yun2025 发表于 9 分钟前

拓斯达发布三套工业场景解决方案,推出双足人形机器人Star

拓斯达发布三套工业场景解决方案,推出双足人形机器人Star

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_XNjU4MDI0NTI3Mg==.html
https://v.youku.com/v_show/id_XNjU4MDI0NTI2NA==.html
https://v.youku.com/v_show/id_XNjU4MDI0ODA0OA==.html
https://v.youku.com/v_show/id_XNjU4MDIzOTExNg==.html
https://v.youku.com/v_show/id_XNjU4MDIzNzUzMg==.html
https://v.youku.com/v_show/id_XNjU4MDI0MzQwOA==.html
https://v.youku.com/v_show/id_XNjU4OTkxOTM5Mg==.html
https://v.youku.com/v_show/id_XNjU4MDIzOTAwNA==.html
https://v.youku.com/v_show/id_XNjU4MDIzNzI3Mg==.html
https://v.youku.com/v_show/id_XNjU4OTkxNTE5Ng==.html
https://v.youku.com/v_show/id_XNjU4MDIzODg1Mg==.html
https://v.youku.com/v_show/id_XNjU4OTkxOTIwNA==.html
https://v.youku.com/v_show/id_XNjU4OTkxNTE3Mg==.html
https://v.youku.com/v_show/id_XNjU4MDIzNzIyMA==.html
https://v.youku.com/v_show/id_XNjU4OTkxNTEyMA==.html
https://v.youku.com/v_show/id_XNjU4MDIzODc1Mg==.html
https://v.youku.com/v_show/id_XNjU4MDIzNzA3Ng==.html
https://v.youku.com/v_show/id_XNjU4OTkxMzM5Mg==.html
https://v.youku.com/v_show/id_XNjU4OTkxNzcwOA==.html
https://v.youku.com/v_show/id_XNjU4MDIzNTU1Mg==.html
https://v.youku.com/v_show/id_XNjU4OTkwNzE5Mg==.html
https://v.youku.com/v_show/id_XNjU4OTkxMTYxMg==.html
https://v.youku.com/v_show/id_XNjU4OTkxMDExMg==.html
https://v.youku.com/v_show/id_XNjU4OTkxMDA2MA==.html
https://v.youku.com/v_show/id_XNjU4OTkwNzExNg==.html
https://v.youku.com/v_show/id_XNjU4OTkxMTQyMA==.html
https://v.youku.com/v_show/id_XNjU4OTkwNzAzMg==.html
https://v.youku.com/v_show/id_XNjU4OTkxMTMwOA==.html
https://v.youku.com/v_show/id_XNjU4MDIzNDA2MA==.html
https://v.youku.com/v_show/id_XNjU4MDIzMjc0OA==.html
https://v.youku.com/v_show/id_XNjU4OTkxMjkwNA==.html
https://v.youku.com/v_show/id_XNjU4OTkxNDQ3Mg==.html
https://v.youku.com/v_show/id_XNjU4OTkxMjkwMA==.html
https://v.youku.com/v_show/id_XNjU4MDIzMTAwNA==.html
https://v.youku.com/v_show/id_XNjU4OTkwOTc1Ng==.html
https://v.youku.com/v_show/id_XNjU4MDIzMDg3Mg==.html
https://v.youku.com/v_show/id_XNjU4MDIzMjUxMg==.html
https://v.youku.com/v_show/id_XNjU4OTkwNTI0MA==.html
https://v.youku.com/v_show/id_XNjU4MDIyOTYwMA==.html
https://v.youku.com/v_show/id_XNjU4MDIzMjMxNg==.html
https://v.youku.com/v_show/id_XNjU4MDIzMjMwOA==.html
https://v.youku.com/v_show/id_XNjU4OTkwNTIwMA==.html
https://v.youku.com/v_show/id_XNjU4MDIyOTUyMA==.html
https://v.youku.com/v_show/id_XNjU4MDIzMjIwOA==.html
https://v.youku.com/v_show/id_XNjU4OTkwMzM2NA==.html
https://v.youku.com/v_show/id_XNjU4MDIyNzY1Ng==.html
https://v.youku.com/v_show/id_XNjU4OTkwMDU4MA==.html
https://v.youku.com/v_show/id_XNjU4MDIxOTgzNg==.html
https://v.youku.com/v_show/id_XNjU4MDIyNDQyNA==.html
https://v.youku.com/v_show/id_XNjU4MDIyNzQ2NA==.html
https://v.youku.com/v_show/id_XNjU4MDIzMDUwNA==.html
https://v.youku.com/v_show/id_XNjU4MDIyNDM4MA==.html
https://v.youku.com/v_show/id_XNjU4MDIyNzM0OA==.html
https://v.youku.com/v_show/id_XNjU4MDIyNDI4MA==.html
https://v.youku.com/v_show/id_XNjU4OTkwMDI1Ng==.html

瓦罗伦 发表于 6 分钟前

路过,支持一下啦
页: [1]
查看完整版本: 拓斯达发布三套工业场景解决方案,推出双足人形机器人Star