åäººã«æç©ããŠã«ãããªã³ãŽãåã£ãŠæž¡ããŠãããããé Œããšãããæ³åããŠãã ãããå人ã¯ããŠã«ãèŠãŠãã©ãããªã³ãŽãã倿ããã€ã¶ããªãçšåºŠã®åå æžã§æã¡äžããŠãããªãã«ææž¡ããŸãã
ã§ã¯åãããšããã£ãããããã«é Œãã ãã©ãã§ãããããªã³ãŽã®æã¡æ¹ãã©ã®æã䜿ãã¹ãããããããã®åå æžãŸã§æ£ç¢ºã«èª¬æã¯ã§ããŸãããå®éã«æã¡äžããããšã¯ã§ããŸãããæããªãããªã³ãŽã®è§Šæãããããªãããã§ãã
ãšã³ããã£ãAI ã¯ããã®ã®ã£ãããåããããšããåéã§ããç°¡åã«èšãã°ãç©ççãªèº«äœïŒãããããè»ããããŒã³ãªã©ïŒãæã¡ãåã«æç« ãèªãã®ã§ã¯ãªããçŸå®äžçã§è¡åããããšã§åŠç¿ããAIã§ããåãèãæ¹ãæãçšèªãšããŠããã£ãžã«ã«AIãããã䜿ãããŸãã
ããŸããã®åéã¯å€§ããªè»¢æ©ãè¿ããŠããŸãã2026幎1æã®CESã§ãNVIDIAã®Jensen Huangã¯ãããããããã£ã¯ã¹ã«ãããChatGPTã®ç¬éã ãšè¡šçŸããŸããã
è³éãæµã蟌ãã§ããŸããCNBC ãå ±ããDealroomã®ããŒã¿ã«ãããšããããã£ã¯ã¹äŒæ¥ã¯2026幎6æåæ¬ãŸã§ã«558åãã« ã調éããéå»ã®å¹Žéèšé²ãã»ãŒåå¢ãããŸããã
æ¬èšäºã§ã¯æ¬¡ã®å
å®¹ãæ±ããŸãã
ãšã³ããã£ãAIãšã¯äœããLLMãå€å
žçãããã£ã¯ã¹ãšã®æ¯èŒ
ãªããã£ãžã«ã«AIã¯ããžã¿ã«AIãããã¯ããã«é£ããã®ã
ç¥èŠã»æšè«ã»è¡åã«ãŒãã«ãããšã³ããã£ãAIã®ä»çµã¿
ãããããã·ãã¥ã¬ãŒã·ã§ã³ã§èšç·Žãããæ¹æ³ãšãã·ã ã»ãã¥ã»ãªã¢ã«ã®ã£ãããšã¯äœã
2026幎ã«ããããšã³ããã£ãAIã®æŽ»çšåéãšæ³šç®äŒæ¥
ããããã¹ãŠãããŒã¿ãµã€ãšã³ãã£ã¹ãã«ãšã£ãŠæå³ããããš
ããããã«è§ŠããããšããªããŠãå¿é
ã¯ãããŸãããæ©æ¢°åŠç¿ã«å€å°ã®éŠŽæã¿ãããã°å©ãã«ãªããŸãããã©ã¡ãã«ããåºç€ããé ã远ã£ãŠè§£èª¬ããã®ã§ãç¡çãªãèªã¿é²ããããŸãã
ãšã³ããã£ãAIã®èŠç¹
ãšã³ããã£ãAIã¯ããããããè»ããããŒã³ã®ãããªç©ççãªèº«äœã«çµã¿èŸŒãŸããããã¹ããç»åã ãã§ãªãå®äžçã§è¡åããããšã§åŠç¿ããAIã§ãã
æå€§ã®ããã«ããã¯ã¯ããŒã¿ã§ããæå€§çŽã®ãªãŒãã³ãªãããã£ã¯ã¹ããŒã¿ã»ããã®1ã€ã§ããOpen X-Embodimentã§ãããçŸå®ã®è»è·¡ã¯100äžè¶
ã«éããŸãããäžæ¹ã§LLMã¯å
åäœã®ããŒã¯ã³ã§åŠç¿ããŸãã
ããŒã ã¯ã·ãã¥ã¬ãŒã·ã§ã³ããã¡ã€ã³ã©ã³ãã åãäºååŠç¿æžã¿ã®ããžã§ã³ã»èšèªããã¯ããŒã³ã§ãããè£ããŸãã
2026幎æç¹ã§ã¯ãå庫ãããã¿ã¯ã·ãŒã§åæã®æ¬çªéçšãé²ã¿ããã¥ãŒããã€ãã®å€ãã¯éå®çãªãã€ãããæ®µéã«ãããŸãã
ãšã³ããã£ãAIãšã¯ïŒ
ãšã³ããã£ãAIã¯ããããããèªåé転è»ããããŒã³ã®ãããªç©ççãªèº«äœã«çµã¿èŸŒãŸããã»ã³ãµãŒã§ç°å¢ãç¥èŠ ããäœãããããæšè« ããã¢ãŒã¿ãŒãéããŠäžçã«äœçš ããèªãã®è¡åã®çµæããåŠç¿ãã人工ç¥èœã·ã¹ãã ã§ãã
ããã§ã®éµã¯ãšã³ããã£ãïŒèº«äœæ§ïŒ ã§ãã
å€ãã®AIã¢ãã«ã¯ãä»è
ãåéããããŒã¿ãèŠ³å¯ ããŠåŠã³ãŸããããã«å¯Ÿãããšã³ããã£ããªã·ã¹ãã ã¯ãã³ãããæŒããŠãã³ãããæ»ãã®ã芳å¯ããæŒãããšãã«ã³ãããã©ãæ¯ãèãããšããç¥èãæŽæ°ããããšãã£ãå
·åã«ãç°å¢ãšçžäºäœçš ããŠåŠã³ãŸãã
äŸãæããŸããããæ°çŸäžæã®ãã¢ã®åçã§åŠç¿ããããžã§ã³ã¢ãã«ã¯ããã¢ãã©ã®ããã«èŠããããç¥ã£ãŠããŸããäžæ¹ã§ãå®éã«500æã®ãã¢ãéããããšããããããã¯ãæŒããã¢ãããã°åŒããã¢ããããéããã®ãããã°ãå
ã«ãã³ãã«ãæŒãäžããå¿
èŠããããã®ãããããšç¥ã£ãŠããŸãã
ãã®ã¬ãã«ã®ç¥èã¯åçããã¯åŸãããŸããã
ãã衚çŸããã®ã奜ãã§ããå€ãã®AIã¯äžçã«ã€ããŠãèªãã§ãåŠã¶ãããšã³ããã£ãAIã¯äžçã«ãè§Šããäœéšãããããšã§åŠã¶ ã®ã§ãã
ãã®éãããå¿
èŠãªããŒã¿ãåŠç¿æ¹æ³ããããŠæ³å®å€ã®äºæ
ã®èµ·ããæ¹ãäžå€ãããŸãã
ãšã³ããã£ãAI vs. å€§èŠæš¡èšèªã¢ãã« vs. åŸæ¥åãããã£ã¯ã¹
ãããŸã§ããšã³ããã£ãAIããã£ããããããšæ¯ã¹ãŠããŸãããæ¬¡ã«ã人ã
ãããæ··åãã2ã€ãããªãã¡å€§èŠæš¡èšèªã¢ãã«ïŒLLMïŒãšåŸæ¥ã®ã«ãŒã«ããŒã¹ãããã£ã¯ã¹ãšæ¯èŒããŠã¿ãŸãããã
 ãšã³ããã£ãAI å€§èŠæš¡èšèªã¢ãã«ïŒLLMïŒ åŸæ¥ã®ã«ãŒã«ããŒã¹ãããã£ã¯ã¹ ç°å¢ããåŠç¿ãã ã¯ããçžäºäœçšãšãã£ãŒãããã¯ãéã㊠ã»ãŒããããåºå®ã®ããã¹ãã³ãŒãã¹ããåŠç¿ ããããæåã¯æäœæ¥ã§ããã°ã©ã ç©ççãªè¡åããšã ã¯ã ããã ã¯ã ã€ã³ã¿ãŒãããããŒã¿ã§åŠç¿ãã äžéšïŒããžã§ã³ãšèšèªã®ããã¯ããŒã³ïŒ ã¯ããå
åäœã®ããŒã¯ã³ ããã æ°ããç°å¢ãžã®æ±å äžçšåºŠãæ¥éã«æ¹åäž è¯å¥œïŒèšèªã¿ã¹ã¯å
ã§ïŒ äžååãèšå®ãå€ãããšç Žç¶» 2026幎ã®åçšæç床 åææ¬çªïŒå庫ãå·¥å Žã§ã®ãã€ãããïŒ æçãåºãå±é æçãå·¥å Žã§æ°å幎ã®å®çžŸ
ç¹ã«æåŸã®2è¡ã«æ³šç®ããŠãã ããã
ã«ãŒã«ããŒã¹ã®ç£æ¥çšã¢ãŒã ã¯æ°å幎ã«ãããèªåè»ã溶æ¥ããŠããŸãããæ¥µããŠä¿¡é Œæ§ãé«ãã®ã¯ãäœæ¥ã»ã«å
ã®äœãå€åããªãããã§ãããšã³ããã£ãAIã¯ããã®ä¿¡é Œæ§ãä¿ã¡ãªãããçŸå®äžçã®ä¹±éãã蚱容ããããšããŸããç ç©¶è
ã¯ãããæ±åãšåŒã³ãŸãã
LLMãšãšã³ããã£ãAIã¯ããããèšå€§ãªããŒã¿ããåŠç¿ããŸãããè§£ãåé¡ã¯å€§ããç°ãªããŸããLLM ã¯ããã¹ããå
¥åãšããŠæ¬¡ã«æ¥ãããŒã¯ã³ãäºæž¬ããŸããäžæ¹ããšã³ããã£ãã·ã¹ãã ã¯ã«ã¡ã©ç»åãé¢ç¯è§åºŠãè§ŠèŠã®èªã¿åãå€ãå
¥åãšããŠã次ã«è¡ãã¹ãåããäºæž¬ããŸãã
èŠããã«ãLLMã¯ç©çäžçãç¥ã£ãŠ ããããšã³ããã£ãAIã¯ç©çäžçã§è¡å ããŸãããªã³ãŽã®äŸã«æ»ããšãLLMã¯æã¡äžãæ¹ãæ£ç¢ºã«èª¬æã§ãããšã³ããã£ãAIã®ããããã¯å®éã«ãããè¡ããŸãã
誀ãã®ä»£åããŸã£ããéããŸããLLMãééãã°ãããå¥åŠãªæç« ã«ãªãã ãã§ãããšã³ããã£ãã·ã¹ãã ãééãã°ãã°ã©ã¹ãåºã«èœã¡ããããããŸãããããã£ãšæªãäºæ
ãããåŸãŸãã
ãããŠå€ãã®äººãèŠèœãšãç¹ããããŸããäž¡è
ã¯é£æºãããšããããšã§ãã
VLAïŒããžã§ã³ã»èšèªã»ã¢ã¯ã·ã§ã³ïŒã¢ãã« ã§ã¯ãäºååŠç¿æžã¿ã®ããžã§ã³ã»èšèªã¢ãã«ãã·ã¹ãã ã®äžæ žã«äœçœ®ããŸããããã¯ã«ã¡ã©ç»åãšããªãã®æç€ºïŒãæç©ãããŠã«ã«å
¥ããŠãïŒãèªã¿åãããã®çè§£ãã¢ã¯ã·ã§ã³ãã³ãŒãã«æž¡ããå®éã®ã¢ãŒã¿ãŒã³ãã³ããçæããŸãã
äºååŠç¿æžã¿ããžã§ã³ã»èšèªã¢ãã«ãÏâïŒãã€ã»ãŒãïŒã§ããããã®è¡åã«æ¥ç¶ãããä»çµã¿ãåºå
žç»åïŒBlack et al., 2024
ãªããã£ãžã«ã«AIã¯ããžã¿ã«AIãããé£ããã®ãïŒ
ãã£ãžã«ã«AIãããžã¿ã«AIããé£ããäž»å ã¯ããŒã¿ ã§ãã
èšèªã¢ãã«ã¯ãäœå幎ã«ããããããã¹ããã³ãŒããç»åã®ãªã³ã©ã€ã³å
±æã«ããæ¥éã«é²æ©ããŸããããçŸå®äžçã®ã»ã³ãµãŒæ
å ±ã«çžåœããå
±ææºã¯ååšããŸãããæ¥åžžç©äœãšçžäºäœçšããããããããåŸãããé¢ç¯è§ãåèŠãã«ã¡ã©ãã¬ãŒã ã®æµãã¯ã¯ããã«å°ãªãã®ã§ãã
ããŒã¿èŠä»¶ã詳ããèŠãŠã¿ãŸãããããšã³ããã£ãã·ã¹ãã ã¯ãå
¥æãé£ãã3çš®é¡ã®ããŒã¿ãå¿
èŠãšããŸãã
ã»ã³ãµãŒããŒã¿ ïŒã«ã¡ã©ã深床ã»ã³ãµãŒãã©ã€ããŒãè§ŠèŠã»ã³ãµãŒãªã©ãããããã®èŠç¹ããèšé²ããããã®
ç©äœã®ç©ç ïŒæ»ããæ²ããã転åãç Žæãšãã£ãæå
è¡åã®çµæ ïŒãããããäœããããã®åŸã©ããªã£ããã®èšé²
ããã§å
·äœçãªæ°åãèŠãŠã¿ãŸãã
æå
端ã®LLMã¯å
åäœã®ããŒã¯ã³ã§åŠç¿ããŸããæå€§çŽã®ãªãŒãã³ãããã£ã¯ã¹ããŒã¿ã»ããã§ããOpen X-Embodiment ã¯ã21æ©é¢ã»22çš®ã®ããããããããŒã¿ãéçŽããŠããçŸå®ã®è»è·¡ã¯100äžåŒ·ã«ãšã©ãŸããŸããã
Open X-EmbodimentããŒã¿ã»ããã«çµ±åããããããããšã¿ã¹ã¯ãåºå
žç»åïŒOpen X-Embodiment Collaboration, 2023
ãªãããã»ã©å°ãªãã®ããåè»è·¡ã¯ãå®åšã®ãããããå®ã¿ã¹ã¯ãå®è¡ããå¿
èŠãããããããŠãã¯äººéããã¬ãªãã§æäœããŸããããã¯æéãã³ã¹ããããããŸãã
ãã®ã®ã£ããããããã£ãžã«ã«AIã®æ±åãããžã¿ã«AIããé
ã çç±ã§ããããŸããã¢ãã«ã¯ãé¡äŒŒã®äºäŸãèŠãŠããã°ããã»ã©å€åã«åŒ·ããªããŸããã100äžã®è»è·¡ã§ã¯ãå
åäœã®ããŒã¯ã³ãæç« ãã«ããŒããå¹
ã«æ¯ã¹ãŠããããã³ãç
§æãç©äœåœ¢ç¶ã®ããªãšãŒã·ã§ã³ã倧å¹
ã«åãããŒããŸãã
ãã®èšäºã®æ®ãã®éšåã§ãããã®ããŒã¿åé¡ãé ã®çé
ã«çœ®ããŠãã ããã以äžã§åºãŠããèšèšå€æã®å€ãïŒã·ãã¥ã¬ãŒã·ã§ã³ããã¡ã€ã³ã©ã³ãã åãäºååŠç¿æžã¿ã®ããžã§ã³ã»èšèªããã¯ããŒã³ã®æŽ»çšãªã©ïŒã¯ããã¹ãŠãã®åé¡ãžã®å¯ŸåŠæ³ã§ãã
ãšã³ããã£ãAIã¯ã©ãåãã®ãïŒç¥èŠã»æšè«ã»è¡åã«ãŒã
ãšã³ããã£ãAIã¯ã次ã®3段éã®é£ç¶ã«ãŒãã§åäœããŸãã
ç¥èŠ ïŒäžçãæãåã
æšè« ïŒäœããããæ±ºãã
è¡å ïŒèº«äœãåãã
ãã®ã«ãŒãã¯æ¯ç§äœåºŠãç¹°ãè¿ãããããããåããæ¬¡ã®ç¥èŠã«åœ±é¿ããããããããµã€ã¯ã«ã®åºåã¯æ¬¡ã®ãµã€ã¯ã«ã®å
¥åã«ãªããŸãã
åãã«ãŒãã¯ã¯ãŒã«ãã¢ãã« ã®èåŸã«ããããŸããHaãšSchmidhuberã®2018幎ã®ãWorld Modelsãè«æ ã¯ããšãŒãžã§ã³ããããžã§ã³ãã¡ã¢ãªãã³ã³ãããŒã©ã«åããã¬ãŒã¹ã²ãŒã ã®è»ã§ãã¹ãããŸããããšã³ããã£ãAIã¯åãçºæ³ããã«ã®ããããã«é©çšããŸãã
ãã®ã«ãŒããçè§£ããè¯ãæ¹æ³ã¯ãããŒã«ããã£ããããå Žé¢ãæ³åããããšã§ãã
ãŸããç®ã§ããŒã«ã远ããäœçœ®ãšéåºŠãææ¡ããŸãã
次ã«ãè³ã0.5ç§åŸã®ããŒã«ã®äœçœ®ãäºæž¬ããæã®è¡ãå
ãæ±ºããŸãã
æåŸã«è
ãåããããŒã«ãè¿ã¥ãã«ã€ããŠèª¿æŽãç¶ããŸãã
ãããããåãã§ãç®ã®ä»£ããã«ã«ã¡ã©ãçèã®ä»£ããã«ã¢ãŒã¿ãŒã䜿ãã ãã§ãã
ã§ã¯å段éãèŠãŠãããŸãããã
ç¥èŠïŒç©çäžçãçè§£ãã
ç¥èŠã¯ãçã®ã»ã³ãµãŒå€ããã·ã¹ãã å
šäœãæšè«å¯èœãªè¡šçŸã«å€æããæ®µéã§ããåäžã®ã«ã¡ã©ã ãã§ã¯è¶³ããªãããšãå€ãããããŠãè€æ°ã®ã»ã³ãµãŒãçµã¿åãããŸãã
RGBã«ã¡ã© ïŒè²ãšå€èгãã·ãŒã³ã®ã¬ã€ã¢ãŠããææ¡ãããããè€æ°å°ãæèŒããã®ãäžè¬ç
深床ã»ã³ãµãŒ ãã©ã€ã㌠ïŒè·é¢ãš3D圢ç¶
åºææèŠïŒããããªãªã»ãã·ã§ã³ïŒ ïŒé¢ç¯è§ãé床ããã«ã¯ãªã©ãããããèªèº«ã®èº«äœæèŠ
è§ŠèŠã»ã³ãµãŒ ïŒæå
ã§ã®æ¥è§Šãå§åãæ»ã
Gemini Robotics-ER 1.5ã®ç¥èŠåºåäŸãåºå
žç»åïŒGoogle DeepMind
ç¥èŠã¢ãã«ã¯ããããã®èªã¿åãå€ã空éå°å³ãç©äœã®åå®ãé害ç©ã®äœçœ®ãã·ãŒã³å
šäœã®çè§£ãžãšå€æããŸãã
å€ãã¯æšæºçãªã³ã³ãã¥ãŒã¿ããžã§ã³ ã§ãããç©äœæ€åº ãã»ã°ã¡ã³ããŒã·ã§ã³ ãæ·±åºŠæšå®ãªã©ãå«ãŸããéåžžã¯DINOv2ãSigLIPã®ãããªäºååŠç¿æžã¿ã®ããžã§ã³ãã©ã³ã¹ãã©ãŒã㌠ã®äžã«æ§ç¯ãããŸãã
ãšã¯ãããè§ŠèŠã¯ä»ãã£ãšãéå°è©äŸ¡ãããŠããç¥èŠèŠçŽ ã ãšç§ã¯æããŸããã«ã¡ã©ã¯ããŒãã«ã®äžã«ã°ã©ã¹ãããããšã¯æããŠãããŸãããã°ã©ã¹ãæããæ»ãå§ããŠããããšãæããã®ã¯è§ŠèŠã§ãã
çŸç¶ã®äžäŸãšããŠãXELA Roboticsã¯Automate 2026ã§ ããããéšã«30åã®3軞 åèŠã»ã³ãµãŒãã€ã³ããå
èµããããããæå
ãæ«é²ããŸãããå瀟ã¯uSkinã»ã³ãµãŒã0.1ã°ã©ã éçžåœã®åãŸã§æ€åºå¯èœ ã ãšãè¿°ã¹ãŠããŸãã
äžè»žè§ŠèŠã»ã³ãµãŒãã€ã³ãé
åãå
èµããuSkinããããæå
ãåºå
žç»åïŒXELA Robotics
æšè«ïŒã¯ãŒã«ãã¢ãã«ãšãã©ã³ãã³ã°
æšè«ã¯ã次ã«äœãããããæ±ºããæ®µéã§ããæ°ããã·ã¹ãã ã§ã¯ãéåžž2å±€æ§é ã«ãªã£ãŠããŸãã
ã¯ãŒã«ãã¢ãã« ïŒäžäœå±€ïŒïŒãããããè¡åãèµ·ããåã« ãç°å¢ãã©ãåå¿ããããäºæž¬ããå
éšè¡šçŸ
ãã©ã³ãã³ã° ïŒäžäœå±€ïŒïŒå€§ããªç®æšãå°ããªã¹ãããã«åè§£
ã¯ãŒã«ãã¢ãã«ã¯ãããããã«ãè¡åã®åã«æ³åãããããã®ã§ãã
DreamerV3 ã¯è¯ãäŸã§ããç°å¢ã®ã³ã³ãã¯ããªã¢ãã«ãåŠç¿ãããã®æ³åäžã®äžçã®äžã§ã»ãŒãã¹ãŠã®æ¹çåŠç¿ãè¡ããŸãã
ç§èªèº«ã®ç ç©¶ã§ãDreamerV3ã䜿ã£ãŠããŸããããã£ãšãé©ããã®ã¯ããšãŒãžã§ã³ããå®ç°å¢ããããé ã®äžãã§ç·Žç¿ããæéã®æ¹ãã¯ããã«é·ãããšã§ããããã¯ãåŠç¿ãéããã¯ããã«äœã³ã¹ãã«ãªãããšãæå³ããŸãã
DreamerV3ã¯ãã¯ãŒã«ãã¢ãã«ã®æ³åããŒã«ã¢ãŠãã§æ¹çãåŠç¿ããŸããåºå
žç»åïŒHafner et al., 2023
äžæ¹ããã©ã³ãã³ã°ã¯èšèªã¢ãã«ãæ
ãããšãå¢ããŠããŸãã
奜äŸãGoogle DeepMindã®Gemini Robotics-ER 1.5ã§ãé«ã¬ãã«ã®ãã©ã³ããŒãšããŠæ©èœããŸããå°åã®ãªãµã€ã¯ã«èŠåã«åŸã£ãŠå»æ£ç©ãä»åãããããªã¿ã¹ã¯ã§ã¯ãGoogleæ€çŽ¢ã§èŠåã調ã¹ãäœæ¥ãã¹ãããã«åè§£ããåã¹ããããç©çäœæ¥ãè¡ãGemini Robotics 1.5ã®VLAã¢ãã«ã«æž¡ããŸãã
èšèªã¢ãã«ã¯ç®¡çè
ãVLAã¢ãã«ã¯å®éã«äœæ¥ãããäœæ¥è
ããšèãããšããããããã§ãããã
Gemini Robotics-ER 1.5ã¯ã¿ã¹ã¯ãèšç»ããç©ççãªå®è¡ãGemini Robotics 1.5 VLAã«å§è²ããŸããåºå
žç»åïŒGoogle DeepMind, 2025
è¡åïŒæææ±ºå®ãåãã«å€ãã
è¡åã¯ãæææ±ºå®ãåé¢ç¯ã»ã¢ãŒã¿ãŒãžã®ç²Ÿå¯ãªã³ãã³ãã«å€æããæ®µéã§ãéåžžã¯æ¯ç§æ°åãæ°çŸåïŒHzïŒè¡ãããŸãããã®æ®µéã®äœã¬ãã«éšåãå¶åŸ¡ïŒã³ã³ãããŒã«ïŒã§ãã
äŸãã°ãã°ãªãããŒãå·Šã«5cmåããããšããã³ãã³ãã¯åçŽã«èãããŸããã7é¢ç¯ã¢ãŒã ã§ã¯åé¢ç¯ã¢ãŒã¿ãŒã«å¯Ÿããå
·äœçãªãã«ã¯ã«å€æããåäœäžã¯ç¶ç¶çã«åèšç®ãç¶ããã°ãªããŸããã
é£ããã®ã¯ãçŸå®ãããããã®äºæž¬ã©ããã«ãªããªã ããšãå€ãç¹ã§ããç©äœã¯æ»ããåºã¯ãããã«äžæŽã§ããã©ã€ã³ããéãããšç
§æãå€ãããã±ãŒãã«ã¯äºæž¬ãšç°ãªãæ²ããæ¹ãããŸãã
åªããã³ã³ãããŒã©ã¯ãäºæž¬ãšå®éã®å·®ç°ã«æ°ã¥ããå³åº§ã«è£æ£ããŸãã
ç§ã¯ãéå€ã§æ§é åãããŠããªãç°å¢ã§ã¯ãè¡åïŒå¶åŸ¡ïŒããã£ãšãé£ããæ®µéã ãšèããŸããç¥èŠã®èª€ãã¯èŠçŽãã°ä¿®æ£ã§ããŸãããèšç»ã®èª€ãã¯åèšç»ã§ä¿®æ£ã§ããŸãããå¶åŸ¡ã®èª€ãã¯ãªã¢ã«ã¿ã€ã ã§çºçããŸãã
ãšã³ããã£ãAIã¯ã©ãèšç·ŽããããïŒã·ãã¥ã¬ãŒã·ã§ã³ã®åœ¹å²
ãšã³ããã£ãAIã¯ãã·ãã¥ã¬ãŒã·ã§ã³ ãšå®äžçããŒã¿ãçµã¿åãããŠåŠç¿ããŸããã·ãã¥ã¬ãŒã·ã§ã³ãšã¯ãç©ççã«æ£ç¢ºãª3Dãªããžã§ã¯ãã§æºããããä»®æ³ç°å¢ã®ããšã§ãããããã¯çŸå®ã®ããŒããŠã§ã¢ã«è§Šããåã«äœçŸäžåãç·Žç¿ã§ããŸãã
åè¿°ã®ããŒã¿åé¡ãæãåºããŠãã ãããã·ãã¥ã¬ãŒã·ã§ã³ã¯ãç¹ã«æ©è¡ããããã¥ã¬ãŒã·ã§ã³ã®åŒ·ååŠç¿ã«ãããäž»ãªåé¿çã®ã²ãšã€ã§ããÏâã®ãããªåºç€ã¢ãã«ã¯äŸç¶ãšããŠå®ãã¢ã«å€§ããäŸåãããããå€ãã®ããŒã ã¯äž¡æ¹ã䜿ããŸãã
å®äžçã§ã®ããŒã¿åéã«ã¯3ã€ã®å€§ããªåé¡ããããŸãã
é
ã ïŒ1å°ã®ããããã1æ¥ã«åéã§ãããã¢ã¯æ°çŸä»¶çšåºŠ
é«äŸ¡ ïŒããããã¯å£ããŸããã人éã®ç£ç£ãå¿
èŠ
å±éº ïŒç¹ã«éããã¥ãŒããã€ããè»ã§ã¯
ã·ãã¥ã¬ãŒã¿ã¯ããã3ã€ãåæã«è§£æ±ºããŸããåäžGPUäžã§äœåãã®ä»®æ³ããããã䞊å皌åãããå¿
èŠãªã ã倱æã»åéãç¹°ãè¿ããŸããããã¯æ°å¹Žåã®ããããçµéšãæ°æéã«åçž®ããŸãã
è¯ãã·ãã¥ã¬ãŒã·ã§ã³ç°å¢ã®æ¡ä»¶
ãã¹ãŠã®ã·ãã¥ã¬ãŒã¿ãããããåŠç¿ã«çããæçšãšããããã§ã¯ãããŸãããç§èªèº«ãããããã¢ãŒã ãã·ãã¥ã¬ãŒã·ã§ã³ã§æ±ã£ãŠããçµéšãããè¯ãç°å¢ã«ã¯æ¬¡ã®3æ¡ä»¶ãå¿
èŠã§ãã
ç©ç粟床 ïŒæ¥è§Šãæ©æŠãå€åœ¢ãçŸå®ã©ããã«æ¯ãèãããš
ç©äœã®å€æ§æ§ ïŒåããã°ã«ãããååã§ã¯ãªããç°ãªããã°ã«ãããäœåçš®ãèŠããããããš
ãã¡ã€ã³ã©ã³ãã å ïŒåŠç¿ãšããœãŒãããšã«ç
§æããã¯ã¹ãã£ã質éãæ©æŠãªã©ãå€åãããããããšïŒè©³çްã¯åŸè¿°ïŒ
æãããç®ã«ãããã©ãããã©ãŒã ã¯ãNVIDIA Isaac SimïŒããã³Isaac LabïŒãšMuJoCoã§ãã
ãã©ãããã©ãŒã éçºå
åŸæåé æé©ãªçšé NVIDIA Isaac Sim ãš Isaac Lab NVIDIA ãã©ããªã¢ã«ãªã¬ã³ããªã³ã°ãã»ã³ãµãŒã·ãã¥ã¬ãŒã·ã§ã³ãGPUäžã§ã®æ°åç°å¢ã®äžŠåå®è¡ å€§èŠæš¡åŒ·ååŠç¿ãšåæã«ã¡ã©ããŒã¿ MuJoCo Google DeepMindïŒãªãŒãã³ãœãŒã¹ïŒ é«éã§æ£ç¢ºãªæ¥è§Šç©çã軜éãå€§èŠæš¡ç ç©¶ã³ãã¥ãã㣠ç ç©¶ãæ©è¡ã»ãããã¥ã¬ãŒã·ã§ã³ã®ãã³ãããŒã¯
ææ°ã®è¿œå ãNewton ã§ãNVIDIAãGoogle DeepMindãDisney Researchãæ§ç¯ããLinux Foundationã®ç®¡çäž ã«ããGPUå éã®ãªãŒãã³ãœãŒã¹ç©çãšã³ãžã³ã§ãã
Newton 1.0ã¯2026幎3æã®NVIDIA GTCã§ãªãªãŒã¹ ãããŸãããIsaac Labã®ç©çããã¯ãšã³ããšããŠæ¥ç¶ã§ããSolverãšããŠMuJoCo Warpãåããã±ãŒãã«ãåžã®ãããªå¯å€åœ¢ç©äœåãã®è¿œå SolverãæäŸããŸãã
å¯å€åœ¢ç©ã¯èŠãç®ä»¥äžã«éèŠã§ããå€ãã·ãã¥ã¬ãŒã¿ã¯ã±ãŒãã«ãåžã®æ±ããäžåŸæã§ãå·¥å Žã§ã¯ãã®äž¡æ¹ãæ±ãããããããæ±ããããŠããŸãã
ã·ã ã»ãã¥ã»ãªã¢ã«ã®ã£ãã
ã·ã ã»ãã¥ã»ãªã¢ã«ã®ã£ãããšã¯ãã·ãã¥ã¬ãŒã·ã§ã³ã§åŠç¿ããæ¹çã宿©ã«ç§»ããéã«çããæ§èœäœäžã®ããšã§ããšã³ããã£ãã·ã¹ãã æå€§ã®èª²é¡ã®ã²ãšã€ã§ãã
äŸãã°ãã·ãã¥ã¬ãŒã·ã§ã³ã®æ©æŠã¯åžžã«å°ããããã«ã¡ã©ã¯ããããããç©äœã¯çæ³åãããã¡ã§ãããã®ãããã·ãã¥ã¬ãŒã·ã§ã³ã§ã¯95%æåããŠããæ¹çãã宿©ã«åºäŒã£ãé端ã«ç Žç¶»ããããšããããŸãã
ãã®ã®ã£ãããåããäž»èŠãªææ³ã¯2ã€ãããŸãã
åŠç¿äžã®ãã¡ã€ã³ã©ã³ãã å
ã·ãã¥ã¬ãŒã¿ãå®ç§ã«è¿ã¥ããã®ã§ã¯ãªããããŸããŸãªåœ¢ã§ã©ã³ãã åããŸãã
TobinãïŒ2017ïŒ ã¯ãã¯ã¹ãã£ãç
§æã倧ããã©ã³ãã åããå®äžçãã¢ãã«ã«ãšã£ãŠãããã²ãšã€ã®ããªãšãŒã·ã§ã³ãã«èŠããããã«ããŸãããOpenAIã®ã«ãŒããã¯ãã¥ãŒãçšãããããã³ã ã¯ãæ¹çãäžéããã«ã€ããŠã©ã³ãã åã®å¹
ãèªåçã«åºããŸããã
ã·ãã¥ã¬ãŒã·ã§ã³åŸã®å®ããŒã¿ã§ã®ãã¡ã€ã³ãã¥ãŒãã³ã°
ã·ãã¥ã¬ãŒã·ã§ã³ã§äºååŠç¿ããæ¹çã¯ãã¿ã¹ã¯ã®ãåãããã§ã«åŠãã§ãããããé©å¿ã®ããã«å¿
èŠãªå®ãã¢ã¯å°æ°ã§æžãããšãå€ãã§ãã
ã©ã¡ãã®ææ³ãã®ã£ãããå®å
šã«ã¯æ¶ããªããããçŸå®äžçã§ã®æ§èœãžã®åœ±é¿ãæããã¹ããåããŒã ã¯åãçµã¿ãç¶ããŠããŸãã
2026幎ã®ãšã³ããã£ãAIäºäŸ
ãšã³ããã£ãAIã¯ãåéã«ãã£ãŠæ¿æ·¡ã¯ãããã®ã®ããã§ã«ç 究宀ã®å€ã§çšŒåããŠããŸãã
ç§ã®èŠç«ãŠã§ã¯ã人æã§ã¹ã¯ãªããåããã«ã¯ç°å¢å€åã倧ããããåé¡ãèµ·ããã人éãä»å
¥ã§ããå Žæããé ã«å°å
¥ãé²ã¿ãŸãã
èªåé転è»
èªåé転ã¯ããã£ãšãããŒã¿ãå¿
èŠãšãããšã³ããã£ãAIã®äžçš®ã§ãã
Waymo Driverã¯å®å
šèªåã§çŽ2åãã€ã«ãèµ°è¡ããããã«ã·ãã¥ã¬ãŒã·ã§ã³ã§æ°ååãã€ã«ã®åŠç¿ã»ãã¹ã ãè¡ã£ãŠãããšã2026幎2æã®Waymoã®World Modelã«é¢ããæçš¿ã§è¿°ã¹ãŠããŸããã·ãã¥ã¬ãŒã·ã§ã³ã«ãããå®éã®ã€ã³ã·ãã³ããåçŸããããå®è·¯ã§ã¯ãã¹ãè»äž¡ãééããªããããªã¬ã¢ã±ãŒã¹ïŒäŸïŒé§è»è»äž¡ã®éããåã©ããé£ã³åºãïŒãçæã§ããŸãã
èªåé転ã¯ãç¥èŠã»æšè«ã»è¡åã«ãŒããå
šéã§åã£ãŠãã奜äŸã§ããããŸããWaymoã®è»äž¡ã¯ã«ã¡ã©ãã©ã€ããŒãã¬ãŒããŒã§ç¥èŠããæ©è¡è
ãè»äž¡ã次ã«åããããªè¡åãæšè«ãããã³ãã«ããã¬ãŒããæ¯ç§äœåºŠãå¶åŸ¡ããŠè¡åããŸãã
å庫ãšç©æµ
ç§èŠã§ã¯ãå庫ã¯çŸåšãã£ãšãèªç¶ãªãšã³ããã£ãAIã®åçšçšéã§ãã
åºã®ã©ã€ã³ã«åŸãåºå®çµè·¯åãããããããåçã§éå€ãªåšåº« ã«å¯Ÿå¿ã§ããã·ã¹ãã ãžã®ã·ãããé²ãã§ããŸããäŸãã°40çš®é¡ã®è£œåãäžç·ããã«å
¥ã£ãããŒãããã®ãããã³ã°ãªã©ã§ãã
Agility Roboticsã®ãã¥ãŒããã€ããDigitãã¯ã2024幎ã«ç· çµ ãããè€æ°å¹Žå¥çŽã®ããšGXO Logisticsã§ããŒãæ¬éãè¡ã£ãŠãããBoston Dynamicsã®Stretchã¯ãã©ãã¯ããã®ç®±åºããæ
ããŸãã
Agility Roboticsã®DigitããAMRãšã³ã³ãã€éã§ããŒããæ¬éãåºå
žç»åïŒAgility Robotics/The Robot Report
ãã«ã¹ã±ã¢ãããã£ã¯ã¹
ãã«ã¹ã±ã¢ã¯ããšã³ããã£ãAIããã£ãšãæ
éã«é²ãé åã ãšèŠãŠããŸãã
Intuitiveã®da Vinciã®ãããªæè¡æ¯æŽã·ã¹ãã ã¯ããã§ã«äžçäžã®ç
é¢ã§äœ¿ãããŠããŸãããæäœè
ã¯å€ç§å»ã§ãããã®åéã®AIç ç©¶ã®å€ãã¯ãåšå
·è¿œè·¡ãã«ã¡ã©å¶åŸ¡ãçž«åæ¯æŽãšãã£ãã¢ã·ã¹ãã«çŠç¹ãåœãŠãŠããŸãã
ãã«ã¹ã±ã¢ã§ã¯ãã¢ãã«ã®èœå以äžã«èŠå¶æ¿èªãå¶çŽãšãªãããšãå€ãããããŠããã¯æ£åœã§ããèªåŸæè¡ã«è¿ããã®ãæ¥ãåã«ãã¢ã·ã¹ãããã¬ãŒãã³ã°çšéãå
ã«æ®åãããšå人çã«ã¯äºæ³ããŸãã
å·¥å Žã®ãã¥ãŒããã€ããããã
ãã¥ãŒããã€ãã¯æ³šç®åºŠãé«ãäžæ¹ã§ãå®çžŸã¯éå®çã§ãã
CES 2026ã§Boston Dynamicsã¯Atlasã®éç£ç ãçºè¡šãã2026幎ã«è£œé ããå
šå°æ°ã¯HyundaiãšGoogle DeepMindã«ãã§ã«å²ãåœãŠæžã¿ã ãšè¿°ã¹ãŸãããäžæ¹Figureã¯ããµãŠã¹ã«ãã©ã€ãå·ã¹ããŒã¿ã³ããŒã°ã®BMWå·¥å Žã§ãFigure 02ãã¥ãŒããã€ãã«ããæ¿éããŒãã®11ãæã®å°å
¥ ãè¡ããŸããã
ççŽã«èšãã°ãå€ãã®ãã¥ãŒããã€ãå°å
¥ã¯ãããŸã éå®ã¿ã¹ã¯ã®ãã€ãããæ®µéã§ããäŸãã°Hyundaiã¯Atlasã2028幎ã«éšåã·ãŒã±ã³ã·ã³ã°ã§äœ¿ãå§ããèšç»ã§ãæå€§æã§ãã£ãŠã極ããŠæ
éã«é²ããŠããããšãããããŸãã
2026幎ã«ç¥ã£ãŠããã¹ãäž»èŠãšã³ããã£ãAIäŒæ¥
ã§ã¯ãå®éã«ããããäœã£ãŠããã®ã¯èª°ã§ããããããšã³ããã£ãAIãåããŠåŠã¶æ¹ã«ãæäœéæŒãããŠã»ãã5ã€ã®çµç¹ãæããŸãã
çµç¹ éçºç© éèŠãªçç± NVIDIA Isaac SimãIsaac LabãNewtonãGR00TãCosmos å€ãã®ããŒã ãåŠç¿ã«äœ¿ãã·ãã¥ã¬ãŒã·ã§ã³ãšã³ã³ãã¥ãŒãã®ã¹ã¿ã㯠Physical Intelligence Ïâç³»ã®ããããåºç€ã¢ãã« ãªãŒãã³ãªãã§ãã¯ãã€ã³ããåããæ±çšããããæ¹ç Agility Robotics ãã¥ãŒããã€ãDigit ååº«ç©æµã«ç¹åãããã§ã«é¡§å®¢ãšçšŒå Boston Dynamics AtlasãStretchãSpot Atlasãç ç©¶ãã¢ãã2026幎ã«éç£æ©ãžç§»è¡ AMI LabsïŒYann LeCunïŒ JEPAç³»ã¯ãŒã«ãã¢ãã« ãã¯ã»ã«çæãè¡ããªãã¯ãŒã«ãã¢ãã«ãžã®é匵ã
NVIDIA
NVIDIAã¯ãããããæ¬äœãšãããããããããåãå·»ãããŒã«çŸ€ãäž»ã«éçºããŠããŸãã
Isaac Simã¯ãã©ããªã¢ã«ãªã·ãã¥ã¬ãŒã·ã§ã³ããIsaac Labã¯ãã®äžã§ã®åŒ·ååŠç¿ããNewtonã¯ç©çãæ
ããŸããNVIDIAã¯æ±çšããããã販売ããŠããªãã«ãããããããå€ãã®ãããã£ã¯ã¹ããŒã ãNVIDIAã®ãœãããŠã§ã¢ã§åŠç¿ããŠããŸãã
Physical Intelligence
Physical Intelligenceã¯ãÏâ ãšãã33åãã©ã¡ãŒã¿ã®ããžã§ã³ã»èšèªã»ã¢ã¯ã·ã§ã³ã¢ãã« ãæ§ç¯ãããããŒãããã³ã°ã§æŽæ¿¯ç©ããã¿ã®ãããªã¿ã¹ã¯åãã®æ»ãããªã¢ã¯ã·ã§ã³åãçæããŸãã
å®éã«ããŠã³ããŒãã§ããæ±çšããããã¢ãã«ãšããŠãç§ãç¥ãéãæè¯ã®äŸã§ããopenpiãªããžã㪠ããå
¥æã§ããŸãã
ãåºç€ã¢ãã«ãã®æŠå¿µãåããŠã§ããã°ãIntroduction to Foundation Models ã¬ã€ãããããããã説æããŠããŸãã
Agility Robotics
Agility Roboticsã¯ãDigitã§çãé åã«çµããŸããã
åœåããå庫ã§ã®ããŒãæ¬éã«æé©åãããŠããããã®æç¢ºãªçŠç¹ããä»ã®å€ãã®ãã¥ãŒããã€ãããæ©ãæåã®é¡§å®¢å°å
¥ã«è³ã£ã倧ããªçç±ã§ãã
Boston Dynamics
Boston Dynamicsã¯ãAtlasã§å¯Ÿç
§çãªã¢ãããŒããåããŸããã
ïŒãã«ã¯ãŒã«åç»ã§ããªãã¿ã®ããã«ïŒé·å¹Žããããã®éåæ§èœãéçãŸã§é«ãããã®åŸãGoogle DeepMindã®Gemini Roboticsã¢ãã«ã§é§åããèšç»ã®ããªãŒã«é»åã®å·¥å Žè£œåãžãšè»¢æããŸããã
AMI LabsïŒYann LeCunïŒ
AMI Labsã¯Yann LeCunã®ããªæ ç¹ã®ã¹ã¿ãŒãã¢ããã§ã2026幎3æã«10å3000äžãã«ã®ã·ãŒãã©ãŠã³ãã宿œïŒãã¬ãããŒè©äŸ¡35åãã«ïŒãCathay InnovationãGreycroftãHiro CapitalãHV CapitalãBezos Expeditionsãå
±åäž»å°ããNVIDIAãSamsungãåºè³ããŠããŸãã
LeCunã¯ãå°æ¥ã®åãã¯ã»ã«ãäºæž¬ããã®ã¯ç¡é§ãå€ããšé·å¹Žäž»åŒµããŠãããJoint Embedding Predictive ArchitectureïŒJEPAïŒã¯ã代ããã«æœè±¡è¡šçŸç©ºéã§äºæž¬ãè¡ããŸãããã®ã¢ã€ãã¢ã¯ãMetaããã§ã«V-JEPA 2 ã§æ€èšŒæžã¿ã§ãã
AMIãæ³šç®ã«å€ããã®ã¯ãåéã®äž»æµãããããŠå€ããŠããç¹ã§ãããã£ãšãããŸã äœãåºè·ããŠããªããããçŸæç¹ã§ã¯å®èšŒæžã¿ææ³ãšããããã倧åã®è³éãä»ããç ç©¶äžã®è³ããšèŠãã®ã劥åœã§ãããã
ããŒã¿ãµã€ãšã³ãã£ã¹ãã®ããã®ãšã³ããã£ãAIïŒã©ãã«è²¢ç®ã§ããã
ãšã³ããã£ãAIã¯ãããã£ã¯ã¹å·¥åŠã ãã®åé¡ã§ã¯ãããŸãããäœæ¥ã®å€ããããã倧åã¯æ©æ¢°åŠç¿ã®ä»äºã ãšèšã£ãŠããã§ãããã
ãã£ãšãçŽçµããã¹ãã«ã¯ä»¥äžã§ãã
ç¥èŠã»æšè«ã»è¡åã«ãŒããèªåã§è©Šã
éå§ã«ããããã¯äžèŠã§ããgymnasiumã©ã€ãã©ãªã«ã¯MuJoCoç°å¢ã忢±ãããŠãããããpip install "gymnasium[mujoco]"ãå®è¡ããã°ãã·ãã¥ã¬ãŒã·ã§ã³ããããã§ã«ãŒãå
šäœãåãããŸãã
import gymnasium as gym
env = gym. make( "HalfCheetah-v5" )
observation, info = env. reset( seed= 42 )
for step in range ( 1000 ) :
action = env. action_space. sample( )
observation, reward, terminated, truncated, info = env. step( action)
if terminated or truncated:
observation, info = env. reset( )
env. close( ) æäŸïŒ
çŸæç¹ã§ã¯ããã§ãŒã¿ã¯ãæšè«ãã¹ããããenv.action_space.sample()ã§æ¯åã©ã³ãã ã«è¡åãéžã¶ããããã ãã¿ãã¿ããã ãã§ãã
ãã®1è¡ãåŠç¿æžã¿ã®æ¹çã«çœ®ãæããã®ãããŸãã«åŒ·ååŠç¿ã®åœ¹å²ã§ãããã倧ããªé¡æã«é²ãåã«ããã²è©ŠããŠã¿ãŠãã ããã
ãšã³ããã£ãAIã®éçã¯ïŒ
ãšã³ããã£ãAIã¯æ¥éã«é²æ©ããŠããŸããããªã4ã€ã®åŒ±ç¹ããããç¥ã£ãŠãã䟡å€ããããŸãã
倱æããã®å埩 ãåŠç¿ããŒã¿ã®å€ãã¯æåäŸã§ãããã€ãã¿æããªã©äœæ¥ã®éäžã§èµ·ãã倱æãžã®å¯ŸåŠãåŠã¶æ©äŒãå°ãªãã
é·ãã¿ã¹ã¯ ãå°ããªèª€å·®ã®çޝç©ãåã¹ããã95%æåã»å€±æãç¬ç«ãªãã20ã¹ãããã®ã¿ã¹ã¯å®äºçã¯çŽ36%ã«èœã¡ãŸãã
ããããäžã§ã®é«éåŠç ãæ°ååãã©ã¡ãŒã¿ã®ã¢ãã«ã宿©ããŒãã§ãªã¢ã«ã¿ã€ã å¶åŸ¡å¯èœãªé床ã§åããã®ã¯äŸç¶ãšããŠé£ããããã®ããå€ãã®ã·ã¹ãã ã¯é
ãæšè«ãšéãå¶åŸ¡ãåé¢ããé
ãéšåãããããå€ã§åããããšããããŸãã
äžæ
£ããªç°å¢ ãåŠç¿ååžã«è¿ãã·ãŒã³ã¯åŸæã§ãããã®å€ã§ã¯æããã«æ§èœãèœã¡ãŸããçµå±ã¯ãåé ã§è¿°ã¹ãããŒã¿åé¡ã«æ»ããŸãã
ãŸãšã
ãšã³ããã£ãAIã¯ãæ©æ¢°ã«ç©ççãªèº«äœãšãããã䜿ãããªãç¥èœãäžããŸããç¥èŠã»æšè«ã»è¡åã®ã«ãŒãã§åäœããäž»ãªå¶çŽã¯ç©çããŒã¿ã®äžè¶³ã§ãã2026幎çŸåšãç 究宀ãåºãŠãå庫ãåæã®å·¥å Žãã€ããããžãšé²ã¿ã€ã€ãããŸãã
äœããé©ãã®ã¯ãåãã®åããå€ãã£ãããšã§ããæ°å¹Žåã¯ãLLMããããã£ã¯ã¹ç ç©¶ãäžèŠã«ããã®ã§ã¯ããšåãããŸãããå®éã«ã¯ãLLMã¯ãšã³ããã£ãã·ã¹ãã å
éšã®æšè«å±€ ãšãªããäž¡åéã¯çžäºã«èåãã€ã€ãããŸãã
èšäºåé ã®ãªã³ãŽãæãåºããŠãã ãããæã¡æ¹ããç¥ã£ãŠãããããšãšãå®éã«ãæã¡äžãããããšã¯ããŸã£ããå¥ã®åé¡ã§ããããšã³ããã£ãAIã¯ããã®åŸè
ã«åãçµãåéã§ãã
ãã®åºç€ãšãªãMLãåŠã³ãããªãããŸãã¯Reinforcement Learning in Pythonãã©ã㯠ããå§ããŠãã ãããã¹ã¿ãã¯ã®èšèªåŽã¯ãLarge Language Models (LLMs) Concepts ã³ãŒã¹ããDeveloping Large Language Models ãã©ãã¯ãè¯ã次ã®äžæ©ã«ãªããŸãã
åäººã«æç©ããŠã«ãããªã³ãŽãåã£ãŠæž¡ããŠãããããé Œããšãããæ³åããŠãã ãããå人ã¯ããŠã«ãèŠãŠãã©ãããªã³ãŽãã倿ããã€ã¶ããªãçšåºŠã®åå æžã§æã¡äžããŠãããªãã«ææž¡ããŸãã
ã§ã¯åãããšããã£ãããããã«é Œãã ãã©ãã§ãããããªã³ãŽã®æã¡æ¹ãã©ã®æã䜿ãã¹ãããããããã®åå æžãŸã§æ£ç¢ºã«èª¬æã¯ã§ããŸãããå®éã«æã¡äžããããšã¯ã§ããŸãããæããªãããªã³ãŽã®è§Šæãããããªãããã§ãã
ãšã³ããã£ãAIã¯ããã®ã®ã£ãããåããããšããåéã§ããç°¡åã«èšãã°ãç©ççãªèº«äœïŒãããããè»ããããŒã³ãªã©ïŒãæã¡ãåã«æç« ãèªãã®ã§ã¯ãªããçŸå®äžçã§è¡åããããšã§åŠç¿ããAIã§ããåãèãæ¹ãæãçšèªãšããŠããã£ãžã«ã«AIãããã䜿ãããŸãã
ããŸããã®åéã¯å€§ããªè»¢æ©ãè¿ããŠããŸãã2026幎1æã®CESã§ãNVIDIAã®Jensen Huangã¯ãããããããã£ã¯ã¹ã«ãããChatGPTã®ç¬éããšè¡šçŸããŸããã
è³éãæµã蟌ãã§ããŸããCNBCãå ±ããDealroomã®ããŒã¿ã«ãããšããããã£ã¯ã¹äŒæ¥ã¯2026幎6æåæ¬ãŸã§ã«558åãã«ã調éããéå»ã®å¹Žéèšé²ãã»ãŒåå¢ãããŸããã
æ¬èšäºã§ã¯æ¬¡ã®å å®¹ãæ±ããŸãã
ããããã«è§ŠããããšããªããŠãå¿é ã¯ãããŸãããæ©æ¢°åŠç¿ã«å€å°ã®éŠŽæã¿ãããã°å©ãã«ãªããŸãããã©ã¡ãã«ããåºç€ããé ã远ã£ãŠè§£èª¬ããã®ã§ãç¡çãªãèªã¿é²ããããŸãã
ãšã³ããã£ãAIã®èŠç¹
ãšã³ããã£ãAIãšã¯ïŒ
ãšã³ããã£ãAIã¯ããããããèªåé転è»ããããŒã³ã®ãããªç©ççãªèº«äœã«çµã¿èŸŒãŸããã»ã³ãµãŒã§ç°å¢ãç¥èŠããäœãããããæšè«ããã¢ãŒã¿ãŒãéããŠäžçã«äœçšããèªãã®è¡åã®çµæããåŠç¿ãã人工ç¥èœã·ã¹ãã ã§ãã
ããã§ã®éµã¯ãšã³ããã£ãïŒèº«äœæ§ïŒã§ãã
å€ãã®AIã¢ãã«ã¯ãä»è ãåéããããŒã¿ã芳å¯ããŠåŠã³ãŸããããã«å¯Ÿãããšã³ããã£ããªã·ã¹ãã ã¯ãã³ãããæŒããŠãã³ãããæ»ãã®ã芳å¯ããæŒãããšãã«ã³ãããã©ãæ¯ãèãããšããç¥èãæŽæ°ããããšãã£ãå ·åã«ãç°å¢ãšçžäºäœçšããŠåŠã³ãŸãã
äŸãæããŸããããæ°çŸäžæã®ãã¢ã®åçã§åŠç¿ããããžã§ã³ã¢ãã«ã¯ããã¢ãã©ã®ããã«èŠããããç¥ã£ãŠããŸããäžæ¹ã§ãå®éã«500æã®ãã¢ãéããããšããããããã¯ãæŒããã¢ãããã°åŒããã¢ããããéããã®ãããã°ãå ã«ãã³ãã«ãæŒãäžããå¿ èŠããããã®ãããããšç¥ã£ãŠããŸãã
ãã®ã¬ãã«ã®ç¥èã¯åçããã¯åŸãããŸããã
ãã衚çŸããã®ã奜ãã§ããå€ãã®AIã¯äžçã«ã€ããŠãèªãã§ãåŠã¶ãããšã³ããã£ãAIã¯äžçã«ãè§Šããäœéšãããããšã§åŠã¶ã®ã§ãã
ãã®éãããå¿ èŠãªããŒã¿ãåŠç¿æ¹æ³ããããŠæ³å®å€ã®äºæ ã®èµ·ããæ¹ãäžå€ãããŸãã
ãšã³ããã£ãAI vs. å€§èŠæš¡èšèªã¢ãã« vs. åŸæ¥åãããã£ã¯ã¹
ãããŸã§ããšã³ããã£ãAIããã£ããããããšæ¯ã¹ãŠããŸãããæ¬¡ã«ã人ã ãããæ··åãã2ã€ãããªãã¡å€§èŠæš¡èšèªã¢ãã«ïŒLLMïŒãšåŸæ¥ã®ã«ãŒã«ããŒã¹ãããã£ã¯ã¹ãšæ¯èŒããŠã¿ãŸãããã
ç¹ã«æåŸã®2è¡ã«æ³šç®ããŠãã ããã
ã«ãŒã«ããŒã¹ã®ç£æ¥çšã¢ãŒã ã¯æ°å幎ã«ãããèªåè»ã溶æ¥ããŠããŸãããæ¥µããŠä¿¡é Œæ§ãé«ãã®ã¯ãäœæ¥ã»ã«å ã®äœãå€åããªãããã§ãããšã³ããã£ãAIã¯ããã®ä¿¡é Œæ§ãä¿ã¡ãªãããçŸå®äžçã®ä¹±éãã蚱容ããããšããŸããç ç©¶è ã¯ãããæ±åãšåŒã³ãŸãã
LLMãšãšã³ããã£ãAIã¯ããããèšå€§ãªããŒã¿ããåŠç¿ããŸãããè§£ãåé¡ã¯å€§ããç°ãªããŸããLLMã¯ããã¹ããå ¥åãšããŠæ¬¡ã«æ¥ãããŒã¯ã³ãäºæž¬ããŸããäžæ¹ããšã³ããã£ãã·ã¹ãã ã¯ã«ã¡ã©ç»åãé¢ç¯è§åºŠãè§ŠèŠã®èªã¿åãå€ãå ¥åãšããŠã次ã«è¡ãã¹ãåããäºæž¬ããŸãã
èŠããã«ãLLMã¯ç©çäžçãç¥ã£ãŠããããšã³ããã£ãAIã¯ç©çäžçã§è¡åããŸãããªã³ãŽã®äŸã«æ»ããšãLLMã¯æã¡äžãæ¹ãæ£ç¢ºã«èª¬æã§ãããšã³ããã£ãAIã®ããããã¯å®éã«ãããè¡ããŸãã
誀ãã®ä»£åããŸã£ããéããŸããLLMãééãã°ãããå¥åŠãªæç« ã«ãªãã ãã§ãããšã³ããã£ãã·ã¹ãã ãééãã°ãã°ã©ã¹ãåºã«èœã¡ããããããŸãããããã£ãšæªãäºæ ãããåŸãŸãã
ãããŠå€ãã®äººãèŠèœãšãç¹ããããŸããäž¡è ã¯é£æºãããšããããšã§ãã
VLAïŒããžã§ã³ã»èšèªã»ã¢ã¯ã·ã§ã³ïŒã¢ãã«ã§ã¯ãäºååŠç¿æžã¿ã®ããžã§ã³ã»èšèªã¢ãã«ãã·ã¹ãã ã®äžæ žã«äœçœ®ããŸããããã¯ã«ã¡ã©ç»åãšããªãã®æç€ºïŒãæç©ãããŠã«ã«å ¥ããŠãïŒãèªã¿åãããã®çè§£ãã¢ã¯ã·ã§ã³ãã³ãŒãã«æž¡ããå®éã®ã¢ãŒã¿ãŒã³ãã³ããçæããŸãã
äºååŠç¿æžã¿ããžã§ã³ã»èšèªã¢ãã«ãÏâïŒãã€ã»ãŒãïŒã§ããããã®è¡åã«æ¥ç¶ãããä»çµã¿ãåºå žç»åïŒBlack et al., 2024
ãªããã£ãžã«ã«AIã¯ããžã¿ã«AIãããé£ããã®ãïŒ
ãã£ãžã«ã«AIãããžã¿ã«AIããé£ããäž»å ã¯ããŒã¿ã§ãã
èšèªã¢ãã«ã¯ãäœå幎ã«ããããããã¹ããã³ãŒããç»åã®ãªã³ã©ã€ã³å ±æã«ããæ¥éã«é²æ©ããŸããããçŸå®äžçã®ã»ã³ãµãŒæ å ±ã«çžåœããå ±ææºã¯ååšããŸãããæ¥åžžç©äœãšçžäºäœçšããããããããåŸãããé¢ç¯è§ãåèŠãã«ã¡ã©ãã¬ãŒã ã®æµãã¯ã¯ããã«å°ãªãã®ã§ãã
ããŒã¿èŠä»¶ã詳ããèŠãŠã¿ãŸãããããšã³ããã£ãã·ã¹ãã ã¯ãå ¥æãé£ãã3çš®é¡ã®ããŒã¿ãå¿ èŠãšããŸãã
ããã§å ·äœçãªæ°åãèŠãŠã¿ãŸãã
æå 端ã®LLMã¯å åäœã®ããŒã¯ã³ã§åŠç¿ããŸããæå€§çŽã®ãªãŒãã³ãããã£ã¯ã¹ããŒã¿ã»ããã§ããOpen X-Embodimentã¯ã21æ©é¢ã»22çš®ã®ããããããããŒã¿ãéçŽããŠããçŸå®ã®è»è·¡ã¯100äžåŒ·ã«ãšã©ãŸããŸããã
Open X-EmbodimentããŒã¿ã»ããã«çµ±åããããããããšã¿ã¹ã¯ãåºå žç»åïŒOpen X-Embodiment Collaboration, 2023
ãªãããã»ã©å°ãªãã®ããåè»è·¡ã¯ãå®åšã®ãããããå®ã¿ã¹ã¯ãå®è¡ããå¿ èŠãããããããŠãã¯äººéããã¬ãªãã§æäœããŸããããã¯æéãã³ã¹ããããããŸãã
ãã®ã®ã£ããããããã£ãžã«ã«AIã®æ±åãããžã¿ã«AIããé ãçç±ã§ããããŸããã¢ãã«ã¯ãé¡äŒŒã®äºäŸãèŠãŠããã°ããã»ã©å€åã«åŒ·ããªããŸããã100äžã®è»è·¡ã§ã¯ãå åäœã®ããŒã¯ã³ãæç« ãã«ããŒããå¹ ã«æ¯ã¹ãŠããããã³ãç §æãç©äœåœ¢ç¶ã®ããªãšãŒã·ã§ã³ãå€§å¹ ã«åãããŒããŸãã
ãã®èšäºã®æ®ãã®éšåã§ãããã®ããŒã¿åé¡ãé ã®çé ã«çœ®ããŠãã ããã以äžã§åºãŠããèšèšå€æã®å€ãïŒã·ãã¥ã¬ãŒã·ã§ã³ããã¡ã€ã³ã©ã³ãã åãäºååŠç¿æžã¿ã®ããžã§ã³ã»èšèªããã¯ããŒã³ã®æŽ»çšãªã©ïŒã¯ããã¹ãŠãã®åé¡ãžã®å¯ŸåŠæ³ã§ãã
ãšã³ããã£ãAIã¯ã©ãåãã®ãïŒç¥èŠã»æšè«ã»è¡åã«ãŒã
ãšã³ããã£ãAIã¯ã次ã®3段éã®é£ç¶ã«ãŒãã§åäœããŸãã
ãã®ã«ãŒãã¯æ¯ç§äœåºŠãç¹°ãè¿ãããããããåããæ¬¡ã®ç¥èŠã«åœ±é¿ããããããããµã€ã¯ã«ã®åºåã¯æ¬¡ã®ãµã€ã¯ã«ã®å ¥åã«ãªããŸãã
åãã«ãŒãã¯ã¯ãŒã«ãã¢ãã«ã®èåŸã«ããããŸããHaãšSchmidhuberã®2018幎ã®ãWorld Modelsãè«æã¯ããšãŒãžã§ã³ããããžã§ã³ãã¡ã¢ãªãã³ã³ãããŒã©ã«åããã¬ãŒã¹ã²ãŒã ã®è»ã§ãã¹ãããŸããããšã³ããã£ãAIã¯åãçºæ³ããã«ã®ããããã«é©çšããŸãã
ãã®ã«ãŒããçè§£ããè¯ãæ¹æ³ã¯ãããŒã«ããã£ããããå Žé¢ãæ³åããããšã§ãã
ãããããåãã§ãç®ã®ä»£ããã«ã«ã¡ã©ãçèã®ä»£ããã«ã¢ãŒã¿ãŒã䜿ãã ãã§ãã
ã§ã¯å段éãèŠãŠãããŸãããã
ç¥èŠïŒç©çäžçãçè§£ãã
ç¥èŠã¯ãçã®ã»ã³ãµãŒå€ããã·ã¹ãã å šäœãæšè«å¯èœãªè¡šçŸã«å€æããæ®µéã§ããåäžã®ã«ã¡ã©ã ãã§ã¯è¶³ããªãããšãå€ãããããŠãè€æ°ã®ã»ã³ãµãŒãçµã¿åãããŸãã
Gemini Robotics-ER 1.5ã®ç¥èŠåºåäŸãåºå žç»åïŒGoogle DeepMind
ç¥èŠã¢ãã«ã¯ããããã®èªã¿åãå€ã空éå°å³ãç©äœã®åå®ãé害ç©ã®äœçœ®ãã·ãŒã³å šäœã®çè§£ãžãšå€æããŸãã
å€ãã¯æšæºçãªã³ã³ãã¥ãŒã¿ããžã§ã³ã§ãããç©äœæ€åºãã»ã°ã¡ã³ããŒã·ã§ã³ã深床æšå®ãªã©ãå«ãŸããéåžžã¯DINOv2ãSigLIPã®ãããªäºååŠç¿æžã¿ã®ããžã§ã³ãã©ã³ã¹ãã©ãŒããŒã®äžã«æ§ç¯ãããŸãã
ãšã¯ãããè§ŠèŠã¯ä»ãã£ãšãéå°è©äŸ¡ãããŠããç¥èŠèŠçŽ ã ãšç§ã¯æããŸããã«ã¡ã©ã¯ããŒãã«ã®äžã«ã°ã©ã¹ãããããšã¯æããŠãããŸãããã°ã©ã¹ãæããæ»ãå§ããŠããããšãæããã®ã¯è§ŠèŠã§ãã
çŸç¶ã®äžäŸãšããŠãXELA Roboticsã¯Automate 2026ã§ããããéšã«30åã®3軞åèŠã»ã³ãµãŒãã€ã³ããå èµããããããæå ãæ«é²ããŸãããå瀟ã¯uSkinã»ã³ãµãŒã0.1ã°ã©ã éçžåœã®åãŸã§æ€åºå¯èœã ãšãè¿°ã¹ãŠããŸãã
äžè»žè§ŠèŠã»ã³ãµãŒãã€ã³ãé åãå èµããuSkinããããæå ãåºå žç»åïŒXELA Robotics
æšè«ïŒã¯ãŒã«ãã¢ãã«ãšãã©ã³ãã³ã°
æšè«ã¯ã次ã«äœãããããæ±ºããæ®µéã§ããæ°ããã·ã¹ãã ã§ã¯ãéåžž2å±€æ§é ã«ãªã£ãŠããŸãã
ã¯ãŒã«ãã¢ãã«ã¯ãããããã«ãè¡åã®åã«æ³åãããããã®ã§ãã
DreamerV3ã¯è¯ãäŸã§ããç°å¢ã®ã³ã³ãã¯ããªã¢ãã«ãåŠç¿ãããã®æ³åäžã®äžçã®äžã§ã»ãŒãã¹ãŠã®æ¹çåŠç¿ãè¡ããŸãã
ç§èªèº«ã®ç ç©¶ã§ãDreamerV3ã䜿ã£ãŠããŸããããã£ãšãé©ããã®ã¯ããšãŒãžã§ã³ããå®ç°å¢ããããé ã®äžãã§ç·Žç¿ããæéã®æ¹ãã¯ããã«é·ãããšã§ããããã¯ãåŠç¿ãéããã¯ããã«äœã³ã¹ãã«ãªãããšãæå³ããŸãã
DreamerV3ã¯ãã¯ãŒã«ãã¢ãã«ã®æ³åããŒã«ã¢ãŠãã§æ¹çãåŠç¿ããŸããåºå žç»åïŒHafner et al., 2023
äžæ¹ããã©ã³ãã³ã°ã¯èšèªã¢ãã«ãæ ãããšãå¢ããŠããŸãã
奜äŸãGoogle DeepMindã®Gemini Robotics-ER 1.5ã§ãé«ã¬ãã«ã®ãã©ã³ããŒãšããŠæ©èœããŸããå°åã®ãªãµã€ã¯ã«èŠåã«åŸã£ãŠå»æ£ç©ãä»åãããããªã¿ã¹ã¯ã§ã¯ãGoogleæ€çŽ¢ã§èŠåã調ã¹ãäœæ¥ãã¹ãããã«åè§£ããåã¹ããããç©çäœæ¥ãè¡ãGemini Robotics 1.5ã®VLAã¢ãã«ã«æž¡ããŸãã
èšèªã¢ãã«ã¯ç®¡çè ãVLAã¢ãã«ã¯å®éã«äœæ¥ãããäœæ¥è ããšèãããšããããããã§ãããã
Gemini Robotics-ER 1.5ã¯ã¿ã¹ã¯ãèšç»ããç©ççãªå®è¡ãGemini Robotics 1.5 VLAã«å§è²ããŸããåºå žç»åïŒGoogle DeepMind, 2025
è¡åïŒæææ±ºå®ãåãã«å€ãã
è¡åã¯ãæææ±ºå®ãåé¢ç¯ã»ã¢ãŒã¿ãŒãžã®ç²Ÿå¯ãªã³ãã³ãã«å€æããæ®µéã§ãéåžžã¯æ¯ç§æ°åãæ°çŸåïŒHzïŒè¡ãããŸãããã®æ®µéã®äœã¬ãã«éšåãå¶åŸ¡ïŒã³ã³ãããŒã«ïŒã§ãã
äŸãã°ãã°ãªãããŒãå·Šã«5cmåããããšããã³ãã³ãã¯åçŽã«èãããŸããã7é¢ç¯ã¢ãŒã ã§ã¯åé¢ç¯ã¢ãŒã¿ãŒã«å¯Ÿããå ·äœçãªãã«ã¯ã«å€æããåäœäžã¯ç¶ç¶çã«åèšç®ãç¶ããã°ãªããŸããã
é£ããã®ã¯ãçŸå®ãããããã®äºæž¬ã©ããã«ãªããªãããšãå€ãç¹ã§ããç©äœã¯æ»ããåºã¯ãããã«äžæŽã§ããã©ã€ã³ããéãããšç §æãå€ãããã±ãŒãã«ã¯äºæž¬ãšç°ãªãæ²ããæ¹ãããŸãã
åªããã³ã³ãããŒã©ã¯ãäºæž¬ãšå®éã®å·®ç°ã«æ°ã¥ããå³åº§ã«è£æ£ããŸãã
ç§ã¯ãéå€ã§æ§é åãããŠããªãç°å¢ã§ã¯ãè¡åïŒå¶åŸ¡ïŒããã£ãšãé£ããæ®µéã ãšèããŸããç¥èŠã®èª€ãã¯èŠçŽãã°ä¿®æ£ã§ããŸãããèšç»ã®èª€ãã¯åèšç»ã§ä¿®æ£ã§ããŸãããå¶åŸ¡ã®èª€ãã¯ãªã¢ã«ã¿ã€ã ã§çºçããŸãã
ãšã³ããã£ãAIã¯ã©ãèšç·ŽããããïŒã·ãã¥ã¬ãŒã·ã§ã³ã®åœ¹å²
ãšã³ããã£ãAIã¯ãã·ãã¥ã¬ãŒã·ã§ã³ãšå®äžçããŒã¿ãçµã¿åãããŠåŠç¿ããŸããã·ãã¥ã¬ãŒã·ã§ã³ãšã¯ãç©ççã«æ£ç¢ºãª3Dãªããžã§ã¯ãã§æºããããä»®æ³ç°å¢ã®ããšã§ãããããã¯çŸå®ã®ããŒããŠã§ã¢ã«è§Šããåã«äœçŸäžåãç·Žç¿ã§ããŸãã
åè¿°ã®ããŒã¿åé¡ãæãåºããŠãã ãããã·ãã¥ã¬ãŒã·ã§ã³ã¯ãç¹ã«æ©è¡ããããã¥ã¬ãŒã·ã§ã³ã®åŒ·ååŠç¿ã«ãããäž»ãªåé¿çã®ã²ãšã€ã§ããÏâã®ãããªåºç€ã¢ãã«ã¯äŸç¶ãšããŠå®ãã¢ã«å€§ããäŸåãããããå€ãã®ããŒã ã¯äž¡æ¹ã䜿ããŸãã
å®äžçã§ã®ããŒã¿åéã«ã¯3ã€ã®å€§ããªåé¡ããããŸãã
ã·ãã¥ã¬ãŒã¿ã¯ããã3ã€ãåæã«è§£æ±ºããŸããåäžGPUäžã§äœåãã®ä»®æ³ããããã䞊å皌åãããå¿ èŠãªã ã倱æã»åéãç¹°ãè¿ããŸããããã¯æ°å¹Žåã®ããããçµéšãæ°æéã«åçž®ããŸãã
è¯ãã·ãã¥ã¬ãŒã·ã§ã³ç°å¢ã®æ¡ä»¶
ãã¹ãŠã®ã·ãã¥ã¬ãŒã¿ãããããåŠç¿ã«çããæçšãšããããã§ã¯ãããŸãããç§èªèº«ãããããã¢ãŒã ãã·ãã¥ã¬ãŒã·ã§ã³ã§æ±ã£ãŠããçµéšãããè¯ãç°å¢ã«ã¯æ¬¡ã®3æ¡ä»¶ãå¿ èŠã§ãã
æãããç®ã«ãããã©ãããã©ãŒã ã¯ãNVIDIA Isaac SimïŒããã³Isaac LabïŒãšMuJoCoã§ãã
ææ°ã®è¿œå ãNewtonã§ãNVIDIAãGoogle DeepMindãDisney Researchãæ§ç¯ããLinux Foundationã®ç®¡çäžã«ããGPUå éã®ãªãŒãã³ãœãŒã¹ç©çãšã³ãžã³ã§ãã
Newton 1.0ã¯2026幎3æã®NVIDIA GTCã§ãªãªãŒã¹ãããŸãããIsaac Labã®ç©çããã¯ãšã³ããšããŠæ¥ç¶ã§ããSolverãšããŠMuJoCo Warpãåããã±ãŒãã«ãåžã®ãããªå¯å€åœ¢ç©äœåãã®è¿œå SolverãæäŸããŸãã
å¯å€åœ¢ç©ã¯èŠãç®ä»¥äžã«éèŠã§ããå€ãã·ãã¥ã¬ãŒã¿ã¯ã±ãŒãã«ãåžã®æ±ããäžåŸæã§ãå·¥å Žã§ã¯ãã®äž¡æ¹ãæ±ãããããããæ±ããããŠããŸãã
ã·ã ã»ãã¥ã»ãªã¢ã«ã®ã£ãã
ã·ã ã»ãã¥ã»ãªã¢ã«ã®ã£ãããšã¯ãã·ãã¥ã¬ãŒã·ã§ã³ã§åŠç¿ããæ¹çã宿©ã«ç§»ããéã«çããæ§èœäœäžã®ããšã§ããšã³ããã£ãã·ã¹ãã æå€§ã®èª²é¡ã®ã²ãšã€ã§ãã
äŸãã°ãã·ãã¥ã¬ãŒã·ã§ã³ã®æ©æŠã¯åžžã«å°ããããã«ã¡ã©ã¯ããããããç©äœã¯çæ³åãããã¡ã§ãããã®ãããã·ãã¥ã¬ãŒã·ã§ã³ã§ã¯95%æåããŠããæ¹çãã宿©ã«åºäŒã£ãé端ã«ç Žç¶»ããããšããããŸãã
ãã®ã®ã£ãããåããäž»èŠãªææ³ã¯2ã€ãããŸãã
åŠç¿äžã®ãã¡ã€ã³ã©ã³ãã å
ã·ãã¥ã¬ãŒã¿ãå®ç§ã«è¿ã¥ããã®ã§ã¯ãªããããŸããŸãªåœ¢ã§ã©ã³ãã åããŸãã
TobinãïŒ2017ïŒã¯ãã¯ã¹ãã£ãç §æã倧ããã©ã³ãã åããå®äžçãã¢ãã«ã«ãšã£ãŠãããã²ãšã€ã®ããªãšãŒã·ã§ã³ãã«èŠããããã«ããŸãããOpenAIã®ã«ãŒããã¯ãã¥ãŒãçšãããããã³ãã¯ãæ¹çãäžéããã«ã€ããŠã©ã³ãã åã®å¹ ãèªåçã«åºããŸããã
ã·ãã¥ã¬ãŒã·ã§ã³åŸã®å®ããŒã¿ã§ã®ãã¡ã€ã³ãã¥ãŒãã³ã°
ã·ãã¥ã¬ãŒã·ã§ã³ã§äºååŠç¿ããæ¹çã¯ãã¿ã¹ã¯ã®ãåãããã§ã«åŠãã§ãããããé©å¿ã®ããã«å¿ èŠãªå®ãã¢ã¯å°æ°ã§æžãããšãå€ãã§ãã
ã©ã¡ãã®ææ³ãã®ã£ãããå®å šã«ã¯æ¶ããªããããçŸå®äžçã§ã®æ§èœãžã®åœ±é¿ãæããã¹ããåããŒã ã¯åãçµã¿ãç¶ããŠããŸãã
2026幎ã®ãšã³ããã£ãAIäºäŸ
ãšã³ããã£ãAIã¯ãåéã«ãã£ãŠæ¿æ·¡ã¯ãããã®ã®ããã§ã«ç 究宀ã®å€ã§çšŒåããŠããŸãã
ç§ã®èŠç«ãŠã§ã¯ã人æã§ã¹ã¯ãªããåããã«ã¯ç°å¢å€åã倧ããããåé¡ãèµ·ããã人éãä»å ¥ã§ããå Žæããé ã«å°å ¥ãé²ã¿ãŸãã
èªåé転è»
èªåé転ã¯ããã£ãšãããŒã¿ãå¿ èŠãšãããšã³ããã£ãAIã®äžçš®ã§ãã
Waymo Driverã¯å®å šèªåã§çŽ2åãã€ã«ãèµ°è¡ããããã«ã·ãã¥ã¬ãŒã·ã§ã³ã§æ°ååãã€ã«ã®åŠç¿ã»ãã¹ããè¡ã£ãŠãããšã2026幎2æã®Waymoã®World Modelã«é¢ããæçš¿ã§è¿°ã¹ãŠããŸããã·ãã¥ã¬ãŒã·ã§ã³ã«ãããå®éã®ã€ã³ã·ãã³ããåçŸããããå®è·¯ã§ã¯ãã¹ãè»äž¡ãééããªããããªã¬ã¢ã±ãŒã¹ïŒäŸïŒé§è»è»äž¡ã®éããåã©ããé£ã³åºãïŒãçæã§ããŸãã
èªåé転ã¯ãç¥èŠã»æšè«ã»è¡åã«ãŒããå šéã§åã£ãŠãã奜äŸã§ããããŸããWaymoã®è»äž¡ã¯ã«ã¡ã©ãã©ã€ããŒãã¬ãŒããŒã§ç¥èŠããæ©è¡è ãè»äž¡ã次ã«åããããªè¡åãæšè«ãããã³ãã«ããã¬ãŒããæ¯ç§äœåºŠãå¶åŸ¡ããŠè¡åããŸãã
å庫ãšç©æµ
ç§èŠã§ã¯ãå庫ã¯çŸåšãã£ãšãèªç¶ãªãšã³ããã£ãAIã®åçšçšéã§ãã
åºã®ã©ã€ã³ã«åŸãåºå®çµè·¯åãããããããåçã§éå€ãªåšåº«ã«å¯Ÿå¿ã§ããã·ã¹ãã ãžã®ã·ãããé²ãã§ããŸããäŸãã°40çš®é¡ã®è£œåãäžç·ããã«å ¥ã£ãããŒãããã®ãããã³ã°ãªã©ã§ãã
Agility Roboticsã®ãã¥ãŒããã€ããDigitãã¯ã2024幎ã«ç· çµãããè€æ°å¹Žå¥çŽã®ããšGXO Logisticsã§ããŒãæ¬éãè¡ã£ãŠãããBoston Dynamicsã®Stretchã¯ãã©ãã¯ããã®ç®±åºããæ ããŸãã
Agility Roboticsã®DigitããAMRãšã³ã³ãã€éã§ããŒããæ¬éãåºå žç»åïŒAgility Robotics/The Robot Report
ãã«ã¹ã±ã¢ãããã£ã¯ã¹
ãã«ã¹ã±ã¢ã¯ããšã³ããã£ãAIããã£ãšãæ éã«é²ãé åã ãšèŠãŠããŸãã
Intuitiveã®da Vinciã®ãããªæè¡æ¯æŽã·ã¹ãã ã¯ããã§ã«äžçäžã®ç é¢ã§äœ¿ãããŠããŸãããæäœè ã¯å€ç§å»ã§ãããã®åéã®AIç ç©¶ã®å€ãã¯ãåšå ·è¿œè·¡ãã«ã¡ã©å¶åŸ¡ãçž«åæ¯æŽãšãã£ãã¢ã·ã¹ãã«çŠç¹ãåœãŠãŠããŸãã
ãã«ã¹ã±ã¢ã§ã¯ãã¢ãã«ã®èœå以äžã«èŠå¶æ¿èªãå¶çŽãšãªãããšãå€ãããããŠããã¯æ£åœã§ããèªåŸæè¡ã«è¿ããã®ãæ¥ãåã«ãã¢ã·ã¹ãããã¬ãŒãã³ã°çšéãå ã«æ®åãããšå人çã«ã¯äºæ³ããŸãã
å·¥å Žã®ãã¥ãŒããã€ããããã
ãã¥ãŒããã€ãã¯æ³šç®åºŠãé«ãäžæ¹ã§ãå®çžŸã¯éå®çã§ãã
CES 2026ã§Boston Dynamicsã¯Atlasã®éç£çãçºè¡šãã2026幎ã«è£œé ããå šå°æ°ã¯HyundaiãšGoogle DeepMindã«ãã§ã«å²ãåœãŠæžã¿ã ãšè¿°ã¹ãŸãããäžæ¹Figureã¯ããµãŠã¹ã«ãã©ã€ãå·ã¹ããŒã¿ã³ããŒã°ã®BMWå·¥å Žã§ãFigure 02ãã¥ãŒããã€ãã«ããæ¿éããŒãã®11ãæã®å°å ¥ãè¡ããŸããã
ççŽã«èšãã°ãå€ãã®ãã¥ãŒããã€ãå°å ¥ã¯ãããŸã éå®ã¿ã¹ã¯ã®ãã€ãããæ®µéã§ããäŸãã°Hyundaiã¯Atlasã2028幎ã«éšåã·ãŒã±ã³ã·ã³ã°ã§äœ¿ãå§ããèšç»ã§ãæå€§æã§ãã£ãŠã極ããŠæ éã«é²ããŠããããšãããããŸãã
2026幎ã«ç¥ã£ãŠããã¹ãäž»èŠãšã³ããã£ãAIäŒæ¥
ã§ã¯ãå®éã«ããããäœã£ãŠããã®ã¯èª°ã§ããããããšã³ããã£ãAIãåããŠåŠã¶æ¹ã«ãæäœéæŒãããŠã»ãã5ã€ã®çµç¹ãæããŸãã
NVIDIA
NVIDIAã¯ãããããæ¬äœãšãããããããããåãå·»ãããŒã«çŸ€ãäž»ã«éçºããŠããŸãã
Isaac Simã¯ãã©ããªã¢ã«ãªã·ãã¥ã¬ãŒã·ã§ã³ããIsaac Labã¯ãã®äžã§ã®åŒ·ååŠç¿ããNewtonã¯ç©çãæ ããŸããNVIDIAã¯æ±çšããããã販売ããŠããªãã«ãããããããå€ãã®ãããã£ã¯ã¹ããŒã ãNVIDIAã®ãœãããŠã§ã¢ã§åŠç¿ããŠããŸãã
Physical Intelligence
Physical Intelligenceã¯ãÏâãšãã33åãã©ã¡ãŒã¿ã®ããžã§ã³ã»èšèªã»ã¢ã¯ã·ã§ã³ã¢ãã«ãæ§ç¯ãããããŒãããã³ã°ã§æŽæ¿¯ç©ããã¿ã®ãããªã¿ã¹ã¯åãã®æ»ãããªã¢ã¯ã·ã§ã³åãçæããŸãã
å®éã«ããŠã³ããŒãã§ããæ±çšããããã¢ãã«ãšããŠãç§ãç¥ãéãæè¯ã®äŸã§ããopenpiãªããžããªããå ¥æã§ããŸãã
ãåºç€ã¢ãã«ãã®æŠå¿µãåããŠã§ããã°ãIntroduction to Foundation Modelsã¬ã€ãããããããã説æããŠããŸãã
Agility Robotics
Agility Roboticsã¯ãDigitã§çãé åã«çµããŸããã
åœåããå庫ã§ã®ããŒãæ¬éã«æé©åãããŠããããã®æç¢ºãªçŠç¹ããä»ã®å€ãã®ãã¥ãŒããã€ãããæ©ãæåã®é¡§å®¢å°å ¥ã«è³ã£ã倧ããªçç±ã§ãã
Boston Dynamics
Boston Dynamicsã¯ãAtlasã§å¯Ÿç §çãªã¢ãããŒããåããŸããã
ïŒãã«ã¯ãŒã«åç»ã§ããªãã¿ã®ããã«ïŒé·å¹Žããããã®éåæ§èœãéçãŸã§é«ãããã®åŸãGoogle DeepMindã®Gemini Roboticsã¢ãã«ã§é§åããèšç»ã®ããªãŒã«é»åã®å·¥å Žè£œåãžãšè»¢æããŸããã
AMI LabsïŒYann LeCunïŒ
AMI Labsã¯Yann LeCunã®ããªæ ç¹ã®ã¹ã¿ãŒãã¢ããã§ã2026幎3æã«10å3000äžãã«ã®ã·ãŒãã©ãŠã³ãã宿œïŒãã¬ãããŒè©äŸ¡35åãã«ïŒãCathay InnovationãGreycroftãHiro CapitalãHV CapitalãBezos Expeditionsãå ±åäž»å°ããNVIDIAãSamsungãåºè³ããŠããŸãã
LeCunã¯ãå°æ¥ã®åãã¯ã»ã«ãäºæž¬ããã®ã¯ç¡é§ãå€ããšé·å¹Žäž»åŒµããŠãããJoint Embedding Predictive ArchitectureïŒJEPAïŒã¯ã代ããã«æœè±¡è¡šçŸç©ºéã§äºæž¬ãè¡ããŸãããã®ã¢ã€ãã¢ã¯ãMetaããã§ã«V-JEPA 2ã§æ€èšŒæžã¿ã§ãã
AMIãæ³šç®ã«å€ããã®ã¯ãåéã®äž»æµãããããŠå€ããŠããç¹ã§ãããã£ãšãããŸã äœãåºè·ããŠããªããããçŸæç¹ã§ã¯å®èšŒæžã¿ææ³ãšããããã倧åã®è³éãä»ããç ç©¶äžã®è³ããšèŠãã®ã劥åœã§ãããã
ããŒã¿ãµã€ãšã³ãã£ã¹ãã®ããã®ãšã³ããã£ãAIïŒã©ãã«è²¢ç®ã§ããã
ãšã³ããã£ãAIã¯ãããã£ã¯ã¹å·¥åŠã ãã®åé¡ã§ã¯ãããŸãããäœæ¥ã®å€ããããã倧åã¯æ©æ¢°åŠç¿ã®ä»äºã ãšèšã£ãŠããã§ãããã
ãã£ãšãçŽçµããã¹ãã«ã¯ä»¥äžã§ãã
ç¥èŠã»æšè«ã»è¡åã«ãŒããèªåã§è©Šã
éå§ã«ããããã¯äžèŠã§ãã
gymnasiumã©ã€ãã©ãªã«ã¯MuJoCoç°å¢ã忢±ãããŠãããããpip install "gymnasium[mujoco]"ãå®è¡ããã°ãã·ãã¥ã¬ãŒã·ã§ã³ããããã§ã«ãŒãå šäœãåãããŸããçŸæç¹ã§ã¯ããã§ãŒã¿ã¯ãæšè«ãã¹ãããã
env.action_space.sample()ã§æ¯åã©ã³ãã ã«è¡åãéžã¶ããããã ãã¿ãã¿ããã ãã§ãããã®1è¡ãåŠç¿æžã¿ã®æ¹çã«çœ®ãæããã®ãããŸãã«åŒ·ååŠç¿ã®åœ¹å²ã§ãããã倧ããªé¡æã«é²ãåã«ããã²è©ŠããŠã¿ãŠãã ããã
ãšã³ããã£ãAIã®éçã¯ïŒ
ãšã³ããã£ãAIã¯æ¥éã«é²æ©ããŠããŸããããªã4ã€ã®åŒ±ç¹ããããç¥ã£ãŠãã䟡å€ããããŸãã
ãŸãšã
ãšã³ããã£ãAIã¯ãæ©æ¢°ã«ç©ççãªèº«äœãšãããã䜿ãããªãç¥èœãäžããŸããç¥èŠã»æšè«ã»è¡åã®ã«ãŒãã§åäœããäž»ãªå¶çŽã¯ç©çããŒã¿ã®äžè¶³ã§ãã2026幎çŸåšãç 究宀ãåºãŠãå庫ãåæã®å·¥å Žãã€ããããžãšé²ã¿ã€ã€ãããŸãã
äœããé©ãã®ã¯ãåãã®åããå€ãã£ãããšã§ããæ°å¹Žåã¯ãLLMããããã£ã¯ã¹ç ç©¶ãäžèŠã«ããã®ã§ã¯ããšåãããŸãããå®éã«ã¯ãLLMã¯ãšã³ããã£ãã·ã¹ãã å éšã®æšè«å±€ãšãªããäž¡åéã¯çžäºã«èåãã€ã€ãããŸãã
èšäºåé ã®ãªã³ãŽãæãåºããŠãã ãããæã¡æ¹ããç¥ã£ãŠãããããšãšãå®éã«ãæã¡äžãããããšã¯ããŸã£ããå¥ã®åé¡ã§ããããšã³ããã£ãAIã¯ããã®åŸè ã«åãçµãåéã§ãã
ãã®åºç€ãšãªãMLãåŠã³ãããªãããŸãã¯Reinforcement Learning in Pythonãã©ãã¯ããå§ããŠãã ãããã¹ã¿ãã¯ã®èšèªåŽã¯ãLarge Language Models (LLMs) Conceptsã³ãŒã¹ããDeveloping Large Language Modelsãã©ãã¯ãè¯ã次ã®äžæ©ã«ãªããŸãã