Tesla launches FSD v15 in the US with Reality View
According to IT之家, Tesla launched FSD v15 in the US with Tesla Reality View, a new interface that generates a virtual view of surrounding roads in real time using the vehicle’s AI computer. The interface supports full-screen, map and split-screen views, and users can select a destination in the visualization to guide the vehicle there. Tesla AI chief Ashok Elswamy said the neural network expanded from about 1 billion to about 10 billion parameters. Tesla also said its upgraded vision encoder is shared by FSD, Robotaxi and Optimus and was trained on millions of internet videos.
The release combines a substantially larger driving neural network with a real-time AI-generated visualization interface, making it a notable update to Tesla’s autonomous-driving software stack.
特斯拉FSD(全自动驾驶系统)最新版本新增了一项颇具趣味的功能,允许用户像在《GTA》中畅游莱昂尼达州一样,围绕车辆自由旋转虚拟摄像机。考虑到车辆需要实时生成周围环境的画面,这项功能显得相当有意思。此外,系统还会显示一个悬浮箭头,提前告知车辆准备驶出环岛的哪个出口。
这项功能名为Tesla Reality View,现已随FSD V15向早期体验用户推送。此前,车辆的泊车界面仅通过轮廓线显示周围物体,而现在系统能够实时重建周围街道环境,并展示FSD计划采取的转向或泊车动作。用户可以选择全屏显示、与导航地图分屏显示,或者直接切换回地图。此外,新版本还加入了更安静的车内模式,转向灯提示音也会在响起几次后逐渐减弱。
特斯拉对FSD V15最引人注目的宣传是安全性提升,但这一说法仍有待进一步验证。该公司声称,新版本在需要进行关键干预的情况下,安全性提升了25倍。此次升级的视觉识别软件经过数百万条互联网视频训练,相关技术也被应用于Robotaxi和Optimus项目。