Discuz! Board

 找回密码
 立即注册
搜索
热搜: 活动 交友 discuz
查看: 13|回复: 0

GPT-6.1 Sol参与活体小鼠脑手术:连续操作三次 定位误差减半

[复制链接]

7万

主题

7万

帖子

21万

积分

论坛元老

Rank: 8Rank: 8

积分
219781
发表于 4 天前 | 显示全部楼层 |阅读模式
GPT-6.1 Sol参与活体小鼠脑手术:连续操作三次 定位误差减半

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://v.youku.com/v_show/id_XNjU3ODI5MDAyOA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4OTk4NA==.html
https://v.youku.com/v_show/id_XNjU4Nzk5MTM2OA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4NTU4NA==.html
https://v.youku.com/v_show/id_XNjU3ODI4ODc4MA==.html
https://v.youku.com/v_show/id_XNjU4Nzk5MTI5Ng==.html
https://v.youku.com/v_show/id_XNjU4Nzk4Mzk4MA==.html
https://v.youku.com/v_show/id_XNjU3ODI4ODcyOA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4ODQwOA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4NzAwNA==.html
https://v.youku.com/v_show/id_XNjU3ODI4OTgzMg==.html
https://v.youku.com/v_show/id_XNjU4Nzk3NzU1Ng==.html
https://v.youku.com/v_show/id_XNjU3ODI4ODY0OA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4MjMwNA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4MDY5Ng==.html
https://v.youku.com/v_show/id_XNjU4Nzk4NTM5Mg==.html
https://v.youku.com/v_show/id_XNjU4Nzk4MjI1Ng==.html
https://v.youku.com/v_show/id_XNjU3ODI4ODU2OA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4OTczMg==.html
https://v.youku.com/v_show/id_XNjU4Nzk4MjIxNg==.html
https://v.youku.com/v_show/id_XNjU4Nzk4Njg0MA==.html
https://v.youku.com/v_show/id_XNjU4Nzk3NzQwMA==.html
https://v.youku.com/v_show/id_XNjU4Nzk3NzM2MA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4Njc2MA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4MDQ4NA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4NTA4OA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4MjA1Ng==.html
https://v.youku.com/v_show/id_XNjU4Nzk4NjY4NA==.html
https://v.youku.com/v_show/id_XNjU4Nzk3OTA1Mg==.html
https://v.youku.com/v_show/id_XNjU3ODI4ODM1Mg==.html
https://v.youku.com/v_show/id_XNjU4Nzk4NTA2MA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4MzUyNA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4MTk2MA==.html
https://v.youku.com/v_show/id_XNjU3ODI4ODIyNA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4MzQ1Mg==.html
https://v.youku.com/v_show/id_XNjU4Nzk3NzE1Ng==.html
https://v.youku.com/v_show/id_XNjU3ODI4NzAwNA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4MzQyNA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4MTgzMg==.html
https://v.youku.com/v_show/id_XNjU4Nzk4NjQ2NA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4NjQ0NA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4NDkxNg==.html
https://v.youku.com/v_show/id_XNjU4Nzk4NjQwNA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4NDg3Ng==.html
https://v.youku.com/v_show/id_XNjU3ODI4NjkxNg==.html
https://v.youku.com/v_show/id_XNjU4Nzk3NDM0MA==.html
https://v.youku.com/v_show/id_XNjU4Nzk4MzI5Ng==.html
https://v.youku.com/v_show/id_XNjU4Nzk4NDgzMg==.html
https://v.youku.com/v_show/id_XNjU4Nzk3Njk3Ng==.html
https://v.youku.com/v_show/id_XNjU3ODI4NTU0MA==.html
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则

Archiver|手机版|小黑屋|DiscuzX

GMT+8, 2026-11-15 12:28 , Processed in 0.049140 second(s), 18 queries .

Powered by Discuz! X3.4

Copyright © 2001-2020, Tencent Cloud.

快速回复 返回顶部 返回列表