d\u7684\u80fd\u529b\u3002<\/p>\n\n\n\n

\"graphical
\u56fe1\uff1a\u667a\u80fd\u4f53\u5728\u7269\u7406\u4e16\u754c\u548c\u865a\u62df\u4e16\u754c\u7684\u53d1\u5c55<\/figcaption><\/figure>\n\n\n\n

\u5728\u6700\u65b0\u7684\u7814\u7a76\u4e2d\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u8003\u5bdf\u4e86 GUI Grounding \u4ece\u6570\u636e\u6536\u96c6\u5230\u6a21\u578b\u8bad\u7ec3\u7684\u5404\u4e2a\u7ec6\u8282\uff0c\u63d0\u51fa\u4e86 Phi-Ground \u7cfb\u5217\u6a21\u578b\uff0c\u80fd\u591f\u6839\u636e\u7528\u6237\u6307\u4ee4\u51c6\u786e\u5730\u5728\u7535\u8111\u5c4f\u5e55\u4e2d\u5b9a\u4f4d\u5230\u6240\u9700\u4f4d\u7f6e\u3002<\/p>\n\n\n\n

\u5728\u6280\u672f\u62a5\u544a Phi-Ground Tech Report: Advancing Perception in GUI Grounding \u4e2d\uff0c\u7814\u7a76\u5458\u4eec\u7cfb\u7edf\u6027\u5730\u4ecb\u7ecd\u4e86\u8fd9\u4e00\u6a21\u578b\u7684\u7b97\u6cd5\u8bbe\u8ba1\u4e0e\u5de5\u7a0b\u5b9e\u73b0\u3002<\/p>\n\n\n\n

Phi-Ground Tech Report: Advancing Perception in GUI Grounding<\/em><\/strong><\/p>\n\n\n\n