½ñÈÕî¿Ïµ²¿·ÖÅû¶ÖØ´óÏ£Íû·Æ×Üͳ½ôÆÈ×÷·Ï¶à¸ö¹ûÕæÐÐ³Ì ºÜÐË·ÜΪÄú½â´ðÕâ¸öÎÊÌ⣬£¬£¬ÈÃÎÒÀ´°ïÄúÏêϸ˵Ã÷һϡ£¡£¡£¡£¡£Æ·ÅÆÊÚȨ±¨Ð޵绰£¬£¬£¬¿ìËÙÉÏÃÅ·þÎñ

ËÄ´¨Ê¡¹ã°²ÊÐÔÀ³ØÏؽ­Î÷Ê¡ÒË´ºÊÐÒË·áÏر±º£Êк£³ÇÇø¹ÌÔ­ÊÐÎ÷¼ªÏØÏòÑôÇø¶¹¸÷ׯµØÇø¹ã¶«Ê¡½­ÃÅÊÐËÄ´¨Ê¡µÂÑôÊй㺺ÊÐÎäÇåÇøãâ¹Á¸ÛÕòÎ÷³ÇÇøÔÂ̳½ÖµÀ½­Î÷ʡӥ̶ÊÐÓ཭Çøɽ¶«Ê¡¶«ÓªÊпÑÀûÇø¸£½¨Ê¡ÄÏƽÊÐËÄ´¨Ê¡µÂÑôÊй㺺ÊÐɽÎ÷Ê¡ÂÀÁºÊÐÖÐÑôÏØÏòÑôÇøСºìÃŵØÇøËÄ´¨Ê¡¸Ê×βØ×å×ÔÖÎÖÝãò¶¨ÏØÄÏÑôÊÐÄÚÏçÏØËÄ´¨Ê¡Ò˱öÊÐçîÏØɽ¶«Ê¡µÂÖÝÊÐÆëºÓÏØËÄ´¨Ê¡Á¹É½ÒÍ×å×ÔÖÎÖÝÎ÷²ýÊÐËÄ´¨Ê¡¸Ê×βØ×å×ÔÖÎÖݾÅÁúÏØ·ą́Çø³¤ÐÁµêÕò²ýƽÇøСÌÀɽÕòº£ÄÏÊ¡ÈýɳÊÐÎ÷ɳÇøÑÓÇìÇø¿µ×¯Õòºþ±±Ê¡ÏåÑôÊÐÏåÖÝÇøº£ÄÏÊ¡º£¿£¿£¿£¿£¿ÚÊÐÃÀÀ¼ÇøÎ÷²ØɽÄÏÊй±¸ÂÏØÏòÑôÇøË«¾®½ÖµÀ¹ã¶«Ê¡½­ÃÅÊб±³½Çø½­ËÕÊ¡ÐìÖÝÊÐÐÂÒÊÊÐÉÂÎ÷Ê¡±¦¼¦ÊÐǧÑôÏزýƽÇø»ØÁú¹Û½ÖµÀ¹ÌÔ­ÊÐÎ÷¼ªÏØ»³ÈáÇøÑãÆܵØÇøÁÉÄþÊ¡ÌúÁëÊÐÇåºÓÇøÎߺþÊÐÄÏÁêÏØÔÞ»ÊÏØÎ÷ÁúÃÅÏç°¢¿ËËÕµØÇøкÍÏØÖØÇìÊÐÏØÎ×ɽÏس¤°²Çø¹ã°²½ÖµÀ¹ȪÇø˼ÒׯÕò²©¶ûËþÀ­ÃɹÅ×ÔÖÎÖÝÎÂȪÏØÏòÑôÇøСºìÃŵØÇø½­Î÷Ê¡¸ÓÖÝÊÐʯ³ÇÏØÄÚÃɹÅÐË°²ÃËÎÚÀ¼ºÆÌØÊкÓÎ÷ÇøÌÒÔ°½ÖµÀËÄ´¨Ê¡¸Ê×βØ×å×ÔÖÎÖÝãò¶¨ÏØɽÎ÷Ê¡ÐÃÖÝÊж¨ÏåÏؽ­ËÕÊ¡ÄϾ©ÊÐÆÖ¿ÚÇøɽ¶«Ê¡¶«ÓªÊпÑÀûÇøºþ±±Ê¡Ð¢¸ÐÊкº´¨ÊÐÉÂÎ÷Ê¡ººÖÐÊÐÄÏÖ£Çø½­Î÷Ê¡ÒË´ºÊÐÒË·áÏØÔÆÄÏÊ¡ºìºÓ¹þÄá×åÒÍ×å×ÔÖÎÖÝÂÌ´ºÏØÇàÐãÇø½­Î÷Ê¡¾Å½­ÊÐÎäÄþÏظ£½¨Ê¡ÆÎÌïÊгÇÏáÇø´óÐËÇø°¢¿ËËÕµØÇøкÍÏغÚÁú½­Ê¡¹þ¶û±õÊаÍÑåÏØÏòÑôÇø¶¹¸÷ׯµØÇø½­ËÕÊ¡ÐìÖÝÊÐÐÂÒÊÊкÚÁú½­Ê¡´óÐË°²ÁëµØÇøºôÂêÏز©¶ûËþÀ­ÃɹÅ×ÔÖÎÖÝÎÂȪÏØ»³ÈáÇø±¦É½Õò¹óÖÝÊ¡Í­ÈÊÊÐÓñÆÁ¶±×å×ÔÖÎÏØ·À³Ç¸ÛÊж«ÐËÊкӶ«Çø´óÖ±¹Á½ÖµÀ±±³½Çø¸ÊËàÊ¡ÇìÑôÊÐÕòÔ­ÏØÎ÷²Ø°¢ÀïµØÇøÈÕÍÁÏعóÖÝÊ¡Ç­¶«ÄÏÃç×嶱×å×ÔÖÎÖÝÌìÖùÏØËÄ´¨Ê¡Á¹É½ÒÍ×å×ÔÖÎÖÝÎ÷²ýÊÐÇÅÎ÷ÇøÁôÓª½ÖµÀ¹óÖÝÊ¡Ç­Äϲ¼ÒÀ×åÃç×å×ÔÖÎÖÝ»ÝË®ÏØÎäÇåÇøÉÏÂį́Õòɽ¶«Ê¡Ì©°²ÊÐ̩ɽÇøÇàÐãÇø¼»ÖÝÇø¶«¶þÓªÕò¹óÖÝÊ¡Í­ÈÊÊÐÓñÆÁ¶±×å×ÔÖÎÏØ¿¦Ê²µØÇøÒ¶³ÇÏØÃÜÔÆÇø²»ÀÏÍÍÕò±±º£Êк£³ÇÇøËÄ´¨Ê¡µÂÑôÊй㺺ÊÐËÄ´¨Ê¡³É¶¼»á½ðÅ£ÇøÔªÊÏÏØÉÂÎ÷Ê¡ÏÌÑôÊÐÐËƽÊÐËÄ´¨Ê¡³É¶¼»áж¼Çø½­ËÕÊ¡ÐìÖÝÊÐÐÂÒÊÊи£½¨Ê¡ÈýÃ÷ÊÐ÷ÁÐÇøÎ÷ÇàÇø¾«ÎäÕòºÍÌïµØÇøÔÞ»ÊÏØԺͷÕòÃÜÔÆÇøºÓÄÏÕ¯ÕòËÄ´¨Ê¡Á¹É½ÒÍ×å×ÔÖÎÖÝÕѾõÏØÎ÷ÇàÇø¾«ÎäÕòËÄ´¨Ê¡Ò˱öÊдäÆÁÇø³ç×óÊÐ

¿ËÈÕÐÐÒµ±¨¸æÐû²¼ÐÂЧ¹û±¾ÖÜ AI ÏîÄ¿ÍƼö£ºWorldModelBench£¬£¬£¬WorldScore£¬£¬£¬WorldLens¡­ÌìÏÂÄ£×Ó¾®Å磬£¬£¬ÐèҪеÄBenchmark£¬£¬£¬ºÜÐË·ÜΪÄú½â´ðÕâ¸öÎÊÌ⣬£¬£¬ÈÃÎÒÀ´°ïÄúÏêϸ˵Ã÷Ò»ÏÂ:¹Ù·½·þÎñרÏߣ¬£¬£¬Ö§³Ö¶àÆ·ÅƱ¨ÐÞ

ÐÅÑôÊÐƽÇÅÇøÇå¾²ÇøÄÏÊнֵÀºþ±±Ê¡Ð¢¸ÐÊкº´¨Êй㶫ʡÉÇÍ·ÊÐÁúºþÇø¹óÖÝÊ¡Ç­Äϲ¼ÒÀ×åÃç×å×ÔÖÎÖÝ»ÝË®ÏØ°¢¿ËËÕµØÇø°Ý³ÇÏظÊËàÊ¡¸ÊÄϲØ×å×ÔÖÎÖÝËÄ´¨Ê¡Á¹É½ÒÍ×å×ÔÖÎÖݸÊÂåÏØÁÉÄþÊ¡ÉòÑôÊлëÄÏÇøÉÂÎ÷Ê¡ÉÌÂåÊÐɽÑôÏØÇå¾²ÇøÄÏÊнֵÀºÚÁú½­Ê¡¼Ñľ˹Êи»½õÊÐÐÅÑôÊÐƽÇÅÇøÎ÷²ØɽÄÏÊÐÄÚÃɹÅÎýÁÖ¹ùÀÕÃËÕýÏâ°×ÆìÇຣʡº£ÄϲØ×å×ÔÖÎÖݹóµÂÏظ£½¨Ê¡¸£ÖÝÊÐÂÞÔ´ÏØÁÉÄþÊ¡ÌúÁëÊпªÔ­Êй㶫ʡÔƸ¡ÊÐÐÂÐËÏØÄÚÃɹÅÎÚº£ÊÐÎÚ´ïÇøɽ¶«Ê¡ÑĮ̀ÊÐIJƽÇøºþ±±Ê¡Ð¢¸ÐÊкº´¨ÊÐÎ÷²ØɽÄÏÊй±¸ÂÏØÑÓÇìÇø¿µ×¯ÕòÎ÷³ÇÇøÌìÇŽֵÀ¡°²ÏØ·À³Ç¸ÛÊж«ÐËÊй㶫ʡÉÇÍ·ÊÐÁúºþÇøºþ±±Ê¡Ò˲ýÊÐïö¹éÏؽ­ËÕÊ¡ÄÏͨÊÐÆô¶«ÊкÚÁú½­Ê¡¼Ñľ˹ÊÐÌÀÔ­ÏغÚÁú½­Ê¡º×¸ÚÊи£½¨Ê¡ÕÄÖÝÊÐÁúº£ÊÐÁÉÄþÊ¡ÉòÑôÊлëÄÏÇø¸£½¨Ê¡ÄÏƽÊÐËÄ´¨Ê¡¹ãÔªÊÐÇÕÖÝÊкÚÁú½­Ê¡¹þ¶û±õÊаÍÑåÏØÇຣʡº£Î÷ÃɹÅ×å²Ø×å×ÔÖÎÖݵÂÁî¹þÊи£½¨Ê¡ÆÎÌïÊгÇÏáÇøɽÎ÷Ê¡Ô˶¼»áÏÄÏØ°ÍÒô¹ùÀãÃɹÅ×ÔÖÎÖݺÍ˶ÏظÊËàÊ¡°×ÒøÊо°Ì©ÏØÒÁÀç¹þÈø¿Ë×ÔÖÎÖÝÕÑËÕÏØɽ¶«Ê¡µÂÖÝÊÐÆëºÓÏØ·ą́Çø³¤ÐÁµêÕò¸ÊËàÊ¡°×ÒøÊо°Ì©ÏØÎÚ³ľÆëÊÐɳÒÀ°Í¿ËÇøÁÉÄþÊ¡ÉòÑôÊлëÄÏÇøËÄ´¨Ê¡¸Ê×βØ×å×ÔÖÎÖݾÅÁúÏØÐÐÌÆÏØÁúÖÝÕòÃÜÔÆÇø¹Å±±¿ÚÕò¹ã¶«Ê¡ÉîÛÚÊÐËÄ´¨Ê¡¸Ê×βØ×å×ÔÖÎÖݵ¤°ÍÏØÏòÑôÇøС¹Ø½ÖµÀÎߺþÊÐÄÏÁêÏØɽÎ÷Ê¡Ë·ÖÝÊÐɽÒõÏؽ­Î÷Ê¡¼ª°²ÊÐÓÀÐÂÏØËÄ´¨Ê¡¸Ê×βØ×å×ÔÖÎÖݵ¤°ÍÏØÂåÑôÊÐÈêÑôÏØÔÆÄÏÊ¡ÓñϪÊÐÐÂƽÒÍ×å´ö×å×ÔÖÎÏظÊËàÊ¡°×ÒøÊо°Ì©Ïؾ²º£Çø´óÇñׯÕò¹ã¶«Ê¡½­ÃÅÊиÊËàʡ¤ÄÏÊÐÉÂÎ÷Ê¡Î÷°²ÊÐδÑëÇø»³ÈáÇøÑãÆܵØÇøÇÕÖÝÊÐËÄ´¨Ê¡¹ãÔªÊÐÉϺ£ÊÐÊÐϽÇø¼Î¶¨ÇøƽɽÏØÕ¬±±Ïç°¢¿ËËÕµØÇø°Ý³ÇÏØɽÎ÷Ê¡½ú¶¼»áÔóÖÝÏع㶫ʡÔƸ¡ÊÐÐÂÐËÏØÐí²ýÊн¨°²Çø½­ËÕÊ¡Õò½­ÊоäÈÝÊй㶫ʡÉÇÍ·ÊÐÄÏ°ÄÏزýƽÇøÑÓÊÙÕòºÚÁú½­Ê¡¼Ñľ˹ÊÐÌÀÔ­ÏØËÄ´¨Ê¡Á¹É½ÒÍ×å×ÔÖÎÖݸÊÂåÏغþ±±Ê¡¶÷Ê©ÍÁ¼Ò×åÃç×å×ÔÖÎÖݶ÷Ê©Êо®ÚêÏØÎâ¼ÒÒ¤ÏçÏòÑôÇøСºìÃŵØÇøÐÅÑôÊÐƽÇÅÇøÔÆÄÏÊ¡ÎÄɽ׳×åÃç×å×ÔÖÎÖݹãÄÏÏؽ­Î÷Ê¡¼ª°²Êмª°²ÏØËÄ´¨Ê¡ÀÖɽÊо®ÚêÏØÎâ¼ÒÒ¤Ïç½­ËÕÊ¡ÐìÖÝÊкþ±±Ê¡Ð¢¸ÐÊкº´¨Êй㶫ʡ»ÝÖÝÊÐÁúÃÅÏØ¿¦Ê²µØÇøÂó¸ÇÌáÏزýƽÇøÑÓÊÙÕòºÚÁú½­Ê¡Æß̨ºÓÊÐÌÒɽÇøºÓ¶«Çø¶«Ð½ֵÀÇÅÎ÷ÇøÁôÓª½ÖµÀ¸£½¨Ê¡ÆÎÌïÊÐÏÉÓÎÏغÚÁú½­Ê¡¹þ¶û±õÊаÍÑåÏØɽ¶«Ê¡×Ͳ©ÊÐÁÙ×ÍÇøºÏ·ÊÊÐÑþº£Çø

È«Çò·þÎñÇøÓò:½­Î÷Ê¡¼ª°²Êмª°²ÏØÇຣʡ¹ûÂå²Ø×å×ÔÖÎÖÝÁÉÄþÊ¡ÌúÁëÊпªÔ­ÊÐÄÏÑôÊÐÄÚÏçÏغ£ÄÏÊ¡º£¿£¿£¿£¿£¿ÚÊÐÃÀÀ¼Çø½­Î÷Ê¡ÒË´ºÊзîÐÂÏظ£½¨Ê¡¸£ÖÝÊÐÂÞÔ´ÏØɽ¶«Ê¡µÂÖÝÊÐÆëºÓÏغþ±±Ê¡Ò˲ýÊÐÒÄÁêÇøÏòÑôÇøСºìÃŵØÇø¸ÊËàʡ¤ÄÏÊÐÎ䶼ÇøºÚÁú½­Ê¡¼Ñľ˹ÊÐÌÀÔ­ÏØ°ö²ºÊÐÁú×ÓºþÇø±±º£Êк£³ÇÇø½­Î÷Ê¡ÒË´ºÊÐÒË·áÏظ£½¨Ê¡ÈýÃ÷ÊÐ÷ÁÐÇø¹óÖÝÊ¡Ç­¶«ÄÏÃç×嶱×å×ÔÖÎÖÝÌìÖùÏØ¿¦Ê²µØÇøÂó¸ÇÌáÏØÏòÑôÇøС¹Ø½ÖµÀÄÏ¿ªÇøÌåÓýÖÐÐĽֵÀ¾²º£ÇøÎ÷µÔׯÕòƽɽÏض«»ØÉáÕò½­Î÷Ê¡¾°µÂÕòÊвý½­Çø½­Î÷Ê¡¾°µÂÕòÊвý½­ÇøÕÔÏØɳºÓµêÕòÇÅÎ÷Çø¶«»ª½ÖµÀƽɽÏØС¾õÕòºÓ¶«Çø¶«Ð½ֵÀɽÎ÷Ê¡Ë·ÖÝÊÐɽÒõÏØ»³ÈáÇøÁúɽ½ÖµÀ½­ËÕÊ¡ÐìÖÝÊзáÏØËÄ´¨Ê¡Ò˱öÊдäÆÁÇøɽ¶«Ê¡ÁĶ¼»á¹ÚÏØÎߺþÊÐÄÏÁêÏØÎ÷²ØɽÄÏÊй±¸ÂÏØÎ÷ÇàÇø¾«ÎäÕòÇàÐãÇøËÄ´¨Ê¡ãòÖÝÊн­ÑôÇøÔÆÄÏÊ¡´óÀí°××å×ÔÖÎÖÝÔÆÁúÏØ´óÐËÇøÉÂÎ÷Ê¡±¦¼¦ÊÐǧÑôÏØÇຣʡº£Î÷ÃɹÅ×å²Ø×å×ÔÖÎÖݵÂÁî¹þÊк£ÄÏÊ¡ÈýɳÊÐÎ÷ɳÇøÁÉÄþÊ¡½õÖÝÊÐÒåÏØÔÞ»ÊÏØԺͷÕòÄÚÃɹÅÐË°²ÃËÎÚÀ¼ºÆÌØÊÐÂí°°É½Êв©ÍûÇøɽ¶«Ê¡ÁĶ¼»áÜÝƽÇøºÚÁú½­Ê¡ÒÁ´ºÊнðÁÖÇø±¦ÛæÇøÅ£¼ÒÅÆÕò¿ª·âÊÐͨÐíÏغ£ÄÏÊ¡º£¿£¿£¿£¿£¿ÚÊÐÃÀÀ¼Çø¹ȪÇø°×¹ȪÏçÎ÷²ØɽÄÏÊм»ÖÝÇø¹ÙׯÕòÕÔÏØɳºÓµêÕò¹ðÁÖÊÐÐã·åÇø¹óÖÝÊ¡Ç­¶«ÄÏÃç×嶱×å×ÔÖÎÖÝÌìÖùÏØÔªÊÏÏØÒó´åÕò½­Î÷ʡӥ̶ÊÐÓ཭ÇøËÄ´¨Ê¡³É¶¼»á½ðÅ£ÇøÁÉÄþÊ¡ÉòÑôÊÐÉòºÓÇøƽ¶¥É½ÊÐÕ¿ºÓÇø³ç×óÊÐɽÎ÷Ê¡ÂÀÁºÊÐÖÐÑôÏØÉÂÎ÷Ê¡±¦¼¦ÊÐǧÑôÏظ£½¨Ê¡ÄÏƽÊн¨ê±ÊкϷÊÊЮÑôÇøËÄ´¨Ê¡¹ã°²ÊÐÔÀ³ØÏØÇÕÖÝÊйȪÇø˼ÒׯÕòÄÚÃɹÅÎÚº£ÊÐÎÚ´ïÇøÄÏÑôÊÐÄÚÏçÏØÔÆÄÏÊ¡Àö½­ÊÐÄþÝõÒÍ×å×ÔÖÎÏظ£½¨Ê¡ÏÃÃÅÊкþÀïÇø¸£½¨Ê¡¸£ÖÝÊÐÓÀÌ©ÏØÁÉÄþÊ¡ÉòÑôÊлëÄÏÇøÔÆÄÏÊ¡ÓñϪÊÐÐÂƽÒÍ×å´ö×å×ÔÖÎÏØƽɽÏØС¾õÕòºÚÁú½­Ê¡Ë绯ÊÐÃ÷Ë®ÏØɽ¶«Ê¡Ì©°²Êж«Æ½ÏØ·ą́ÇøÓÒ°²ÃŽֵÀÄÏÑôÊÐÄÚÏçÏØÒø´¨ÊкØÀ¼ÏØÇÅÎ÷ÇøÁôÓª½ÖµÀÃÜÔÆÇø²»ÀÏÍÍÕòºþ±±Ê¡Ò˲ýÊÐïö¹éÏØÏòÑôÇøСºìÃŵØÇøÁÉÄþÊ¡ÌúÁëÊпªÔ­ÊÐÇÕÖÝÊкþ±±Ê¡¶÷Ê©ÍÁ¼Ò×åÃç×å×ÔÖÎÖݽ¨Ê¼ÏØÔÆÄÏÊ¡ÓñϪÊÐÐÂƽÒÍ×å´ö×å×ÔÖÎÏØÕÔÏØɳºÓµêÕòɽ¶«Ê¡Íþº£ÊвýƽÇø»ØÁú¹Û½ÖµÀÁÉÄþÊ¡±¾ÏªÊÐϪºþÇøËÄ´¨Ê¡Ò˱öÊÐçîÏغþ±±Ê¡¶÷Ê©ÍÁ¼Ò×åÃç×å×ÔÖÎÖݶ÷Ê©ÊÐƽ¶¥É½ÊÐÕ¿ºÓÇøºþ±±Ê¡¶÷Ê©ÍÁ¼Ò×åÃç×å×ÔÖÎÖݶ÷Ê©ÊÐ

±¾ÔÂî¿Ïµ²¿·Ö¹ûÕæ×îж¯Ì¬±¾ÖÜ AI ÏîÄ¿ÍƼö£ºWorldModelBench£¬£¬£¬WorldScore£¬£¬£¬WorldLens¡­ÌìÏÂÄ£×Ó¾®Å磬£¬£¬ÐèҪеÄBenchmark£¬£¬£¬ºÜÐË·ÜΪÄú½â´ðÕâ¸öÎÊÌ⣬£¬£¬ÈÃÎÒÀ´°ïÄúÏêϸ˵Ã÷Ò»ÏÂ:ÊÛºó·þÎñάÐÞÖÐÐĵ绰£¬£¬£¬Ö§³Ö¶àÇþµÀ·þÎñ

ÌìÏ·þÎñÇøÓò£º½­ËÕÊ¡ËÕÖÝÊÐÏà³ÇÇøÉÂÎ÷Ê¡ººÖÐÊÐÎ÷ÏçÏؽ­ËÕÊ¡ÄϾ©ÊÐÆÖ¿ÚÇøÉÂÎ÷Ê¡ÏÌÑôÊÐÐËƽÊвýƽÇøÑÓÊÙÕòɽ¶«Ê¡×Ͳ©ÊÐÁÙ×ÍÇøɽÎ÷Ê¡½úÖÐÊÐÌ«¹ÈÇøÁÉÄþÊ¡ÁÉÑôÊа×ËþÇø·À³Ç¸ÛÊж«ÐËÊкþ±±Ê¡Ò˲ýÊÐÒ˶¼»áËÄ´¨Ê¡Ò˱öÊдäÆÁÇøºÚÁú½­Ê¡º×¸ÚÊн­ËÕÊ¡ÐìÖÝÊÐÐÂÒÊÊиÊËàÊ¡°×ÒøÊо°Ì©ÏØɽÎ÷Ê¡Ô˶¼»áÏÄÏسç×óÊÐÄþÃ÷ÏØÉÂÎ÷Ê¡Î÷°²ÊÐδÑëÇø¿¦Ê²µØÇøÂó¸ÇÌáÏس¤°²Çø¹ã°²½ÖµÀÑÓÇìÇø¿µ×¯ÕòºÚÁú½­Ê¡º×¸ÚÊÐÃÜÔÆÇøºÓÄÏÕ¯Õòº£ÄÏÊ¡º£¿£¿£¿£¿£¿ÚÊÐÃÀÀ¼ÇøËÄ´¨Ê¡¹ã°²ÊÐÔÀ³ØÏØɽ¶«Ê¡ÔæׯÊÐ̨¶ùׯÇø±¦ÛæÇøÅ£¼ÒÅÆÕò¹ã¶«Ê¡ÉÇÍ·ÊÐÄÏ°ÄÏØÁÉÄþÊ¡½õÖÝÊÐÒåÏؽ­ËÕÊ¡Á¬ÔƸÛÊиÓÓÜÇøºÓ¶«Çø´óÖ±¹Á½ÖµÀÉÂÎ÷Ê¡±¦¼¦ÊÐüÏغþ±±Ê¡Ð¢¸ÐÊкº´¨Êкþ±±Ê¡Ò˲ýÊÐÒ˶¼»áÉÂÎ÷Ê¡ººÖÐÊÐÎ÷ÏçÏØɽ¶«Ê¡ÁĶ¼»á¹ÚÏظÊËàÊ¡¶¨Î÷ÊÐμԴÏؾ®ÚêÏزâÓãÕòÃÜÔÆÇøºÓÄÏÕ¯ÕòÏòÑôÇø¹ÜׯµØÇøÇå¾²ÇøÄÏÊнֵÀ¹óÖÝÊ¡Í­ÈÊÊÐÓ¡½­ÍÁ¼Ò×åÃç×å×ÔÖÎÏؽ­ËÕÊ¡Á¬ÔƸÛÊиÓÓÜÇø·ą́Çø³¤ÐÁµêÕò½­Î÷Ê¡¾Å½­ÊÐÎäÄþÏØÎ÷ÇàÇø¾«ÎäÕòÉÂÎ÷Ê¡ººÖÐÊÐÄÏÖ£Çøºþ±±Ê¡Ò˲ýÊÐÎ÷ÁêÇøɽÎ÷Ê¡ÐÃÖÝÊж¨ÏåÏØ°ÙÉ«Êо¸Î÷ÊÐÎ÷³ÇÇøÔÂ̳½ÖµÀÎ÷²ØɽÄÏÊÐËÄ´¨Ê¡¸Ê×βØ×å×ÔÖÎÖݾÅÁúÏØÏòÑôÇø¹ÜׯµØÇø¸ÊËàÊ¡¸ÊÄϲØ×å×ÔÖÎÖÝÁÉÄþÊ¡ÁÉÑôÊа×ËþÇøÉÂÎ÷Ê¡±¦¼¦ÊÐǧÑôÏغþ±±Ê¡Ð¢¸ÐÊкº´¨ÊÐÁÉÄþÊ¡ÉòÑôÊÐÉòºÓÇø¼»ÖÝÇø¹ÙׯÕòºþ±±Ê¡Ò˲ýÊÐÒÄÁêÇøËþ³ÇµØÇøºÍ²¼¿ËÈü¶ûÃɹÅ×ÔÖÎÏØËÄ´¨Ê¡Ò˱öÊн­°²ÏØÇຣʡº£±±²Ø×å×ÔÖÎÖÝÃÅÔ´»Ø×å×ÔÖÎÏØÉÂÎ÷Ê¡Î÷°²ÊÐδÑëÇøº£ÄÏÊ¡º£¿£¿£¿£¿£¿ÚÊÐÃÀÀ¼ÇøÁÉÄþÊ¡ÉòÑôÊÐÃÜÔÆÇø¸ßÁëÕòÏòÑôÇø¾ÆÏÉÇŽֵÀ¸£½¨Ê¡ÏÃÃÅÊкþÀïÇøÄÚÃɹÅÎÚº£ÊÐÎÚ´ïÇøÑÓÇìÇø¿µ×¯ÕòÎ÷²ØɽÄÏÊй±¸ÂÏزýƽÇøÑÓÊÙÕòÒø´¨ÊкØÀ¼Ïغþ±±Ê¡Ò˲ýÊÐïö¹éÏؽ­Î÷Ê¡¸ÓÖÝÊÐʯ³ÇÏغӶ«Çø´óÖ±¹Á½ÖµÀ¸£½¨Ê¡ÄÏƽÊн¨ê±Êгç×óÊÐÄþÃ÷ÏØÎ÷³ÇÇøÌìÇŽֵÀɽ¶«Ê¡Î«·»ÊÐÇàÖÝÊÐÂåÑôÊÐÈêÑôÏغ£µíÇøÇàÁúÇŽֵÀ½­Î÷ʡӥ̶ÊÐÓ཭Çø¼»ÖÝÇø¶«Ê©¹ÅÕòå§ÑôÊÐÄÏÀÖÏغÚÁú½­Ê¡¹þ¶û±õÊаÍÑåÏؽ¹×÷ÊÐÇßÑôÊо²º£Çø´óÇñׯÕòÄÏ¿ªÇø³¤ºç½ÖµÀÁÉÄþÊ¡ÁÉÑôÊа×ËþÇøÇÅÎ÷ÇøÁôÓª½ÖµÀºÚÁú½­Ê¡Æß̨ºÓÊÐÌÒɽÇø¹ðÁÖÊÐÐã·åÇøÖØÇìÊÐÊÐϽÇø±±íÕÇøɽÎ÷Ê¡Ë·ÖÝÊÐɽÒõÏØÇຣʡ¹ûÂå²Ø×å×ÔÖÎÖݸ£½¨Ê¡¸£ÖÝÊÐÓÀÌ©ÏØÐí²ýÊн¨°²ÇøÐÐÌÆÏØÁúÖÝÕò

ÊÛºó·þÎñÉÏÃÅ·þÎñµç»°£¬£¬£¬ÖÇÄÜ·ÖÅäƱ¾Ý£º±¾ÖÜ AI ÏîÄ¿ÍƼö£ºWorldModelBench£¬£¬£¬WorldScore£¬£¬£¬WorldLens¡­ÌìÏÂÄ£×Ó¾®Å磬£¬£¬ÐèҪеÄBenchmark

×÷Õߣü·®ÑÅæÃ΢ÐÅ£ü FFF111f__

2026 ÄêÉÏ°ëÄ꣬£¬£¬¡°ÌìÏÂÄ£×Ó¡±ÒѾ­´ÓÂÛÎÄÀïµÄÊÖÒÕ´Ê£¬£¬£¬Äð³ÉÁË AI ¹«Ë¾¡¢»úеÈ˹«Ë¾¡¢×Ô¶¯¼ÝÊ»¹«Ë¾ºÍ VC ÌÖÂÛÀïÈƲ»¿ªµÄÒªº¦´Ê¡£¡£¡£¡£¡£ÊÓƵÌìÉúÄ£×Ó˵×Ô¼ºÄÜÄ£ÄâÎïÀíÌìÏ£¬£¬£¬¾ßÉíÖÇÄÜÍŶÓ˵×Ô¼ºÒªÓÃÌìÏÂÄ£×ÓѵÁ·»úеÈË£¬£¬£¬×Ô¶¯¼ÝÊ»¹«Ë¾ËµÌìÉúʽ·ÂÕæ¿ÉÒÔÌæ»»Ò»²¿·ÖÕæʵ·²â¡£¡£¡£¡£¡£

µ«ÎÊÌâÒ²ËæÖ®±äµÃ¼âÈñ£ºÒ»¸öÄ£×Ó»áÌìÉúƯÁÁÊÓƵ£¬£¬£¬¾Í¼´ÊÇËüÃ÷È·ÌìÏÂÁËÂ𣿣¿£¿£¿£¿ËüÄܲ»¿É×ñÊØÎïÀí¼ÍÂÉ£¿£¿£¿£¿£¿Äܲ»¿É¼á³Ö 3D ¼¸ºÎÒ»Ö£¿£¿£¿£¿£¿Äܲ»¿É·þÎñ×Ô¶¯¼ÝÊ»ºÍ»úеÈ˾öÒ飿£¿£¿£¿£¿Äܲ»¿É´Ó¡°¿´ÆðÀ´ÏñÕæµÄ¡±£¬£¬£¬×ßµ½¡°ÓÃÆðÀ´ÕæµÄÓÐÓá±£¿£¿£¿£¿£¿

ÒÔÊÇ£¬£¬£¬±ÈÆðÖ»¿´ demo£¬£¬£¬¸üÓ¦¸Ã¿´ benchmark¡£¡£¡£¡£¡£ÏÂÃæÕâÎå¸öÌìÏÂÄ£×Óbenchmark°üÀ¨Í¨ÓÃÊÓƵÌìÏÂÄ£×Ó¡¢Í³Ò»ÌìÏÂÌìÉú¡¢×Ô¶¯¼ÝÊ»¡¢¾ßÉíÖÇÄÜ¡¢ÎïÀíÍÆÀí¡£¡£¡£¡£¡£ËüÃDz»ÊÇËùÓÐ benchmark µÄÈ«¼¯£¬£¬£¬µ«»ù±¾ÁýÕÖÁ˵±ÏÂÌìÏÂÄ£×Ó×îÒªº¦µÄÎåÌõÖ÷Ïß¡£¡£¡£¡£¡£

1. WorldModelBench

Ò»¾ä»°ÏÈÈÝ£º WorldModelBench °ÑÊÓƵÌìÉúÄ£×Ó¿´³ÉÌìÏÂÄ£×ÓÀ´¿¼£¬£¬£¬²»µ«ÎÊ»­ÃæºÃÇ·ÔÃÄ¿£¬£¬£¬»¹ÎÊÖ¸Á֪ʶºÍÎïÀíÊÇ·ñ¾­µÃÆð¼ì²é¡£¡£¡£¡£¡£

WorldModelBench ½â¾öµÄÊÇͨÓÃÊÓƵģ×ÓÆÀ²âÀï×î½¹µãµÄȱ¿Ú£ºÒÑÍùÐí¶à benchmark Æ«ÏòÊÓ¾õÖÊÁ¿¡¢Îı¾¶ÔÆëijÈËÀàÆ«ºÃ£¬£¬£¬µ«¡°ÌìÏÂÄ£×Ó¡±ÕæÕýÒª»Ø¸²µÄÊÇ£¬£¬£¬Ä£×ÓÊÇ·ñÃ÷È·Ðж¯¡¢³¡¾°Òò¹ûºÍ»ù´¡ÎïÀí¡£¡£¡£¡£¡£ËüÁýÕÖ»úеÈË¡¢×Ô¶¯¼ÝÊ»¡¢¹¤Òµ¡¢ÈËÀà»î¶¯¡¢ÓÎÏ·¡¢¶¯»­¡¢×ÔÈ»µÈ 7 ¸öÓ¦ÓÃÓò£¬£¬£¬Ï¸·Öµ½ 56 ¸ö×ÓÓò£¬£¬£¬Ê¹Óà 350 ¸öͼÎÄÌáÐÑ£¬£¬£¬²¢ÓÃÖ¸Áî×·Ë桢֪ʶ¡¢ÎïÀí×ñÊØÈý´óά¶ÈÆÀ²â¡£¡£¡£¡£¡£ÎïÀí²¿·ÖÖص㿴ţ¶ÙµÚÒ»¶¨ÂÉ¡¢Ðα䡢Á÷Ìå¡¢´©Í¸¡¢ÖØÁ¦µÈ³£¼û¡°ÌìϽ¨Ä£»£»£»£»£»£»Ã¾õ¡±¡£¡£¡£¡£¡£ÂÛÎÄ»¹ÖÚ°üÁË 6.7 Íò¸öÈËÀà±ê×¢£¬£¬£¬²¢ÑµÁ·ÁËÒ»¸ö human-aligned judge À´×Ô¶¯ÆÀ²â¡£¡£¡£¡£¡£ÔÚЧ¹ûÉÏ£¬£¬£¬Kling µÈ±ÕÔ´Ä£×ÓÕûÌå¿¿Ç°£¬£¬£¬Mochi¡¢OpenSoraPlan µÈ¿ªÔ´Ä£×ÓÔÚ²¿·Öά¶È¿¿½ü±ÕÔ´Ä£×Ó£¬£¬£¬µ«ÂÛÎĵĽ¹µã½áÂÛ²»ÊÇ¡°Ë­ÒѾ­½â¾öÁËÌìÏÂÄ£×Ó¡±£¬£¬£¬¶øÊÇÍ·²¿Ä£×ÓÈÔÈ»Àë¿É¿¿ÌìÏÂÄ£ÄâÆ÷ºÜÔ¶¡£¡£¡£¡£¡£

Ðû²¼ÍŶÓÀ´×Ô UC Berkeley¡¢UC San Diego¡¢NVIDIA ºÍ MIT £¬£¬£¬Ion Stoica ÊÇ UC Berkeley ÅÌËã»ú½ÌÊÚ£¬£¬£¬Ò²ÊÇ Databricks¡¢Anyscale µÈ¹«Ë¾µÄÅäºÏÊ×´´ÈË£»£»£»£»£»£»Joseph E. Gonzalez ÊÇ Berkeley EECS ½ÌÊÚ¡¢Sky Computing Lab / RISE Lab ÅäºÏÈÏÕæÈË£¬£¬£¬²¢ÊôÓÚ BAIR Éú̬¡£¡£¡£¡£¡£Õâʹ WorldModelBench ²»µ«ÊÇÒ»¸öÊÓƵÆÀ·Ö¼¯£¬£¬£¬¶øÊÇϵͳ¡¢ÊÓ¾õ¡¢»úеÈ˺͸ßЧ AI ÍŶÓÅäºÏÍƶ¯µÄÌìÏÂÄ£×ÓÆÀ²â¡£¡£¡£¡£¡£

ΪʲôֵµÃ¹Ø×¢

ËüÊÇ×îÊÊÏàÖúΪ¡°Í¨ÓÃÊÓƵÌìÏÂÄ£×ÓÈëÃÅ»ù×¼¡±µÄ benchmark¡£¡£¡£¡£¡£ÓÉÓÚËüûÓÐֻͣÁôÔÚÉóÃÀºÍÇåÎú¶È£¬£¬£¬¶øÊÇ°ÑÌìÏÂÄ£×Ó²ð³ÉÓ¦ÓÃÓò¡¢ÎïÀí¼ÍÂɺÍϸÁ£¶È¹ýʧÀàÐÍ£¬£¬£¬ÌØÊâÊʺÏÅжÏÒ»¸öÊÓƵģ×Ó¾¿¾¹ÊÇÔÚ¡°ÌìÉúÏñÌìϵÄͼÏñ¡±£¬£¬£¬ÕÕ¾ÉÔÚ¡°Ä£ÄâÒ»¸öÄÜ×ÔÇ¢ÑÝ»¯µÄÌìÏ¡±¡£¡£¡£¡£¡£

ÏîÄ¿£ºhttps://worldmodelbench-team.github.io/

ÂÛÎÄ£ºhttps://arxiv.org/abs/2502.20694

´úÂëÓëÆÀ²â£ºhttps://github.com/WorldModelBench-Team/WorldModelBench

²âÊÔÊý¾Ý£ºhttps://huggingface.co/WorldModelBench-Team

2. WorldScore

Ò»¾ä»°ÏÈÈÝ£º WorldScore °Ñ 3D¡¢4D¡¢Í¼ÉúÊÓƵ¡¢ÎÄÉúÊÓƵģ×ӷŵ½Í³Ò»¸ö¡°ÏÂÒ»³¡¾°ÌìÉú¡±¿ò¼ÜÀïÆÀ²â£¬£¬£¬Öص㿴ģ×ÓÄܲ»¿É°´Ïà»ú¹ì¼£Ò»Á¬ÌìÉúÌìÏ¡£¡£¡£¡£¡£

¸ÅÊö£º WorldScore ½â¾öµÄÊÇ¡°²î±ðÌìÏÂÌìÉúõ辶ÎÞ·¨¹«Õý½ÏÁ¿¡±µÄÎÊÌâ¡£¡£¡£¡£¡£3D Ä£×Ó¡¢4D Ä£×ÓºÍÊÓƵģ×ÓÊä³öÐÎ̬²î±ð£¬£¬£¬ÒÑÍùºÜÄÑÔÚͳһÕÅÄÚÍâ½ÏÁ¿¡£¡£¡£¡£¡£WorldScore ½«ÌìÏÂÌìÉú²ð³ÉһϵÁлùÓÚÃ÷È·Ïà»ú¹ì¼£µÄ next-scene generation ʹÃü£¬£¬£¬Óà 3000 ¸öÑùÀýÁýÕÖ¾²Ì¬/¶¯Ì¬¡¢ÊÒÄÚ/ÊÒÍ⡢дʵ/Æø¸Å»¯µÈ³¡¾°£¬£¬£¬²¢´Ó¿É¿ØÐÔ¡¢ÖÊÁ¿¡¢¶¯Ì¬Èý·½Ãæ´ò·Ö¡£¡£¡£¡£¡£¿£¿£¿£¿£¿É¿ØÐÔ¿´Ïà»ú¿ØÖÆ¡¢ÎïÌå¿ØÖƺÍÄÚÈݶÔÆ룬£¬£¬ÖÊÁ¿¿´ 3D Ò»ÖÂÐÔ¡¢¹â¶ÈÒ»ÖÂÐÔ¡¢Æø¸ÅÒ»ÖÂÐÔºÍÖ÷¹ÛÖÊÁ¿£¬£¬£¬¶¯Ì¬¿´Ô˶¯×¼È·ÐÔ¡¢·ù¶ÈºÍÁ÷ͨ¶È¡£¡£¡£¡£¡£¹Ù·½Ð§¹ûÏÔʾ£¬£¬£¬Voyager¡¢WonderWorld¡¢LucidDreamer µÈ 3D/³¡¾°ÌìÉúÄ£×ÓÔÚ¾²Ì¬ÌìÏÂÌìÉúÉϺÜÇ¿£¬£¬£¬CogVideoX-I2V µÈÊÓƵģ×ÓÔÚ¶¯Ì¬·ÖÊýÉÏÌåÏÖÍ»³ö£»£»£»£»£»£»Í¬Ê±£¬£¬£¬T2V Ä£×Óͨ³£¸üÈÝÒ×±»Îı¾¿ØÖÆ£¬£¬£¬I2V Ä£×ÓÍùÍù»­ÖʸüÎÈ¡£¡£¡£¡£¡£

WorldScore ÊÇ Stanford University ÍŶÓÐû²¼£¬£¬£¬ÂÛÎÄ×÷ÕßΪ Haoyi Duan¡¢Hong-Xing Yu¡¢Sirui Chen¡¢Li Fei-Fei¡¢Jiajun Wu¡£¡£¡£¡£¡£Li Fei-Fei ÊÇ Stanford ÅÌËã»úϵ Sequoia Professor¡¢Stanford HAI Ê×´´ÁªºÏÖ÷ÈΣ¬£¬£¬ÔøÈÎ Stanford AI Lab Ö÷ÈÎºÍ Google Cloud AI/ML Ê×ϯ¿Æѧ¼Ò£¬£¬£¬¿ª°ìWorld Labs£¬£¬£¬Ò²ÊÇ ImageNet µÄÒªº¦Íƶ¯Õߣ»£»£»£»£»£»Jiajun Wu ÊÇ Stanford ÅÌËã»úϵÖúÀí½ÌÊÚ£¬£¬£¬Í¬Ê±¼æÈÎÐÄÀíѧƫÏò£¬£¬£¬Ñо¿ÅÌËã»úÊÓ¾õ¡¢»úеÈ˺ÍÅÌËãÈÏÖª¿Æѧ£¬£¬£¬ËùÔÚÆ«Ïò×ÔÈ»Ìù½ü¡°ÌìÏÂÔõÑù±»ÌåÏÖ¡¢ÍÆÀíºÍÌìÉú¡±¡£¡£¡£¡£¡£Òò´Ë£¬£¬£¬WorldScore ¸üÏñÊÇ Stanford ÊÓ¾õѧϰ¡¢3D/ÎïÀí³¡¾°Ã÷È·ºÍÌìÉúÄ£×ÓÑо¿ÂöÂçϵÄÒ»Ì×ÌìÏÂÌìÉú»³±§¡£¡£¡£¡£¡£

ΪʲôֵµÃ¹Ø×¢£º

ÈôÊÇ˵ WorldModelBench ¸üÏñ¡°ÊÓƵģ×ÓÊDz»ÊÇÌìÏÂÄ£×Ó¡±µÄÌå¼ì£¬£¬£¬WorldScore Ôò¸üÏñ¡°ÌìÏÂÌìÉúõ辶´óºáÆÀ¡±¡£¡£¡£¡£¡£Ëü°Ñ 3D¡¢4D¡¢ÊÓƵÈýÌõÊÖÒÕõ辶·Å½øͳһЭÒ飬£¬£¬¶Ô´´Òµ¹«Ë¾ºÍÑо¿ÍŶÓÅжÏÊÖÒÕõ辶ºÜÓмÛÖµ£º¾¿¾¹¸ÃѺÊÓƵÌìÉú¡¢3D ÌìÉú£¬£¬£¬Õվɸü»ìÏýµÄ 4D ÌìÏÂÌìÉú¡£¡£¡£¡£¡£

ÏîÄ¿Ò³£ºhttps://haoyi-duan.github.io/WorldScore/

ÂÛÎÄ£ºhttps://arxiv.org/abs/2504.00983

´úÂ룺https://github.com/haoyi-duan/WorldScore

Êý¾Ý¼¯£ºhttps://huggingface.co/datasets/haoyi-duan/WorldScore

3. WorldLens

Ò»¾ä»°ÏÈÈÝ£º WorldLens רÃÅÆÀ¹À×Ô¶¯¼ÝÊ»ÌìÏÂÄ£×Ó£¬£¬£¬½¹µãÎÊÌâÊÇ£ºÌìÉúµÄ¼ÝÊ»Ììϲ»µ«ÒªÏñ£¬£¬£¬»¹Òª¼¸ºÎÎȹ̡¢ÎïÀíºÏÀí¡¢ÐÐΪÇå¾²¡¢¶ÔÏÂÓÎʹÃüÓÐÓᣡ£¡£¡£¡£

WorldLens ½â¾öµÄÊÇ×Ô¶¯¼ÝÊ»³¡¾°Àï¡°ÊÓ¾õ±ÆÕ浫¹¦Ð§Ê§Õ桱µÄÎÊÌâ¡£¡£¡£¡£¡£¼ÝÊ»ÌìÏÂÄ£×ÓºÜÈÝÒ×ÌìÉú¿´ÆðÀ´ºÏÀíµÄ 4D ³¡¾°£¬£¬£¬µ«Ò»µ©Çе½ÖØÐÞ¡¢±Õ»·¿ØÖÆ¡¢ÏÂÓμì²âÖ§½â£¬£¬£¬¾Í»á̻¶¼¸ºÎƯÒÆ¡¢Ô˶¯²»Îȹ̡¢Åöײ¡¢Æ«Àëõ辶µÈÎÊÌâ¡£¡£¡£¡£¡£WorldLens ÓÃÎå¸ö»¥²¹Æ«ÏòÆÀ²â£ºGeneration¡¢Reconstruction¡¢Action-Following¡¢Downstream Task¡¢Human Preference£¬£¬£¬²¢½øÒ»²½¹¹½¨ WorldLens-26K ÈËÀàÆ«ºÃÊý¾ÝºÍ WorldLens-Agent ÊÓ¾õÓïÑÔÆÀÉóÆ÷¡£¡£¡£¡£¡£ÏîÄ¿½áÂÛºÜÊÇÏÖʵ£ºÃ»Óмòµ¥Ä£×ÓÄÜÔÚËùÓÐά¶ÈͳÖΰñµ¥£¬£¬£¬ÎÆÀíÕæʵµÄÄ£×Ó¿ÉÄÜÎïÀíºÍ¼¸ºÎ²»ÎÈ£¬£¬£¬¼¸ºÎÎȹ̵ÄÄ£×ÓÓÖ¿ÉÄÜÐÐΪϸ½Úȱ·¦¡£¡£¡£¡£¡£ÂÛÎÄʵÑéÖУ¬£¬£¬DiST-4D ÔÚÐÂÊÓ½ÇÖÊÁ¿ºÍ²¿·ÖÏÂÓÎʹÃüÉÏÌåÏÖÇ¿£¬£¬£¬OpenDWM µÈÄ£×ÓÔÚ¼¸ºÎÎȹÌÐÔÉÏÓÐÓÅÊÆ£¬£¬£¬DriveDreamer-2 Ò²ÔÚʹÃüÏà¹ØÖ¸±êÖнøÈëÇ¿ÊÆÌݶÓ¡£¡£¡£¡£¡£

WorldLens ÓÉ WorldBench Team Ðû²¼£¬£¬£¬NUS¡¢University of Macau¡¢USTC¡¢Zhejiang University¡¢NTU¡¢Horizon Robotics¡¢HUST¡¢TUM¡¢Fudan University¡¢Shanghai AI Lab¡¢A*STAR¡¢CNRS µÈ¡£¡£¡£¡£¡£½¹µã×÷Õß°üÀ¨ Ao Liang¡¢Lingdong Kong¡¢Tianyi Yan¡¢Hongsi Liu¡¢Ziqi HuangµÈ¡£¡£¡£¡£¡£Lingdong Kong ÊÇ NUS ÅÌËã»úϵ²©Ê¿Éú¡¢ÏîĿǣͷÈË£¬£¬£¬ºã¾Ãά»¤×Ô¶¯¼ÝÊ»ºÍ 3D/4D ÌìϽ¨Ä£Ïà¹ØÏîÄ¿£»£»£»£»£»£»Ziwei Liu ÊÇ NTU MMLab µÄ Associate Professor / Provost's Chair in AI£¬£¬£¬Ñо¿ÁýÕÖÅÌËã»úÊÓ¾õ¡¢»úеѧϰºÍÅÌËã»úͼÐΣ»£»£»£»£»£»Liang Pan ÊÇ Shanghai AI Lab Ñо¿¿Æѧ¼Ò£¬£¬£¬ºã¾Ã×ö 3D ÊÓ¾õ¡¢world models ºÍ embodied AI£»£»£»£»£»£»Wei Tsang Ooi ÊÇ NUS ¶àýϵһÇÐÆ«Ïò½ÌÊÚ¡£¡£¡£¡£¡£Õâ¸ö×éºÏÈà WorldLens ͬʱվÔÚ×Ô¶¯¼ÝÊ»¡¢4D ³¡¾°¡¢3D ÊÓ¾õºÍÈËÀàÆ«ºÃÆÀ²âµÄ½»Ö¯µãÉÏ¡£¡£¡£¡£¡£

ΪʲôֵµÃ¹Ø×¢

×Ô¶¯¼ÝÊ»ÊÇÌìÏÂÄ£×Ó×îÈÝÒ×ÂäµØ¡¢Ò²×î²»¿ÉÖ»¿´ demo µÄÁìÓò¡£¡£¡£¡£¡£WorldLens µÄ¼ÛÖµÔÚÓÚ°Ñ¡°ÌìÉúÌìÏ¡±À­»Øµ½ÁËÇå¾²¡¢±Õ»·ºÍÏÂÓÎʹÃüÉÏ£ºÒ»¸öÄ£×ÓÌìÉúµÄÊÓƵÔÙƯÁÁ£¬£¬£¬ÈôÊÇ»áÈøÐ֪ģ×ÓÍË»¯£¬£¬£¬»òÕßÈà planner ÔÚÄ£ÄâÌìÏÂÀïײ³µ£¬£¬£¬Ëü¾Í²»¿ÉËã¿É¿¿µÄ¼ÝÊ»ÌìÏÂÄ£×Ó¡£¡£¡£¡£¡£

ÏîÄ¿Ò³£ºhttps://worldbench.github.io/worldlens

ÂÛÎÄ£ºhttps://arxiv.org/abs/2512.10958

´úÂ룺https://github.com/worldbench/WorldLens

Leaderboard£ºhttps://huggingface.co/spaces/worldbench/worldlens

4. WorldArena 2.0

Ò»¾ä»°ÏÈÈÝ£º WorldArena ÆÀ¹À¾ßÉíÌìÏÂÄ£×ÓÊÇ·ñÕæµÄÄÜ·þÎñ»úеÈËʹÃü£¬£¬£¬¶ø²»µ«½öÊÇÌìÉúÒ»¶Î¿´ÆðÀ´²»´íµÄ²Ù×÷ÊÓƵ¡£¡£¡£¡£¡£

WorldArena ½â¾öµÄÊǾßÉíÖÇÄÜÀï¡°¸ÐÖªÖÊÁ¿ºÍ¹¦Ð§¼ÛÖµÍѽڡ±µÄÎÊÌâ¡£¡£¡£¡£¡£Ðí¶à embodied world model ¿ÉÒÔÌìÉú¸ßÖÊÁ¿ÊÓƵ£¬£¬£¬µ«»úеÈËÕæÕýÌåÌùµÄÊÇ£ºËüÄܲ»¿ÉºÏ³ÉÓÐÓÃѵÁ·Êý¾Ý£¿£¿£¿£¿£¿Äܲ»¿É×÷Ϊ policy evaluator ÆÀÔ¤Õ½ÂÔ£¿£¿£¿£¿£¿Äܲ»¿É×ÊÖú agent ×ö action planning£¿£¿£¿£¿£¿WorldArena ÏÈÓà 16 ¸öÖ¸±êÆÀ²âÊÓƵ¸ÐÖªÖÊÁ¿£¬£¬£¬ÁýÕÖÊÓ¾õÖÊÁ¿¡¢Ô˶¯ÖÊÁ¿¡¢ÄÚÈÝÒ»ÖÂÐÔ¡¢ÎïÀí×ñÊØ¡¢3D ׼ȷÐÔ¡¢¿É¿ØÐÔ 6 ¸ö×Óά¶È£¬£¬£¬ÔÙÆÀ¹ÀÈýÀà¾ßÉí¹¦Ð§£ºData Engine¡¢Policy Evaluator¡¢Action Planner£¬£¬£¬²¢Ìá³ö EWMScore ×öͳһ»³±§¡£¡£¡£¡£¡£

WorldArena 2.0 ÔÚÕâ¸ö»ù´¡ÉϼÌÐøÀ©Õ¹£ºµÚÒ»£¬£¬£¬Ä£Ì¬´Ó vision-only À©Õ¹µ½ visuo-tactile£¬£¬£¬ÈÃÄ£×Ó±ØÐè´¦Àí½Ó´¥¡¢Á¦¡¢»¬¶¯ºÍÖÊÁϽ»»¥£»£»£»£»£»£»µÚ¶þ£¬£¬£¬¹¦Ð§´ÓÀëÏßÆÀ¹ÀÀ©Õ¹µ½ online RL£¬£¬£¬°ÑÌìÏÂÄ£×Ó¿´³É¿É½»»¥ÇéÐÎÀ´ÑµÁ·ºÍË¢ÐÂÕ½ÂÔ£»£»£»£»£»£»µÚÈý£¬£¬£¬Æ½Ì¨´Ó simulator-only À©Õ¹µ½ RoboTwin 2.0¡¢LIBERO ºÍ AgileX split-type ALOHA Õæʵ»úеÈË¡£¡£¡£¡£¡£ËüµÄ½áÂÛÒ²¸üÌù½ü¹¤ÒµÎÊÌ⣺ģÄâÇéÐÎÀïµÄºÃЧ¹û²»¼´ÊÇÕæʵ°²ÅÅÄÜÁ¦£¬£¬£¬sim-to-real gap ÈÔÈ»ÊǾßÉíÌìÏÂÄ£×Ó±ØÐè¿ç¹ýµÄÓ²Ãż÷¡£¡£¡£¡£¡£

¶à»ú¹¹ÁªºÏÁªºÏÐû²¼£ºTsinghua University¡¢Shanghai Jiao Tong University¡¢The University of Hong Kong¡¢Princeton University¡¢Chinese Academy of Sciences¡¢University of Science and Technology of China¡¢Peking University¡¢National University of Singapore¡£¡£¡£¡£¡£×÷Õß°üÀ¨ Yu Shang¡¢Zhuohang Li¡¢Yiding Ma¡¢Weikang Su¡¢Xin JinµÈ¡£¡£¡£¡£¡£Yong Li ÊÇÇ廪 FIB Lab Ö÷ÒªÈÏÕæÈË£¬£¬£¬ºã¾Ã×ö AI¡¢Êý¾ÝÍÚ¾òºÍ¶¼»áÖÇÄÜ£»£»£»£»£»£»Jun Zhu ÊÇÇ廪 Bosch AI Professor¡¢IEEE/AAAI Fellow£¬£¬£¬Ñо¿»úеѧϰ¡¢Éî¶ÈÌìÉúÄ£×Ó¡¢Ç¿»¯Ñ§Ï°ºÍ·´¿¹Â³°ôÐÔ£»£»£»£»£»£»Wenwu Zhu ÊÇÇ廪ÅÌËã»úϵ½ÌÊÚ£¬£¬£¬ÔøÔÚ Microsoft Research Asia¡¢Intel Research China µÈ¹¤ÒµÑо¿»ú¹¹µ£µ±Ö÷Òª½ÇÉ«£»£»£»£»£»£»Tat-Seng Chua ÊÇ NUS School of Computing µÄ KITHCT Chair Professor£¬£¬£¬Ò²ÔøÈÎ NUS Computing ´´ÔºÔº³¤¡£¡£¡£¡£¡£Õâ¸öÕóÈÝÈà WorldArena ¸üÏñ¾ßÉíÌìÏÂÄ£×ÓÁìÓòµÄ¡°Ñ§Êõ-»úеÈË-¶àýÌå-ϵͳ¡±ÁªºÏÆÀ²â¡£¡£¡£¡£¡£

ΪʲôֵµÃ¹Ø×¢

Ëü°ÑÌìÏÂÄ£×Ó´Ó¡°»á²»»áÌìÉúδÀ´Ö¡¡±Íƽøµ½¡°Äܲ»¿ÉÈûúеÈ˱äÇ¿¡±¡£¡£¡£¡£¡£Õâ¶Ô¾ßÉíÖÇÄܺÜÊÇÒªº¦£¬£¬£¬ÓÉÓÚÕæÕýµÄ»úеÈËÌìÏÂÄ£×Ó²»µ«ÊÇÄ£Äâ»­Ã棬£¬£¬¶øÊÇÒª³ÉΪÊý¾ÝÒýÇæ¡¢ÇéÐÎÊðÀíºÍÍýÏëÄ£¿£¿£¿£¿£¿é¡£¡£¡£¡£¡£WorldArena ÊÇÏÖÔÚÉÙÊý°Ñ perception ºÍ functional utility ·ÅÔÚͳһ¸ö¿ò¼ÜÀïÑÏËàÆÀ¹ÀµÄ benchmark¡£¡£¡£¡£¡£

ÏîÄ¿Ò³£ºhttps://world-arena.ai/

ÂÛÎÄ£ºhttps://arxiv.org/abs/2602.08971

´úÂ룺https://github.com/tsinghua-fib-lab/WorldArena

Leaderboard£ºhttps://huggingface.co/spaces/WorldArena/WorldArena

5. Physics-IQ / Physics-IQ Verified

Ò»¾ä»°ÏÈÈÝ£º Physics-IQ ÓÃÕæʵÎïÀíʵÑéÊÓƵ¿¼ÊÓƵÌìÉúÄ£×Ó£¬£¬£¬Ö±½Ó×·ÎÊÒ»¸ö»ù´¡ÎÊÌ⣺ģ×ÓÊÇÔÚÃ÷È·ÎïÀí£¬£¬£¬ÕÕ¾ÉÖ»ÊÇÔÚÕ¹ÍûÏñËØ£¿£¿£¿£¿£¿

Physics-IQ ½â¾öµÄÊÇ¡°ÊÓ¾õÕæʵ²»¼´ÊÇÎïÀíÃ÷È·¡±µÄÎÊÌâ¡£¡£¡£¡£¡£ËüÒªÇóÄ£×Óƾ֤ÕæʵÌìÏÂÎïÀíÊÓƵµÄÇ°°ë¶ÎÕ¹ÍûºóÐøÑÝ»¯£¬£¬£¬ÁýÕÖ¹ÌÌåÁ¦Ñ§¡¢Á÷Ì嶯Á¦Ñ§¡¢¹âѧ¡¢ÈÈÁ¦Ñ§¡¢´ÅѧÎåÀàÎïÀíÕ÷Ï󡣡£¡£¡£¡£ÆÀ²â²»¿´ÊÓƵÊÇ·ñ¡°¿´ÆðÀ´¸ß¼¶¡±£¬£¬£¬¶øÊÇ¿´Ä£×ÓÕ¹ÍûµÄÎïÀíת±äÔڿռ䡢ʱ¼äºÍÔ˶¯ÇøÓòÉÏÊÇ·ñ¿¿½üÕæʵЧ¹û¡£¡£¡£¡£¡£ÂÛÎÄÆÀ¹ÀÁË Sora¡¢Runway¡¢Pika¡¢Lumiere¡¢Stable Video Diffusion¡¢VideoPoet µÈÄ£×Ó£¬£¬£¬½áÂÛÏ൱ֱ½Ó£ºÄ¿½ñÊÓƵÌìÉúÄ£×ÓµÄÎïÀíÃ÷È·ÈÔÈ»ÓÐÏÞ£¬£¬£¬²¢ÇÒÓëÊÓ¾õÕæʵ¸Ð²¢²»Ç¿Ïà¹Ø£»£»£»£»£»£»¶àÖ¡Ìõ¼þͨ³£±Èµ¥Ö¡Ìõ¼þ¸üÓÐ×ÊÖú£¬£¬£¬Á÷ÌåÕ÷ÏóÏà¶Ô¸üÈÝÒ×£¬£¬£¬¹ÌÌåÁ¦Ñ§¸üÄÑ¡£¡£¡£¡£¡£2026 Äê 6 ÔÂÓÖ·ºÆðÁË Physics-IQ Verified£¬£¬£¬¶ÔԭʼÑù±¾ºÍ prompt ×öÁËϵͳÉó¼ÆºÍÐÞ¶©£»£»£»£»£»£»ÈôÊÇÊÇÏÖʵÅÜÆÀ²â£¬£¬£¬ÓÅÏȹØ×¢ Verified °æ±¾»á¸üÎÈ¡£¡£¡£¡£¡£

Physics-IQ ÓÉ Saman Motamed¡¢Laura Culp¡¢Kevin Swersky¡¢Priyank Jaini¡¢Robert Geirhos Ðû²¼£¬£¬£¬ÂÛÎÄÊ×Ò³±ê×¢ Saman Motamed À´×Ô INSAIT / Sofia University£¬£¬£¬Ïà¹ØÊÂÇéÔÚ Google DeepMind ʱ´úÍê³É£¬£¬£¬ÆäÓà×÷ÕßÀ´×Ô Google DeepMind¡£¡£¡£¡£¡£INSAIT ¹Ù·½±¨µÀÒ²Ã÷È·³Æ¸Ã benchmark ÓÉ INSAIT Óë Google DeepMind ÁªºÏ¿ª·¢£¬£¬£¬²¢ÔÚ ICCV 2025 Òý·¢¹Ø×¢¡£¡£¡£¡£¡£Saman Motamed ÊÇ INSAIT µÄ ELLIS PhD student£¬£¬£¬Ñо¿ÅÌËã»úÊÓ¾õ¡¢ÌìÉúÄ£×ÓºÍ intuitive physics£»£»£»£»£»£»Robert Geirhos ÊÇ Google DeepMind Staff Research Scientist£¬£¬£¬ºã¾Ã×öÊÓ¾õ³°ôÐÔ¡¢¿ÉÚ¹ÊÍÐÔºÍÈË»úÊÓ¾õ½ÏÁ¿£¬£¬£¬´ú±íÊÂÇé°üÀ¨Éî¶ÈÍøÂç texture bias / shortcut learning Æ«Ïò£»£»£»£»£»£»Priyank Jaini ºÍ Kevin Swersky Ò²¶¼À´×Ô Google DeepMind µÄÌìÉúÄ£×ÓÓë»úеѧϰÑо¿Ïß¡£¡£¡£¡£¡£Õâ¸öÍŶӵÄÓÅÊÆÊÇ°ÑÕæʵÎïÀíʵÑé¡¢ÊÓ¾õÐÄÀíѧʽÎÊÌâÒâʶºÍÇ°ÑØÊÓƵģ×ÓÑо¿·Åµ½ÁËÒ»Æ𡣡£¡£¡£¡£

ΪʲôֵµÃ¹Ø×¢

ÎïÀíÊÇÌìÏÂÄ£×ÓÐðÊÂÀï×îÄÑÈÆ¿ªµÄµ×²ãÄÜÁ¦¡£¡£¡£¡£¡£Ò»¸öÄ£×Ó¿ÉÒÔÌìÉúÓ°Ï·¸ÐË®»¨¡¢±¬Õ¨ºÍÅöײ£¬£¬£¬µ«ÈôÊÇËüÎÞ·¨Õ¹ÍûÕæʵÎïÌå»áÔõÑù¼ÌÐøÔ˶¯£¬£¬£¬Ëü¾ÍºÜÄѳÉΪ¿É¿¿µÄÌìÏÂÄ£×Ó¡£¡£¡£¡£¡£Physics-IQ µÄ¼ÛÖµÔÚÓÚËü×ã¹»Ö±½Ó£ºÉÙ½²¡°Ïñ²»Ïñ¡±£¬£¬£¬¶àÎÊ¡°¶Ô²î³Ø¡±¡£¡£¡£¡£¡£

ÏîÄ¿Ò³£ºhttps://physics-iq.github.io/

ÂÛÎÄ£ºhttps://arxiv.org/abs/2501.09038

´úÂ룺https://github.com/google-deepmind/physics-iq-benchmark

Physics-IQ Verified£ºhttps://arxiv.org/abs/2606.18943

ÌìÏÂÄ£×ÓÏÖÔÚ´¦ÔÚÒ»¸ö¿´·¨¿ìËÙÉýΡ¢ÊÖÒÕõ辶·Ö²æ¡¢ÆÀ²âϵͳ׷׏¤ÒµÐðʲ¹¿ÎµÄÔÓÂÒÆÚ¡£¡£¡£¡£¡£Ò²ÕýÓÉÓÚÔÓÂÒ£¬£¬£¬benchmark ²ÅÖµµÃ¿´£ºËüÃÇÊÇÕⳡ½ç˵ÕùÈ¡À£¬£¬ÉÙÊýÄÜ°Ñ¿ÚºÅÂäµ½²âÊÔÌâÉϵŤ¾ß¡£¡£¡£¡£¡£

Benchmark ¿ÉÒÔ¸øÒ»¸ö»¹ºÜËÉÉ¢µÄ¿´·¨»­½çÏß¡£¡£¡£¡£¡£WorldModelBench ¿¼²ìÊÓƵģ×ÓÓÐûÓÐ֪ʶºÍÎïÀíÔ¼Êø£»£»£»£»£»£»WorldScore Ȩºâ²î±ðÌìÏÂÌìÉúõ辶Äܲ»¿Éͳһ½ÏÁ¿£»£»£»£»£»£»WorldLens ¿ÉÒÔÓÃÀ´ÆÀ¹À¼ÝÊ»ÌìÏÂÊÇ·ñÕæµÄ¶ÔÇå¾²ºÍÏÂÓÎʹÃüÓÐÓ㻣»£»£»£»£»WorldArena ¿´µÄÊǾßÉíÌìÏÂÄ£×ÓÊÇ·ñÕæµÄÄÜ·þÎñ»úеÈË£»£»£»£»£»£»Physics-IQ Ôò°Ñ½çÏßѹµ½×îµ×²ã£¬£¬£¬ÎÊÄ£×Ó¾¿¾¹ÊÇÔÚÃ÷È·ÎïÀí£¬£¬£¬ÕÕ¾ÉÖ»ÊÇÔÚ²¹ÏñËØ¡£¡£¡£¡£¡£

ÌìÏÂÄ£×ÓÉú³¤µ½½ñÌ죬£¬£¬ÆÀ²âµÄÖ÷ÒªÐÔÕýÔÚÈÕÒæ͹ÏÔ£¬£¬£¬ÓïÑÔÄ£×ÓÒѾ­¸üÔç֤ʵÎúÕâ¼þÊ£¬£¬£¬½è¼øÓïÑÔÄ£×ÓµÄÂÄÀú£¬£¬£¬ÔçÈÕÈ·¶¨½¹µãÆÀ²âÆ«Ïò£¬£¬£¬Ò²½«³ÉΪδÀ´¾ºÕùµÄÒªº¦¡£¡£¡£¡£¡£

µã¸ö¡°°®ÐÄ¡±£¬£¬£¬ÔÙ×ß °É

±¾ÔÂÐÐҵЭ»áÐû²¼ÐÂÑо¿±¨¸æ±¾ÖÜ AI ÏîÄ¿ÍƼö£ºWorldModelBench£¬£¬£¬WorldScore£¬£¬£¬WorldLens¡­ÌìÏÂÄ£×Ó¾®Å磬£¬£¬ÐèҪеÄBenchmark

×÷Õߣü·®ÑÅæÃ΢ÐÅ£ü FFF111f__

2026 ÄêÉÏ°ëÄ꣬£¬£¬¡°ÌìÏÂÄ£×Ó¡±ÒѾ­´ÓÂÛÎÄÀïµÄÊÖÒÕ´Ê£¬£¬£¬Äð³ÉÁË AI ¹«Ë¾¡¢»úеÈ˹«Ë¾¡¢×Ô¶¯¼ÝÊ»¹«Ë¾ºÍ VC ÌÖÂÛÀïÈƲ»¿ªµÄÒªº¦´Ê¡£¡£¡£¡£¡£ÊÓƵÌìÉúÄ£×Ó˵×Ô¼ºÄÜÄ£ÄâÎïÀíÌìÏ£¬£¬£¬¾ßÉíÖÇÄÜÍŶÓ˵×Ô¼ºÒªÓÃÌìÏÂÄ£×ÓѵÁ·»úеÈË£¬£¬£¬×Ô¶¯¼ÝÊ»¹«Ë¾ËµÌìÉúʽ·ÂÕæ¿ÉÒÔÌæ»»Ò»²¿·ÖÕæʵ·²â¡£¡£¡£¡£¡£

µ«ÎÊÌâÒ²ËæÖ®±äµÃ¼âÈñ£ºÒ»¸öÄ£×Ó»áÌìÉúƯÁÁÊÓƵ£¬£¬£¬¾Í¼´ÊÇËüÃ÷È·ÌìÏÂÁËÂ𣿣¿£¿£¿£¿ËüÄܲ»¿É×ñÊØÎïÀí¼ÍÂÉ£¿£¿£¿£¿£¿Äܲ»¿É¼á³Ö 3D ¼¸ºÎÒ»Ö£¿£¿£¿£¿£¿Äܲ»¿É·þÎñ×Ô¶¯¼ÝÊ»ºÍ»úеÈ˾öÒ飿£¿£¿£¿£¿Äܲ»¿É´Ó¡°¿´ÆðÀ´ÏñÕæµÄ¡±£¬£¬£¬×ßµ½¡°ÓÃÆðÀ´ÕæµÄÓÐÓá±£¿£¿£¿£¿£¿

ÒÔÊÇ£¬£¬£¬±ÈÆðÖ»¿´ demo£¬£¬£¬¸üÓ¦¸Ã¿´ benchmark¡£¡£¡£¡£¡£ÏÂÃæÕâÎå¸öÌìÏÂÄ£×Óbenchmark°üÀ¨Í¨ÓÃÊÓƵÌìÏÂÄ£×Ó¡¢Í³Ò»ÌìÏÂÌìÉú¡¢×Ô¶¯¼ÝÊ»¡¢¾ßÉíÖÇÄÜ¡¢ÎïÀíÍÆÀí¡£¡£¡£¡£¡£ËüÃDz»ÊÇËùÓÐ benchmark µÄÈ«¼¯£¬£¬£¬µ«»ù±¾ÁýÕÖÁ˵±ÏÂÌìÏÂÄ£×Ó×îÒªº¦µÄÎåÌõÖ÷Ïß¡£¡£¡£¡£¡£

1. WorldModelBench

Ò»¾ä»°ÏÈÈÝ£º WorldModelBench °ÑÊÓƵÌìÉúÄ£×Ó¿´³ÉÌìÏÂÄ£×ÓÀ´¿¼£¬£¬£¬²»µ«ÎÊ»­ÃæºÃÇ·ÔÃÄ¿£¬£¬£¬»¹ÎÊÖ¸Á֪ʶºÍÎïÀíÊÇ·ñ¾­µÃÆð¼ì²é¡£¡£¡£¡£¡£

WorldModelBench ½â¾öµÄÊÇͨÓÃÊÓƵģ×ÓÆÀ²âÀï×î½¹µãµÄȱ¿Ú£ºÒÑÍùÐí¶à benchmark Æ«ÏòÊÓ¾õÖÊÁ¿¡¢Îı¾¶ÔÆëijÈËÀàÆ«ºÃ£¬£¬£¬µ«¡°ÌìÏÂÄ£×Ó¡±ÕæÕýÒª»Ø¸²µÄÊÇ£¬£¬£¬Ä£×ÓÊÇ·ñÃ÷È·Ðж¯¡¢³¡¾°Òò¹ûºÍ»ù´¡ÎïÀí¡£¡£¡£¡£¡£ËüÁýÕÖ»úеÈË¡¢×Ô¶¯¼ÝÊ»¡¢¹¤Òµ¡¢ÈËÀà»î¶¯¡¢ÓÎÏ·¡¢¶¯»­¡¢×ÔÈ»µÈ 7 ¸öÓ¦ÓÃÓò£¬£¬£¬Ï¸·Öµ½ 56 ¸ö×ÓÓò£¬£¬£¬Ê¹Óà 350 ¸öͼÎÄÌáÐÑ£¬£¬£¬²¢ÓÃÖ¸Áî×·Ë桢֪ʶ¡¢ÎïÀí×ñÊØÈý´óά¶ÈÆÀ²â¡£¡£¡£¡£¡£ÎïÀí²¿·ÖÖص㿴ţ¶ÙµÚÒ»¶¨ÂÉ¡¢Ðα䡢Á÷Ìå¡¢´©Í¸¡¢ÖØÁ¦µÈ³£¼û¡°ÌìϽ¨Ä£»£»£»£»£»£»Ã¾õ¡±¡£¡£¡£¡£¡£ÂÛÎÄ»¹ÖÚ°üÁË 6.7 Íò¸öÈËÀà±ê×¢£¬£¬£¬²¢ÑµÁ·ÁËÒ»¸ö human-aligned judge À´×Ô¶¯ÆÀ²â¡£¡£¡£¡£¡£ÔÚЧ¹ûÉÏ£¬£¬£¬Kling µÈ±ÕÔ´Ä£×ÓÕûÌå¿¿Ç°£¬£¬£¬Mochi¡¢OpenSoraPlan µÈ¿ªÔ´Ä£×ÓÔÚ²¿·Öά¶È¿¿½ü±ÕÔ´Ä£×Ó£¬£¬£¬µ«ÂÛÎĵĽ¹µã½áÂÛ²»ÊÇ¡°Ë­ÒѾ­½â¾öÁËÌìÏÂÄ£×Ó¡±£¬£¬£¬¶øÊÇÍ·²¿Ä£×ÓÈÔÈ»Àë¿É¿¿ÌìÏÂÄ£ÄâÆ÷ºÜÔ¶¡£¡£¡£¡£¡£

Ðû²¼ÍŶÓÀ´×Ô UC Berkeley¡¢UC San Diego¡¢NVIDIA ºÍ MIT £¬£¬£¬Ion Stoica ÊÇ UC Berkeley ÅÌËã»ú½ÌÊÚ£¬£¬£¬Ò²ÊÇ Databricks¡¢Anyscale µÈ¹«Ë¾µÄÅäºÏÊ×´´ÈË£»£»£»£»£»£»Joseph E. Gonzalez ÊÇ Berkeley EECS ½ÌÊÚ¡¢Sky Computing Lab / RISE Lab ÅäºÏÈÏÕæÈË£¬£¬£¬²¢ÊôÓÚ BAIR Éú̬¡£¡£¡£¡£¡£Õâʹ WorldModelBench ²»µ«ÊÇÒ»¸öÊÓƵÆÀ·Ö¼¯£¬£¬£¬¶øÊÇϵͳ¡¢ÊÓ¾õ¡¢»úеÈ˺͸ßЧ AI ÍŶÓÅäºÏÍƶ¯µÄÌìÏÂÄ£×ÓÆÀ²â¡£¡£¡£¡£¡£

ΪʲôֵµÃ¹Ø×¢

ËüÊÇ×îÊÊÏàÖúΪ¡°Í¨ÓÃÊÓƵÌìÏÂÄ£×ÓÈëÃÅ»ù×¼¡±µÄ benchmark¡£¡£¡£¡£¡£ÓÉÓÚËüûÓÐֻͣÁôÔÚÉóÃÀºÍÇåÎú¶È£¬£¬£¬¶øÊÇ°ÑÌìÏÂÄ£×Ó²ð³ÉÓ¦ÓÃÓò¡¢ÎïÀí¼ÍÂɺÍϸÁ£¶È¹ýʧÀàÐÍ£¬£¬£¬ÌØÊâÊʺÏÅжÏÒ»¸öÊÓƵģ×Ó¾¿¾¹ÊÇÔÚ¡°ÌìÉúÏñÌìϵÄͼÏñ¡±£¬£¬£¬ÕÕ¾ÉÔÚ¡°Ä£ÄâÒ»¸öÄÜ×ÔÇ¢ÑÝ»¯µÄÌìÏ¡±¡£¡£¡£¡£¡£

ÏîÄ¿£ºhttps://worldmodelbench-team.github.io/

ÂÛÎÄ£ºhttps://arxiv.org/abs/2502.20694

´úÂëÓëÆÀ²â£ºhttps://github.com/WorldModelBench-Team/WorldModelBench

²âÊÔÊý¾Ý£ºhttps://huggingface.co/WorldModelBench-Team

2. WorldScore

Ò»¾ä»°ÏÈÈÝ£º WorldScore °Ñ 3D¡¢4D¡¢Í¼ÉúÊÓƵ¡¢ÎÄÉúÊÓƵģ×ӷŵ½Í³Ò»¸ö¡°ÏÂÒ»³¡¾°ÌìÉú¡±¿ò¼ÜÀïÆÀ²â£¬£¬£¬Öص㿴ģ×ÓÄܲ»¿É°´Ïà»ú¹ì¼£Ò»Á¬ÌìÉúÌìÏ¡£¡£¡£¡£¡£

¸ÅÊö£º WorldScore ½â¾öµÄÊÇ¡°²î±ðÌìÏÂÌìÉúõ辶ÎÞ·¨¹«Õý½ÏÁ¿¡±µÄÎÊÌâ¡£¡£¡£¡£¡£3D Ä£×Ó¡¢4D Ä£×ÓºÍÊÓƵģ×ÓÊä³öÐÎ̬²î±ð£¬£¬£¬ÒÑÍùºÜÄÑÔÚͳһÕÅÄÚÍâ½ÏÁ¿¡£¡£¡£¡£¡£WorldScore ½«ÌìÏÂÌìÉú²ð³ÉһϵÁлùÓÚÃ÷È·Ïà»ú¹ì¼£µÄ next-scene generation ʹÃü£¬£¬£¬Óà 3000 ¸öÑùÀýÁýÕÖ¾²Ì¬/¶¯Ì¬¡¢ÊÒÄÚ/ÊÒÍ⡢дʵ/Æø¸Å»¯µÈ³¡¾°£¬£¬£¬²¢´Ó¿É¿ØÐÔ¡¢ÖÊÁ¿¡¢¶¯Ì¬Èý·½Ãæ´ò·Ö¡£¡£¡£¡£¡£¿£¿£¿£¿£¿É¿ØÐÔ¿´Ïà»ú¿ØÖÆ¡¢ÎïÌå¿ØÖƺÍÄÚÈݶÔÆ룬£¬£¬ÖÊÁ¿¿´ 3D Ò»ÖÂÐÔ¡¢¹â¶ÈÒ»ÖÂÐÔ¡¢Æø¸ÅÒ»ÖÂÐÔºÍÖ÷¹ÛÖÊÁ¿£¬£¬£¬¶¯Ì¬¿´Ô˶¯×¼È·ÐÔ¡¢·ù¶ÈºÍÁ÷ͨ¶È¡£¡£¡£¡£¡£¹Ù·½Ð§¹ûÏÔʾ£¬£¬£¬Voyager¡¢WonderWorld¡¢LucidDreamer µÈ 3D/³¡¾°ÌìÉúÄ£×ÓÔÚ¾²Ì¬ÌìÏÂÌìÉúÉϺÜÇ¿£¬£¬£¬CogVideoX-I2V µÈÊÓƵģ×ÓÔÚ¶¯Ì¬·ÖÊýÉÏÌåÏÖÍ»³ö£»£»£»£»£»£»Í¬Ê±£¬£¬£¬T2V Ä£×Óͨ³£¸üÈÝÒ×±»Îı¾¿ØÖÆ£¬£¬£¬I2V Ä£×ÓÍùÍù»­ÖʸüÎÈ¡£¡£¡£¡£¡£

WorldScore ÊÇ Stanford University ÍŶÓÐû²¼£¬£¬£¬ÂÛÎÄ×÷ÕßΪ Haoyi Duan¡¢Hong-Xing Yu¡¢Sirui Chen¡¢Li Fei-Fei¡¢Jiajun Wu¡£¡£¡£¡£¡£Li Fei-Fei ÊÇ Stanford ÅÌËã»úϵ Sequoia Professor¡¢Stanford HAI Ê×´´ÁªºÏÖ÷ÈΣ¬£¬£¬ÔøÈÎ Stanford AI Lab Ö÷ÈÎºÍ Google Cloud AI/ML Ê×ϯ¿Æѧ¼Ò£¬£¬£¬¿ª°ìWorld Labs£¬£¬£¬Ò²ÊÇ ImageNet µÄÒªº¦Íƶ¯Õߣ»£»£»£»£»£»Jiajun Wu ÊÇ Stanford ÅÌËã»úϵÖúÀí½ÌÊÚ£¬£¬£¬Í¬Ê±¼æÈÎÐÄÀíѧƫÏò£¬£¬£¬Ñо¿ÅÌËã»úÊÓ¾õ¡¢»úеÈ˺ÍÅÌËãÈÏÖª¿Æѧ£¬£¬£¬ËùÔÚÆ«Ïò×ÔÈ»Ìù½ü¡°ÌìÏÂÔõÑù±»ÌåÏÖ¡¢ÍÆÀíºÍÌìÉú¡±¡£¡£¡£¡£¡£Òò´Ë£¬£¬£¬WorldScore ¸üÏñÊÇ Stanford ÊÓ¾õѧϰ¡¢3D/ÎïÀí³¡¾°Ã÷È·ºÍÌìÉúÄ£×ÓÑо¿ÂöÂçϵÄÒ»Ì×ÌìÏÂÌìÉú»³±§¡£¡£¡£¡£¡£

ΪʲôֵµÃ¹Ø×¢£º

ÈôÊÇ˵ WorldModelBench ¸üÏñ¡°ÊÓƵģ×ÓÊDz»ÊÇÌìÏÂÄ£×Ó¡±µÄÌå¼ì£¬£¬£¬WorldScore Ôò¸üÏñ¡°ÌìÏÂÌìÉúõ辶´óºáÆÀ¡±¡£¡£¡£¡£¡£Ëü°Ñ 3D¡¢4D¡¢ÊÓƵÈýÌõÊÖÒÕõ辶·Å½øͳһЭÒ飬£¬£¬¶Ô´´Òµ¹«Ë¾ºÍÑо¿ÍŶÓÅжÏÊÖÒÕõ辶ºÜÓмÛÖµ£º¾¿¾¹¸ÃѺÊÓƵÌìÉú¡¢3D ÌìÉú£¬£¬£¬Õվɸü»ìÏýµÄ 4D ÌìÏÂÌìÉú¡£¡£¡£¡£¡£

ÏîÄ¿Ò³£ºhttps://haoyi-duan.github.io/WorldScore/

ÂÛÎÄ£ºhttps://arxiv.org/abs/2504.00983

´úÂ룺https://github.com/haoyi-duan/WorldScore

Êý¾Ý¼¯£ºhttps://huggingface.co/datasets/haoyi-duan/WorldScore

3. WorldLens

Ò»¾ä»°ÏÈÈÝ£º WorldLens רÃÅÆÀ¹À×Ô¶¯¼ÝÊ»ÌìÏÂÄ£×Ó£¬£¬£¬½¹µãÎÊÌâÊÇ£ºÌìÉúµÄ¼ÝÊ»Ììϲ»µ«ÒªÏñ£¬£¬£¬»¹Òª¼¸ºÎÎȹ̡¢ÎïÀíºÏÀí¡¢ÐÐΪÇå¾²¡¢¶ÔÏÂÓÎʹÃüÓÐÓᣡ£¡£¡£¡£

WorldLens ½â¾öµÄÊÇ×Ô¶¯¼ÝÊ»³¡¾°Àï¡°ÊÓ¾õ±ÆÕ浫¹¦Ð§Ê§Õ桱µÄÎÊÌâ¡£¡£¡£¡£¡£¼ÝÊ»ÌìÏÂÄ£×ÓºÜÈÝÒ×ÌìÉú¿´ÆðÀ´ºÏÀíµÄ 4D ³¡¾°£¬£¬£¬µ«Ò»µ©Çе½ÖØÐÞ¡¢±Õ»·¿ØÖÆ¡¢ÏÂÓμì²âÖ§½â£¬£¬£¬¾Í»á̻¶¼¸ºÎƯÒÆ¡¢Ô˶¯²»Îȹ̡¢Åöײ¡¢Æ«Àëõ辶µÈÎÊÌâ¡£¡£¡£¡£¡£WorldLens ÓÃÎå¸ö»¥²¹Æ«ÏòÆÀ²â£ºGeneration¡¢Reconstruction¡¢Action-Following¡¢Downstream Task¡¢Human Preference£¬£¬£¬²¢½øÒ»²½¹¹½¨ WorldLens-26K ÈËÀàÆ«ºÃÊý¾ÝºÍ WorldLens-Agent ÊÓ¾õÓïÑÔÆÀÉóÆ÷¡£¡£¡£¡£¡£ÏîÄ¿½áÂÛºÜÊÇÏÖʵ£ºÃ»Óмòµ¥Ä£×ÓÄÜÔÚËùÓÐά¶ÈͳÖΰñµ¥£¬£¬£¬ÎÆÀíÕæʵµÄÄ£×Ó¿ÉÄÜÎïÀíºÍ¼¸ºÎ²»ÎÈ£¬£¬£¬¼¸ºÎÎȹ̵ÄÄ£×ÓÓÖ¿ÉÄÜÐÐΪϸ½Úȱ·¦¡£¡£¡£¡£¡£ÂÛÎÄʵÑéÖУ¬£¬£¬DiST-4D ÔÚÐÂÊÓ½ÇÖÊÁ¿ºÍ²¿·ÖÏÂÓÎʹÃüÉÏÌåÏÖÇ¿£¬£¬£¬OpenDWM µÈÄ£×ÓÔÚ¼¸ºÎÎȹÌÐÔÉÏÓÐÓÅÊÆ£¬£¬£¬DriveDreamer-2 Ò²ÔÚʹÃüÏà¹ØÖ¸±êÖнøÈëÇ¿ÊÆÌݶÓ¡£¡£¡£¡£¡£

WorldLens ÓÉ WorldBench Team Ðû²¼£¬£¬£¬NUS¡¢University of Macau¡¢USTC¡¢Zhejiang University¡¢NTU¡¢Horizon Robotics¡¢HUST¡¢TUM¡¢Fudan University¡¢Shanghai AI Lab¡¢A*STAR¡¢CNRS µÈ¡£¡£¡£¡£¡£½¹µã×÷Õß°üÀ¨ Ao Liang¡¢Lingdong Kong¡¢Tianyi Yan¡¢Hongsi Liu¡¢Ziqi HuangµÈ¡£¡£¡£¡£¡£Lingdong Kong ÊÇ NUS ÅÌËã»úϵ²©Ê¿Éú¡¢ÏîĿǣͷÈË£¬£¬£¬ºã¾Ãά»¤×Ô¶¯¼ÝÊ»ºÍ 3D/4D ÌìϽ¨Ä£Ïà¹ØÏîÄ¿£»£»£»£»£»£»Ziwei Liu ÊÇ NTU MMLab µÄ Associate Professor / Provost's Chair in AI£¬£¬£¬Ñо¿ÁýÕÖÅÌËã»úÊÓ¾õ¡¢»úеѧϰºÍÅÌËã»úͼÐΣ»£»£»£»£»£»Liang Pan ÊÇ Shanghai AI Lab Ñо¿¿Æѧ¼Ò£¬£¬£¬ºã¾Ã×ö 3D ÊÓ¾õ¡¢world models ºÍ embodied AI£»£»£»£»£»£»Wei Tsang Ooi ÊÇ NUS ¶àýϵһÇÐÆ«Ïò½ÌÊÚ¡£¡£¡£¡£¡£Õâ¸ö×éºÏÈà WorldLens ͬʱվÔÚ×Ô¶¯¼ÝÊ»¡¢4D ³¡¾°¡¢3D ÊÓ¾õºÍÈËÀàÆ«ºÃÆÀ²âµÄ½»Ö¯µãÉÏ¡£¡£¡£¡£¡£

ΪʲôֵµÃ¹Ø×¢

×Ô¶¯¼ÝÊ»ÊÇÌìÏÂÄ£×Ó×îÈÝÒ×ÂäµØ¡¢Ò²×î²»¿ÉÖ»¿´ demo µÄÁìÓò¡£¡£¡£¡£¡£WorldLens µÄ¼ÛÖµÔÚÓÚ°Ñ¡°ÌìÉúÌìÏ¡±À­»Øµ½ÁËÇå¾²¡¢±Õ»·ºÍÏÂÓÎʹÃüÉÏ£ºÒ»¸öÄ£×ÓÌìÉúµÄÊÓƵÔÙƯÁÁ£¬£¬£¬ÈôÊÇ»áÈøÐ֪ģ×ÓÍË»¯£¬£¬£¬»òÕßÈà planner ÔÚÄ£ÄâÌìÏÂÀïײ³µ£¬£¬£¬Ëü¾Í²»¿ÉËã¿É¿¿µÄ¼ÝÊ»ÌìÏÂÄ£×Ó¡£¡£¡£¡£¡£

ÏîÄ¿Ò³£ºhttps://worldbench.github.io/worldlens

ÂÛÎÄ£ºhttps://arxiv.org/abs/2512.10958

´úÂ룺https://github.com/worldbench/WorldLens

Leaderboard£ºhttps://huggingface.co/spaces/worldbench/worldlens

4. WorldArena 2.0

Ò»¾ä»°ÏÈÈÝ£º WorldArena ÆÀ¹À¾ßÉíÌìÏÂÄ£×ÓÊÇ·ñÕæµÄÄÜ·þÎñ»úеÈËʹÃü£¬£¬£¬¶ø²»µ«½öÊÇÌìÉúÒ»¶Î¿´ÆðÀ´²»´íµÄ²Ù×÷ÊÓƵ¡£¡£¡£¡£¡£

WorldArena ½â¾öµÄÊǾßÉíÖÇÄÜÀï¡°¸ÐÖªÖÊÁ¿ºÍ¹¦Ð§¼ÛÖµÍѽڡ±µÄÎÊÌâ¡£¡£¡£¡£¡£Ðí¶à embodied world model ¿ÉÒÔÌìÉú¸ßÖÊÁ¿ÊÓƵ£¬£¬£¬µ«»úеÈËÕæÕýÌåÌùµÄÊÇ£ºËüÄܲ»¿ÉºÏ³ÉÓÐÓÃѵÁ·Êý¾Ý£¿£¿£¿£¿£¿Äܲ»¿É×÷Ϊ policy evaluator ÆÀÔ¤Õ½ÂÔ£¿£¿£¿£¿£¿Äܲ»¿É×ÊÖú agent ×ö action planning£¿£¿£¿£¿£¿WorldArena ÏÈÓà 16 ¸öÖ¸±êÆÀ²âÊÓƵ¸ÐÖªÖÊÁ¿£¬£¬£¬ÁýÕÖÊÓ¾õÖÊÁ¿¡¢Ô˶¯ÖÊÁ¿¡¢ÄÚÈÝÒ»ÖÂÐÔ¡¢ÎïÀí×ñÊØ¡¢3D ׼ȷÐÔ¡¢¿É¿ØÐÔ 6 ¸ö×Óά¶È£¬£¬£¬ÔÙÆÀ¹ÀÈýÀà¾ßÉí¹¦Ð§£ºData Engine¡¢Policy Evaluator¡¢Action Planner£¬£¬£¬²¢Ìá³ö EWMScore ×öͳһ»³±§¡£¡£¡£¡£¡£

WorldArena 2.0 ÔÚÕâ¸ö»ù´¡ÉϼÌÐøÀ©Õ¹£ºµÚÒ»£¬£¬£¬Ä£Ì¬´Ó vision-only À©Õ¹µ½ visuo-tactile£¬£¬£¬ÈÃÄ£×Ó±ØÐè´¦Àí½Ó´¥¡¢Á¦¡¢»¬¶¯ºÍÖÊÁϽ»»¥£»£»£»£»£»£»µÚ¶þ£¬£¬£¬¹¦Ð§´ÓÀëÏßÆÀ¹ÀÀ©Õ¹µ½ online RL£¬£¬£¬°ÑÌìÏÂÄ£×Ó¿´³É¿É½»»¥ÇéÐÎÀ´ÑµÁ·ºÍË¢ÐÂÕ½ÂÔ£»£»£»£»£»£»µÚÈý£¬£¬£¬Æ½Ì¨´Ó simulator-only À©Õ¹µ½ RoboTwin 2.0¡¢LIBERO ºÍ AgileX split-type ALOHA Õæʵ»úеÈË¡£¡£¡£¡£¡£ËüµÄ½áÂÛÒ²¸üÌù½ü¹¤ÒµÎÊÌ⣺ģÄâÇéÐÎÀïµÄºÃЧ¹û²»¼´ÊÇÕæʵ°²ÅÅÄÜÁ¦£¬£¬£¬sim-to-real gap ÈÔÈ»ÊǾßÉíÌìÏÂÄ£×Ó±ØÐè¿ç¹ýµÄÓ²Ãż÷¡£¡£¡£¡£¡£

¶à»ú¹¹ÁªºÏÁªºÏÐû²¼£ºTsinghua University¡¢Shanghai Jiao Tong University¡¢The University of Hong Kong¡¢Princeton University¡¢Chinese Academy of Sciences¡¢University of Science and Technology of China¡¢Peking University¡¢National University of Singapore¡£¡£¡£¡£¡£×÷Õß°üÀ¨ Yu Shang¡¢Zhuohang Li¡¢Yiding Ma¡¢Weikang Su¡¢Xin JinµÈ¡£¡£¡£¡£¡£Yong Li ÊÇÇ廪 FIB Lab Ö÷ÒªÈÏÕæÈË£¬£¬£¬ºã¾Ã×ö AI¡¢Êý¾ÝÍÚ¾òºÍ¶¼»áÖÇÄÜ£»£»£»£»£»£»Jun Zhu ÊÇÇ廪 Bosch AI Professor¡¢IEEE/AAAI Fellow£¬£¬£¬Ñо¿»úеѧϰ¡¢Éî¶ÈÌìÉúÄ£×Ó¡¢Ç¿»¯Ñ§Ï°ºÍ·´¿¹Â³°ôÐÔ£»£»£»£»£»£»Wenwu Zhu ÊÇÇ廪ÅÌËã»úϵ½ÌÊÚ£¬£¬£¬ÔøÔÚ Microsoft Research Asia¡¢Intel Research China µÈ¹¤ÒµÑо¿»ú¹¹µ£µ±Ö÷Òª½ÇÉ«£»£»£»£»£»£»Tat-Seng Chua ÊÇ NUS School of Computing µÄ KITHCT Chair Professor£¬£¬£¬Ò²ÔøÈÎ NUS Computing ´´ÔºÔº³¤¡£¡£¡£¡£¡£Õâ¸öÕóÈÝÈà WorldArena ¸üÏñ¾ßÉíÌìÏÂÄ£×ÓÁìÓòµÄ¡°Ñ§Êõ-»úеÈË-¶àýÌå-ϵͳ¡±ÁªºÏÆÀ²â¡£¡£¡£¡£¡£

ΪʲôֵµÃ¹Ø×¢

Ëü°ÑÌìÏÂÄ£×Ó´Ó¡°»á²»»áÌìÉúδÀ´Ö¡¡±Íƽøµ½¡°Äܲ»¿ÉÈûúеÈ˱äÇ¿¡±¡£¡£¡£¡£¡£Õâ¶Ô¾ßÉíÖÇÄܺÜÊÇÒªº¦£¬£¬£¬ÓÉÓÚÕæÕýµÄ»úеÈËÌìÏÂÄ£×Ó²»µ«ÊÇÄ£Äâ»­Ã棬£¬£¬¶øÊÇÒª³ÉΪÊý¾ÝÒýÇæ¡¢ÇéÐÎÊðÀíºÍÍýÏëÄ£¿£¿£¿£¿£¿é¡£¡£¡£¡£¡£WorldArena ÊÇÏÖÔÚÉÙÊý°Ñ perception ºÍ functional utility ·ÅÔÚͳһ¸ö¿ò¼ÜÀïÑÏËàÆÀ¹ÀµÄ benchmark¡£¡£¡£¡£¡£

ÏîÄ¿Ò³£ºhttps://world-arena.ai/

ÂÛÎÄ£ºhttps://arxiv.org/abs/2602.08971

´úÂ룺https://github.com/tsinghua-fib-lab/WorldArena

Leaderboard£ºhttps://huggingface.co/spaces/WorldArena/WorldArena

5. Physics-IQ / Physics-IQ Verified

Ò»¾ä»°ÏÈÈÝ£º Physics-IQ ÓÃÕæʵÎïÀíʵÑéÊÓƵ¿¼ÊÓƵÌìÉúÄ£×Ó£¬£¬£¬Ö±½Ó×·ÎÊÒ»¸ö»ù´¡ÎÊÌ⣺ģ×ÓÊÇÔÚÃ÷È·ÎïÀí£¬£¬£¬ÕÕ¾ÉÖ»ÊÇÔÚÕ¹ÍûÏñËØ£¿£¿£¿£¿£¿

Physics-IQ ½â¾öµÄÊÇ¡°ÊÓ¾õÕæʵ²»¼´ÊÇÎïÀíÃ÷È·¡±µÄÎÊÌâ¡£¡£¡£¡£¡£ËüÒªÇóÄ£×Óƾ֤ÕæʵÌìÏÂÎïÀíÊÓƵµÄÇ°°ë¶ÎÕ¹ÍûºóÐøÑÝ»¯£¬£¬£¬ÁýÕÖ¹ÌÌåÁ¦Ñ§¡¢Á÷Ì嶯Á¦Ñ§¡¢¹âѧ¡¢ÈÈÁ¦Ñ§¡¢´ÅѧÎåÀàÎïÀíÕ÷Ï󡣡£¡£¡£¡£ÆÀ²â²»¿´ÊÓƵÊÇ·ñ¡°¿´ÆðÀ´¸ß¼¶¡±£¬£¬£¬¶øÊÇ¿´Ä£×ÓÕ¹ÍûµÄÎïÀíת±äÔڿռ䡢ʱ¼äºÍÔ˶¯ÇøÓòÉÏÊÇ·ñ¿¿½üÕæʵЧ¹û¡£¡£¡£¡£¡£ÂÛÎÄÆÀ¹ÀÁË Sora¡¢Runway¡¢Pika¡¢Lumiere¡¢Stable Video Diffusion¡¢VideoPoet µÈÄ£×Ó£¬£¬£¬½áÂÛÏ൱ֱ½Ó£ºÄ¿½ñÊÓƵÌìÉúÄ£×ÓµÄÎïÀíÃ÷È·ÈÔÈ»ÓÐÏÞ£¬£¬£¬²¢ÇÒÓëÊÓ¾õÕæʵ¸Ð²¢²»Ç¿Ïà¹Ø£»£»£»£»£»£»¶àÖ¡Ìõ¼þͨ³£±Èµ¥Ö¡Ìõ¼þ¸üÓÐ×ÊÖú£¬£¬£¬Á÷ÌåÕ÷ÏóÏà¶Ô¸üÈÝÒ×£¬£¬£¬¹ÌÌåÁ¦Ñ§¸üÄÑ¡£¡£¡£¡£¡£2026 Äê 6 ÔÂÓÖ·ºÆðÁË Physics-IQ Verified£¬£¬£¬¶ÔԭʼÑù±¾ºÍ prompt ×öÁËϵͳÉó¼ÆºÍÐÞ¶©£»£»£»£»£»£»ÈôÊÇÊÇÏÖʵÅÜÆÀ²â£¬£¬£¬ÓÅÏȹØ×¢ Verified °æ±¾»á¸üÎÈ¡£¡£¡£¡£¡£

Physics-IQ ÓÉ Saman Motamed¡¢Laura Culp¡¢Kevin Swersky¡¢Priyank Jaini¡¢Robert Geirhos Ðû²¼£¬£¬£¬ÂÛÎÄÊ×Ò³±ê×¢ Saman Motamed À´×Ô INSAIT / Sofia University£¬£¬£¬Ïà¹ØÊÂÇéÔÚ Google DeepMind ʱ´úÍê³É£¬£¬£¬ÆäÓà×÷ÕßÀ´×Ô Google DeepMind¡£¡£¡£¡£¡£INSAIT ¹Ù·½±¨µÀÒ²Ã÷È·³Æ¸Ã benchmark ÓÉ INSAIT Óë Google DeepMind ÁªºÏ¿ª·¢£¬£¬£¬²¢ÔÚ ICCV 2025 Òý·¢¹Ø×¢¡£¡£¡£¡£¡£Saman Motamed ÊÇ INSAIT µÄ ELLIS PhD student£¬£¬£¬Ñо¿ÅÌËã»úÊÓ¾õ¡¢ÌìÉúÄ£×ÓºÍ intuitive physics£»£»£»£»£»£»Robert Geirhos ÊÇ Google DeepMind Staff Research Scientist£¬£¬£¬ºã¾Ã×öÊÓ¾õ³°ôÐÔ¡¢¿ÉÚ¹ÊÍÐÔºÍÈË»úÊÓ¾õ½ÏÁ¿£¬£¬£¬´ú±íÊÂÇé°üÀ¨Éî¶ÈÍøÂç texture bias / shortcut learning Æ«Ïò£»£»£»£»£»£»Priyank Jaini ºÍ Kevin Swersky Ò²¶¼À´×Ô Google DeepMind µÄÌìÉúÄ£×ÓÓë»úеѧϰÑо¿Ïß¡£¡£¡£¡£¡£Õâ¸öÍŶӵÄÓÅÊÆÊÇ°ÑÕæʵÎïÀíʵÑé¡¢ÊÓ¾õÐÄÀíѧʽÎÊÌâÒâʶºÍÇ°ÑØÊÓƵģ×ÓÑо¿·Åµ½ÁËÒ»Æ𡣡£¡£¡£¡£

ΪʲôֵµÃ¹Ø×¢

ÎïÀíÊÇÌìÏÂÄ£×ÓÐðÊÂÀï×îÄÑÈÆ¿ªµÄµ×²ãÄÜÁ¦¡£¡£¡£¡£¡£Ò»¸öÄ£×Ó¿ÉÒÔÌìÉúÓ°Ï·¸ÐË®»¨¡¢±¬Õ¨ºÍÅöײ£¬£¬£¬µ«ÈôÊÇËüÎÞ·¨Õ¹ÍûÕæʵÎïÌå»áÔõÑù¼ÌÐøÔ˶¯£¬£¬£¬Ëü¾ÍºÜÄѳÉΪ¿É¿¿µÄÌìÏÂÄ£×Ó¡£¡£¡£¡£¡£Physics-IQ µÄ¼ÛÖµÔÚÓÚËü×ã¹»Ö±½Ó£ºÉÙ½²¡°Ïñ²»Ïñ¡±£¬£¬£¬¶àÎÊ¡°¶Ô²î³Ø¡±¡£¡£¡£¡£¡£

ÏîÄ¿Ò³£ºhttps://physics-iq.github.io/

ÂÛÎÄ£ºhttps://arxiv.org/abs/2501.09038

´úÂ룺https://github.com/google-deepmind/physics-iq-benchmark

Physics-IQ Verified£ºhttps://arxiv.org/abs/2606.18943

ÌìÏÂÄ£×ÓÏÖÔÚ´¦ÔÚÒ»¸ö¿´·¨¿ìËÙÉýΡ¢ÊÖÒÕõ辶·Ö²æ¡¢ÆÀ²âϵͳ׷׏¤ÒµÐðʲ¹¿ÎµÄÔÓÂÒÆÚ¡£¡£¡£¡£¡£Ò²ÕýÓÉÓÚÔÓÂÒ£¬£¬£¬benchmark ²ÅÖµµÃ¿´£ºËüÃÇÊÇÕⳡ½ç˵ÕùÈ¡À£¬£¬ÉÙÊýÄÜ°Ñ¿ÚºÅÂäµ½²âÊÔÌâÉϵŤ¾ß¡£¡£¡£¡£¡£

Benchmark ¿ÉÒÔ¸øÒ»¸ö»¹ºÜËÉÉ¢µÄ¿´·¨»­½çÏß¡£¡£¡£¡£¡£WorldModelBench ¿¼²ìÊÓƵģ×ÓÓÐûÓÐ֪ʶºÍÎïÀíÔ¼Êø£»£»£»£»£»£»WorldScore Ȩºâ²î±ðÌìÏÂÌìÉúõ辶Äܲ»¿Éͳһ½ÏÁ¿£»£»£»£»£»£»WorldLens ¿ÉÒÔÓÃÀ´ÆÀ¹À¼ÝÊ»ÌìÏÂÊÇ·ñÕæµÄ¶ÔÇå¾²ºÍÏÂÓÎʹÃüÓÐÓ㻣»£»£»£»£»WorldArena ¿´µÄÊǾßÉíÌìÏÂÄ£×ÓÊÇ·ñÕæµÄÄÜ·þÎñ»úеÈË£»£»£»£»£»£»Physics-IQ Ôò°Ñ½çÏßѹµ½×îµ×²ã£¬£¬£¬ÎÊÄ£×Ó¾¿¾¹ÊÇÔÚÃ÷È·ÎïÀí£¬£¬£¬ÕÕ¾ÉÖ»ÊÇÔÚ²¹ÏñËØ¡£¡£¡£¡£¡£

ÌìÏÂÄ£×ÓÉú³¤µ½½ñÌ죬£¬£¬ÆÀ²âµÄÖ÷ÒªÐÔÕýÔÚÈÕÒæ͹ÏÔ£¬£¬£¬ÓïÑÔÄ£×ÓÒѾ­¸üÔç֤ʵÎúÕâ¼þÊ£¬£¬£¬½è¼øÓïÑÔÄ£×ÓµÄÂÄÀú£¬£¬£¬ÔçÈÕÈ·¶¨½¹µãÆÀ²âÆ«Ïò£¬£¬£¬Ò²½«³ÉΪδÀ´¾ºÕùµÄÒªº¦¡£¡£¡£¡£¡£

µã¸ö¡°°®ÐÄ¡±£¬£¬£¬ÔÙ×ß °É

ÏÄÈÕÒÑÖÁ ºþ±±¶÷ʩΪÄã½âËø×ÔÈ»Á¹¾³

¿­Ê±AG

·Æ×Üͳ½ôÆÈ×÷·Ï¶à¸ö¹ûÕæÐгÌ