小鹏汽车第二代VLA全新版本XOS 6.3.0开启推送Xpeng Motors second-generation VLA new version XOS 6.3.0 begins rollout


AI free token/AI免费积分
首页 > 智车之家>汽车新闻
感谢IT之家网友 Autumn_Dream、不一样的感受 的线索投递!

IT之家 9 月 22 日消息,小鹏汽车官方今日宣布,第二代 VLA 全新版本 XOS 6.3.0 正式启动推送。官方宣称:

全新版本,全面升级。从 3D 空间识别跃迁至 4D 时空理解,凭借更强的时序记忆、轨迹预判与决策响应能力,达成看得更远、想得更深、反应更快、开得更稳。

小鹏汽车第二代VLA全新版本XOS 6.3.0开启推送

另外,小鹏汽车第二代 VLA Lite 同步启动推送,单图灵 Max 车型正式升级第二代 VLA。

小鹏汽车第二代VLA全新版本XOS 6.3.0开启推送

小鹏汽车官方此前在 8 月的小鹏物理 AI 分享暨第二代 VLA 全新版本感受日活动表示,第二代 VLA 基于统一工艺底座,已达成 L2 至 L4 能力贯通,进一步向更高阶自动驾驶拓展。

据IT之家从活动现场了解,小鹏借助学习式 Token 压缩与蒸馏训练,在不牺牲输入信息和模型容量的基础上,以更少的有效 Token 达成高质量视觉理解,将第二代 VLA 基座模型能力部署到算力更低的平台,伴随基座模型能力的增强,蒸馏后的图灵 VLA 2.0 Lite 同样取得能力提高。

依据介绍,小鹏第二代 VLA 全新版本模型能力全面增强。小鹏宣布 X-Foresight 预测世界模型首次上车,可预判将来 6 秒会发生什么,推演周边交通参与者将来的可能行为。

另外,第二代 VLA 全新引入 Infini-VLA 长时序架构,记得前 30 秒的世界,驾驶判断开始具备更长的上下文,更多有效信息进入模型。

本文来源:https://www.ithome.com/1/005/528.htm
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

Home > Smart Car Home > Auto News

Thanks to IT Home netizens Autumn_Dream and 不一样的感受 for their news tips!

IT Home, September 22 news: XPeng Motors officially announced today that the all-new second-generation VLA version, XOS 6.3.0, has officially started rolling out. The official statement says:

All-new version, comprehensive upgrade. Leaping from 3D spatial recognition to 4D spatiotemporal understanding, with stronger temporal memory, trajectory prediction, and decision-response capabilities, achieving seeing farther, thinking deeper, reacting faster, and driving more steadily.

Xpeng Motors second-generation VLA new version XOS 6.3.0 begins rollout

In addition, XPeng Motors' second-generation VLA Lite has simultaneously started rolling out, and single Turing Max models have been officially upgraded to the second-generation VLA.

Xpeng Motors second-generation VLA new version XOS 6.3.0 begins rollout

XPeng Motors previously stated at the August XPeng Physical AI Sharing and Second-Generation VLA New Version Experience Day that the second-generation VLA, built on a unified process foundation, has achieved end-to-end capability spanning L2 to L4, further expanding toward higher-level autonomous driving.

According to IT Home's reporting from the event site, XPeng leverages learned token compression and distillation training to achieve high-quality visual understanding with fewer effective tokens without sacrificing input information or model capacity, deploying the second-generation VLA base model's capabilities to platforms with lower computing power. As the base model's capabilities strengthen, the distilled Turing VLA 2.0 Lite has also achieved capability improvements.

According to the introduction, the all-new version of XPeng's second-generation VLA has comprehensively enhanced model capabilities. XPeng announced that the X-Foresight predictive world model has been deployed in vehicles for the first time, able to predict what will happen in the next 6 seconds and infer the possible future behaviors of surrounding traffic participants.

Additionally, the second-generation VLA newly introduces the Infini-VLA long-temporal architecture, remembering the world from the previous 30 seconds, giving driving judgment a longer context with more effective information entering the model.

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry