Wayve英国领先的自主驾驶全端AI开发者

4.6 (5)

审阅者 Daniel Nikulshyn·更新 2026年5月

Autonomous Driving Deep Learning Embodied AI Foundation Models Computer Vision Automotive Research

概览

Wayve 是一家总部位于伦敦的公司，采用端到端深度学习方法构建自动驾驶技术。它的系统不依赖详细的 HD 地图和手工编码规则，而是直接从摄像头输入和真实道路驾驶数据中学习驾驶，旨在实现跨城市和不同车型的泛化能力。该公司开发具身 AI 模型，包括其 AV2.0 平台以及像 GAIA 和 LINGO 这样的基础模型，这些模型融合了视觉、语言和动作。Wayve 与汽车制造商和车队运营商合作，将其驾驶智能引入乘用车和商用车，测试已在英国及其他地区开展。 Wayve面向汽车OEM、出行服务提供商和AI研究人员，定位为传统模块化自动驾驶系统堆栈的可扩展替代方案，优先考虑学习行为和适应性，而非受限于地理围栏的部署。

主要功能

全端深度学习驾驶堆栈
GAIA生成式世界模型
LINGO视觉-语言-动作模型
无地图、摄像头优先视觉
从多样化驾驶数据中学习车队
与制造商合作进行集成

价格

模型: Freemium
分类: Task automation
评分: 4.6 / 5 (5)

使用场景

无地图自主驾驶技术

制造商将Wayve的全端驾驶堆栈集成到乘用车中，实现了无依赖于详细的HD地图或手编排的规则的自治驾驶。

商业车队的自治驾驶

移动服务提供商和车队运营商部署Wayve的AV2.0平台，运用摄像头优先的完全自主驾驶技术为配送和叫车车辆带来自治驾驶

利用GAIA和LINGO进行身体AI的研究

AI研究人员运用Wayve的GAIA生成式世界模型和LINGO视觉-语言-动作模型为其在身体AI和多模态AI中的工作做出了贡献

横跨城市驾驶的普遍适用

从多样化的实际驾驶数据中学习车队，以运用在未来的城市和车辆平台上

优点 & 缺点

优点

全端学习减少了对HD地图的依赖
旨在在各个城市和车型上普遍适用
强大的在身体AI中的研究输出
受到主要汽车和技术投资者的支持

缺点

并非适用于普通消费者的产品
在规模上的实际部署仍具有局限性
各地区的法规审批存在差异
黑盒模型可以更难验证

评测

4.6

5 个评分的平均值。

登录以留下评测。

Leila Hassan

Jan 18, 2026

Use it every day

Honestly didn't expect to like it this much. End-to-end deep learning driving stack is exactly what I needed, and designed to generalize across cities and vehicles. but I reach for it almost every day now and it just clicks.

Tomáš Novák

Jan 3, 2026

Use it every day

Honestly didn't expect to like it this much. Fleet learning from diverse driving data is exactly what I needed, and backed by major automotive and tech investors. but I reach for it almost every day now and it just clicks.

Marcus Bell

Dec 27, 2025

Years in this space

I've evaluated a lot of these over the years. What stands out here is lINGO vision-language-action model — handled better than most — and strong research output in embodied AI. Worth the time if this is your use case.

Diego Fernández

Jun 30, 2025

Compared a few options

Evaluated this against two competitors. Where it wins: partnerships with automakers for integration and backed by major automotive and tech investors. Where it lags: regulatory approval varies by region. On balance the feature set — especially gAIA generative world model — justifies the 4 stars for our use case.

Robert Ainsworth

Jun 23, 2025

Does the job

Pretty happy overall. Map-free, camera-first perception just works and designed to generalize across cities and vehicles. Not a product available to general consumers can be annoying, but no dealbreakers — I'd recommend it to a friend without hesitating.

问答

Who is Wayve intended for, and can individual consumers use it?

Wayve targets automotive OEMs, mobility and fleet operators, and AI researchers. It is not a product sold to general consumers; instead, the company partners with automakers to integrate its driving intelligence into consumer and commercial vehicles.

How does Wayve's approach differ from traditional autonomous driving stacks?

Wayve uses an end-to-end deep learning stack that learns to drive directly from camera input and real-world data, avoiding HD maps and hand-coded rules. This map-free, camera-first design is intended to generalize across different cities and vehicle types.

What are the main limitations to consider before partnering with Wayve?

Real-world deployment remains limited in scale, with testing primarily in the UK and select regions, and regulatory approval varies by market. Its end-to-end models can also be harder to validate than modular stacks due to their black-box nature.