
{"id":8899,"date":"2026-06-07T02:26:38","date_gmt":"2026-06-06T18:26:38","guid":{"rendered":"https:\/\/infernews.com\/blog\/a-vision-language-action-model-empowering-action-generation-through-affordance-a\/"},"modified":"2026-06-07T02:26:38","modified_gmt":"2026-06-06T18:26:38","slug":"a-vision-language-action-model-empowering-action-generation-through-affordance-a","status":"publish","type":"post","link":"https:\/\/infernews.com\/blog\/a-vision-language-action-model-empowering-action-generation-through-affordance-a\/","title":{"rendered":"AffordanceVLA\uff1a\u70ba\u6a5f\u68b0\u81c2\u52a0\u5165\u300c\u5148\u60f3\u518d\u505a\u300d\u7684\u80fd\u529b"},"content":{"rendered":"<figure class=\"wp-block-image size-large\"><img decoding=\"async\" src=\"https:\/\/infernews.com\/blog\/wp-content\/uploads\/2026\/06\/pasted-46cbaed0cf7a.jpg\" alt=\"AffordanceVLA overview\"><\/figure>\n<p>\u6a5f\u68b0\u81c2\u807d\u5230\u300c\u628a\u676f\u5b50\u62ff\u8d77\u4f86\u300d\u9019\u985e\u6307\u4ee4\u6642\uff0c\u50b3\u7d71\u7684\u8996\u89ba\u8a9e\u8a00\u52d5\u4f5c\u6a21\u578b\uff08Vision-Language-Action Model, VLA\uff09\u5f80\u5f80\u8981\u76f4\u63a5\u628a\u5f71\u50cf\u548c\u6587\u5b57\u7ffb\u8b6f\u6210\u95dc\u7bc0\u89d2\u5ea6\uff0c\u4e2d\u9593\u6b20\u7f3a\u4e00\u500b\u300c\u601d\u8003\u300d\u6b65\u9a5f\u3002AffordanceVLA \u5617\u8a66\u5728\u9019\u500b\u9d3b\u6e9d\u4e0a\u67b6\u4e00\u9053\u6a4b\uff1a\u5148\u8b93\u6a21\u578b\u9810\u6e2c\u7d50\u69cb\u5316\u7684\u53ef\u4f9b\u6027\uff08affordance\uff09\uff0c\u518d\u64da\u6b64\u751f\u6210\u52d5\u4f5c\u3002<\/p>\n<p>\u6574\u500b\u6846\u67b6\u7531\u4e09\u500b\u5c08\u5bb6\u6a21\u578b\u7d44\u6210\uff0c\u6309\u55ae\u5411\u7684 UAA \u6ce8\u610f\u529b\u4e32\u63a5\u3002Understanding Expert\uff08M_und\uff09\u4ee5 PaliGemma\uff08SigLIP + Gemma\uff09\u70ba\u9aa8\u5e79\uff0c\u628a\u756b\u9762\u3001\u6307\u4ee4\u8207\u6a5f\u68b0\u81c2\u81ea\u8eab\u72c0\u614b\u878d\u5408\u6210\u7d71\u4e00\u7684\u8a9e\u610f\u8868\u793a\u3002Affordance Generation Expert\uff08M_gen\uff09\u4ee5 Gemma \u642d\u914d\u53ef\u5b78\u7fd2\u67e5\u8a62\uff0c\u628a\u4e0a\u8ff0\u8868\u793a\u89e3\u78bc\u70ba\u4e09\u7a2e\u53ef\u4f9b\u6027\u6a19\u8a18\uff1aWhich2Act \u5224\u65b7\u8981\u64cd\u4f5c\u7684\u7269\u4ef6\u3001Where2Act \u6a19\u51fa\u4e8c\u7dad\u4e92\u52d5\u71b1\u5340\u3001How2Act \u5247\u63a8\u4f30\u4e09\u7dad\u5e7e\u4f55\u8cc7\u8a0a\u3002\u6700\u5f8c Action Expert\uff08M_act\uff09\u4ee5 flow matching \u65b9\u5f0f\u8f38\u51fa\u6574\u6bb5\u52d5\u4f5c\u5e8f\u5217\uff08action chunk\uff09\u3002\u9019\u500b\u8a2d\u8a08\u547c\u61c9\u4e86\u8ad6\u6587\u300cAffordances serve as a perfect bridge\u300d\u7684\u6838\u5fc3\u60f3\u6cd5\u3002<\/p>\n<p>\u7531\u65bc\u73fe\u6210\u6a5f\u5668\u4eba\u6578\u64da\u96c6\u4e2d\u7f3a\u4e4f\u5bc6\u96c6\u7684\u53ef\u4f9b\u6027\u6a19\u8a3b\uff0c\u4f5c\u8005\u984d\u5916\u63d0\u4f9b\u4e86\u4e00\u689d\u81ea\u52d5\u5316\u6a19\u8a3b\u7ba1\u7dda\uff0c\u4e26\u4ee5\u4e09\u968e\u6bb5\u6f38\u9032\u5f0f\u8ab2\u7a0b\u8a13\u7df4 MoT \u67b6\u69cb\u3002\u8a13\u7df4\u6642\u9808\u7559\u610f model.chunk_size \u8207 data.chunk_size \u4fdd\u6301\u4e00\u81f4\uff0c\u5426\u5247\u52d5\u4f5c attention mask \u6703\u932f\u4f4d\uff1bWhich2Act \u7684 Flux loss \u9810\u8a2d\u70ba MSE\uff0c\u53ef\u5728 <code>src\/models\/which2act_decoder.py<\/code> \u9802\u7aef\u5207\u63db\u3002<\/p>\n<p>\u9019\u500b\u9805\u76ee\u9069\u5408\u7814\u7a76 VLA\u3001\u6a5f\u5668\u4eba\u64cd\u4f5c\u7b56\u7565\uff0c\u6216\u5c0d\u53ef\u4f9b\u6027\u8868\u793a\u6709\u8208\u8da3\u7684\u958b\u767c\u8005\u3002\u9700\u8981\u4e00\u652f\u80fd\u8dd1 PaliGemma \u8207 flow matching \u7684 GPU \u74b0\u5883\uff0c\u4e26\u6e96\u5099\u597d\u4eff\u771f\u6216\u5be6\u6a5f\u8a55\u6e2c\u6d41\u7a0b\u3002\u8ad6\u6587\u4e2d\u7684\u6a21\u64ec\u8207\u771f\u6a5f\u5be6\u9a57\u6a6b\u8de8\u591a\u7a2e\u64cd\u4f5c\u5834\u666f\uff0c\u986f\u793a\u52a0\u5165\u53ef\u4f9b\u6027\u4e2d\u9593\u5c64\u80fd\u63d0\u5347\u6cdb\u5316\u8207\u7cbe\u78ba\u5ea6\uff0c\u4f46\u5177\u9ad4\u6578\u5b57\u4ecd\u4ee5\u5b98\u65b9\u5831\u544a\u70ba\u6e96\u3002<\/p>\n<p><strong>\u91cd\u9ede\u6458\u8981<\/strong><\/p>\n<ul>\n<li>\u4ee5 Which2Act\u3001Where2Act\u3001How2Act \u4e09\u6bb5\u5f0f\u53ef\u4f9b\u6027\u4f5c\u70ba\u8996\u89ba\u3001\u8a9e\u8a00\u8207\u52d5\u4f5c\u4e4b\u9593\u7684\u4e2d\u9593\u6a4b\u6a11\u3002<\/li>\n<li>\u63a1\u7528 MoT \u67b6\u69cb\uff0c\u4e32\u63a5 PaliGemma \u70ba\u57fa\u790e\u7684 Understanding\u3001Gemma \u70ba\u57fa\u790e\u7684 Affordance \u8207 Action Expert\u3002<\/li>\n<li>\u914d\u5957\u81ea\u52d5\u5316\u53ef\u4f9b\u6027\u6a19\u8a3b\u7ba1\u7dda\uff0c\u7de9\u89e3\u6a5f\u5668\u4eba\u6578\u64da\u6a19\u8a3b\u4e0d\u8db3\u7684\u554f\u984c\u3002<\/li>\n<li>\u8a13\u7df4\u63a1\u4e09\u968e\u6bb5\u6f38\u9032\u5f0f\u7b56\u7565\uff0cflow matching \u8f38\u51fa\u6574\u6bb5\u52d5\u4f5c\u5e8f\u5217\u3002<\/li>\n<li>\u6a21\u578b\u3001\u8a13\u7df4\u8207\u6a19\u8a3b\u8173\u672c\u7686\u96a8\u958b\u6e90\u9805\u76ee\u91cb\u51fa\uff0c\u6b61\u8fce\u793e\u7fa4\u5ef6\u4f38\u3002<\/li>\n<\/ul>\n<p><strong>GitHub\uff1a<\/strong> <a href=\"https:\/\/github.com\/Skywalker-yqz\/AffordanceVLA\" rel=\"noopener noreferrer\" target=\"_blank\">https:\/\/github.com\/Skywalker-yqz\/AffordanceVLA<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u9019\u9805\u76ee\u628a\u8996\u89ba\u3001\u8a9e\u8a00\u8207\u52d5\u4f5c\u4e09\u8005\u4e4b\u9593\u7684\u843d\u5dee\uff0c\u900f\u904e\u4e09\u6bb5\u5f0f\u53ef\u4f9b\u6027\u9810\u6e2c\u4e32\u8d77\u4f86\uff0c\u8b93\u6a5f\u68b0\u81c2\u5728\u57f7\u884c\u6307\u4ee4\u524d\u5148\u7406\u89e3\u8a72\u78b0\u54ea\u88e1\u3001\u600e\u9ebc\u78b0\u3002<\/p>\n","protected":false},"author":8,"featured_media":8898,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"ai_generated_summary":"","footnotes":""},"categories":[133,164,165,119,76,127,195],"tags":[],"class_list":["post-8899","post","type-post","status-publish","format-standard","hentry","category-133","category-164","category-165","category-119","category-76","category-127","category-195"],"_links":{"self":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/8899","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/comments?post=8899"}],"version-history":[{"count":0,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/posts\/8899\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media\/8898"}],"wp:attachment":[{"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/media?parent=8899"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/categories?post=8899"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/infernews.com\/blog\/wp-json\/wp\/v2\/tags?post=8899"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}