为了L3自动驾驶 上汽真是拼了！联手Momenta成立3亿元合资公司

yun2025

为了L3自动驾驶 上汽真是拼了！联手Momenta成立3亿元合资公司

机器人控制测试发现智能体会误报任务成功

RoboARCade把长序列操作任务放进同一套MuJoCo仿真环境，涉及三种机器人本体。团队扩大了原任务的初始条件范围，让智能体必须在执行过程中利用记忆、判断和视觉信息。综合测试中，GPT-6 Astra得分49.8%，Claude Fable 5.1为43.6%。更丰富的视觉界面带来1.5至6.8个百分点的改善，使用成本同时增至原来的1.7至2.3倍。 Satvik Sharma等人的这项研究收录在11月6日更新的计算机科学预印本列表中。

错误分析显示，主要困难在于选择操作策略，或把正确策略执行得足够精确。智能体在尝试过的子任务中，有24%至39%出现错误宣告成功的情况。在另一个五任务子集上，提示中加入一段成功示范后，Astra从53.5%提高到73.0%。后者属于子集测试，不能当作整套基准上的得分。团队公开了代码、视频与执行轨迹。

https://lightroom.adobe.com/shares/e87da424e51b6c4facfe0b4b1f799182
https://lightroom.adobe.com/shares/b1d8df81382b6841be63f711b9c1a6ac
https://lightroom.adobe.com/shares/7b2158e3582a9b46a510d21a99e64035
https://lightroom.adobe.com/shares/38fc783af36b7146905fb42ac4aeb119
https://lightroom.adobe.com/shares/92f919021275124a82171e6440cef875
https://lightroom.adobe.com/shares/c510d7117f53364e95f499c9b1a692d7
https://lightroom.adobe.com/shares/19ddd56c64f61d478a5b6d88e55683b7
https://lightroom.adobe.com/shares/bdc37252e7a39e429bd9079d6e9af195
https://lightroom.adobe.com/shares/36a574e92619394b9d4b335014113806
https://lightroom.adobe.com/shares/26afbc77d798194b843410ed92b7deb2
https://lightroom.adobe.com/shares/5c9409d93bcd9648b6d9c043728db8da
https://lightroom.adobe.com/shares/6a72663444000f4f819d9e7f05364867
https://lightroom.adobe.com/shares/a50c46bd59a4bf47a5e342a3f2215e3a
https://lightroom.adobe.com/shares/d67d8871dd464248aff239237512519f
https://lightroom.adobe.com/shares/aa713f1df87e744a9ffddab547f5224d
https://lightroom.adobe.com/shares/72f79ea765a86f4fbfbaeccacc4b93af
https://lightroom.adobe.com/shares/35fbb62fbe45f149b6e62ecab13a2062
https://lightroom.adobe.com/shares/3e3c80366002064ea78b054421717a90
https://lightroom.adobe.com/shares/66a46035fe4f144a91c143d1f258c15f
https://lightroom.adobe.com/shares/49ce867c65860143a71e3fd009f07ef7
https://lightroom.adobe.com/shares/bfd7af56b49d1348a67b6ed6bade1cc6
https://lightroom.adobe.com/shares/8f0d08cfb2a91243babe069d115cdda9
https://lightroom.adobe.com/shares/3704e98ebbdc9b44b0baf57446ce773b
https://lightroom.adobe.com/shares/ef4dba044e6c454ba594f0f4005ae1e3
https://lightroom.adobe.com/shares/414082db8a53b1469bfb3e427c130416
https://lightroom.adobe.com/shares/1f413e096d53c54ba824c7b29f12253d
https://lightroom.adobe.com/shares/3e568629c082964a95e8935bc5b9a954
https://lightroom.adobe.com/shares/642dfa6c95cd054ebec301c688ae442d
https://lightroom.adobe.com/shares/4e045bea5d6b88459b09704acf90e516
https://lightroom.adobe.com/shares/5ec02da228ecff47966eed23829e698f
https://lightroom.adobe.com/shares/f64fae507167d64ea7b53ce9d3089184
https://lightroom.adobe.com/shares/3be2d80c29457449ae0070ce8865e714
https://lightroom.adobe.com/shares/e5e639b37dedf247b73893d4adc02508
https://lightroom.adobe.com/shares/6e8613d971443348a6024bd6b73f3053
https://lightroom.adobe.com/shares/4c3630c0c09870499149e2a678073858
https://lightroom.adobe.com/shares/e34ec5be6157f544ad2829c044442ec1
https://lightroom.adobe.com/shares/7e608b29039d3748ae0506375eb5e52f
https://lightroom.adobe.com/shares/8a7870928807374da264b5c6c78b3601
https://lightroom.adobe.com/shares/3f8c4be102aace4dbae4a59ae312a907
https://lightroom.adobe.com/shares/662001e3b00976489813a8f87e0339bb
https://lightroom.adobe.com/shares/364b09d7c46e3f4e84b7d794117b6b4f
https://lightroom.adobe.com/shares/9911871b5680c442ba64fb6cc37771cb
https://lightroom.adobe.com/shares/bfdf6b51c7bbd642b4048c949b8c43a1
https://lightroom.adobe.com/shares/7313faac6764a644b3a65e86b4f50c1e
https://lightroom.adobe.com/shares/91da07d85669324f83c739bc362d7ce1
https://lightroom.adobe.com/shares/b6df694da4efca4398aaf1a452535692
https://lightroom.adobe.com/shares/9ce391ded132df4a995494e8174dd0cc
https://lightroom.adobe.com/shares/993e099fc809fc43a45c16e7c2c81453
https://lightroom.adobe.com/shares/12d577b3aebb7a4e91a7f6697ddf988b
https://lightroom.adobe.com/shares/9d95101e92310345bad73858f04f186a