$比亚迪(SZ002594)$ $特斯拉(TSLA)$ $英伟达(NVDA)$ 来自英国的人工智能初创公司Wayve获得了来自英伟达、微软和软银参与的C轮融资,融资金额为10.5亿美元。
Wayve联合创始人兼首席执行官亚历克斯·肯德尔表示,“本次注入的资金,是对公司自2017年起,将人工智能与自动驾驶融合发展方式的最大肯定。”，说特斯拉FSD都是跟在他的屁股后面的。
Wayve所研发的自动驾驶系统,是基于人工智能框架下的。因此其也被业内比喻为车载版的ChatGPT。2018年起,Wayve已开始在英国公共道路上开启路测,并与Asda和Ocado Group等英国货运公司合作。对此,软银的高管松井健太郎,“这种全新技术是具有革命性的。”而本次融资,将支持Wayve加速推出基于量产乘用车可用的AI自动驾驶产品。
自 2017 年成立以来，Wayve 就一直投入端到端深度学习，并宣称自己是第一个在公共道路上开发和测试仅基于深度学习自动驾驶系统的公司。
相比传统的自动驾驶系统依赖于编程方法或其他 AI 技术，Wayve 完全依赖于深度学习模型来进行决策和操作。Wayve CEO 亚历克斯·肯德尔（Alex Kendall）告诉媒体，他们公司的做法在早期受到其他公司高管的怀疑，但随着当前自动驾驶领域遇到困难和 ChatGPT 等 AI 技术的不断进步，Wayve 的方式开始获得行业关注。
中国自动驾驶产业创新联盟调研员高超认为，Wayve的路线与特斯拉的FSD（Full-Self Driving完全自动驾驶）路线非常相似。“公司最初推出过以模块化架构为核心的自动驾驶系统AV1.0，但依赖手写规则的系统具有很强的局限性，这使得Wayve没有放弃关注和研发端到端的AI算法方向。直到迭代到AV2.0版本后，系统不仅可以自监督式学习，从未标记的数据中训练驾驶技能，还对传感器组合、高精度地图、车辆型号都没有要求。如此一来，一方面为公司节省大量搭建、管理带标记数据集的成本，同时由于具有良好的泛化性，极大地加强了与车企合作搭载的可能性。”
AV2.0 introduces a rapid, continuous, and seamless fleet-learning loop: recording data, training models, evaluating performance, and deploying updated models.
AI capabilities and tools
完全基于深度学习的方法可以自主学习处理各种情况，能更好地适应新的地点和道路上出现的突发事件。该种系统旨在实现完全自动化的驾驶，无需高清地图，仅通过摄像头掌握复杂环境，理论上可以适应全球任何地点。
为了打造更高适应性和决策能力的 AI 系统，Wayve 推出了视觉-语言-行动模型 LINGO-2。该模型由视觉模型和语言模型两大模块组成。视觉模型负责处理摄像头捕捉的连续图像，语言模型基于这些输入预测驾驶轨迹和生成评论文本，随后由车辆控制器执行这一驾驶轨迹。
通过将视觉输入与语言输出相结合，LINGO-2 不仅能实现通过语言指令调整驾驶行为，还具备实时问答能力，可以在驾驶过程中即时回答有关场景问题和提供决策解释，这种实时交互增加了用户对系统的理解和信任。
全部讨论
这家公司为了向OEM销售这套E2E无人驾驶系统，保持独立性，不接受OEM的投资，特斯拉FSD对外licensing遇到对手了：$特斯拉(TSLA)$ $英伟达(NVDA)$ $比亚迪(SZ002594)$
In a Q&A with Hyperdrive, Alex Kendall, Wayve’s co-founder and chief executive officer, compared and contrasted the tack he and Elon Musk are taking to developing this technology. Here are the highlights from the conversation, which are edited and condensed for length and clarity:
It’s a big funding round. How long have you been working on it?
The answer is, a decade. From some of the early AI breakthroughs we made at Cambridge University that showed how it was possible to work 网页链接{without an HD map} and then building out the company. We still had quite significant runway from our Series B, but decided in the fourth quarter to go out and fundraise because the technology got to a point that you saw today.
What will you spend the money on?
People and compute. We’re not going to blitzscale the company. We are disciplined. One of the things that’s been important is making sure we’re being smarter and not harder in terms of we don’t just brute-force solutions — we look for innovative, creative ways to build things efficiently. So we’ve got a fairly lean team today relative to others in the space.
Our big push into automotive came when we were really happy that the tech started to work, call it, a year ago. We’re very careful not to overhype. All of this is setting the right expectations.
How do you plan to deploy the technology?
We 网页链接{don’t want to own and operate} the vehicles. We want to make sure that we can build the safest and more generalizable embodied AI here. By aggregating data over many manufacturers, we can train an AI that is safer than any one can do on their own. And so, we very much want to take a partner-first approach here. We want to build an open platform that can support and give the automotive industry flexibility and choice of how best they want to take advantage of this AI technology.
The best manufacturers are 网页链接{investing in data collection} today. We can’t succeed on our own, and neither can the automotive industry – we’ve both got different expertise. The systems and product manufacturing expertise of carmakers and the data understanding and AI on our side bring these things together.
Did automakers want to invest?
We saw a lot of interest for investment from OEMs, but at this stage, because we are partnering with a number and endeavoring to build an AI that can produce the most and generalizable performance by working with a number across the industry, we wanted to remain neutral in that regard. So we saw a lot of interest for it but we’ve decided to, rather than partner up with one OEM, work more collaboratively with the industry.
How far away are you from launching the tech with carmakers?
We’re not releasing specific launch timelines at this stage. But we’re busy working with a number of manufacturers toward this point and it’s going to be an exciting day when it comes.
Why will your approach win out?
We’ve been all in on an end-to-end AI approach since we started in 2017, and were unique in doing so. It was fun to see Tesla last year pivot to this approach. It’s nice for them to join the club. We’re seeing more and more evidence that AI is what’s going to enable the benefits of autonomous driving to come to the world — certainly, that’s the trajectory we see things on. But the hundreds of billions of dollars that have been invested in this space 网页链接{have gone the other way} so far.
Do consumers want driverless cars?
The experience is magical when you sit in it and try it. It’s pretty clear the errors humans make while driving, and the stress it causes. This technology can really evolve. The key thing is that we need to deploy it responsibly, so making sure we have the appropriate driver-monitoring systems that mean it’s not misused.
But the experience is remarkable. I think consumers haven’t yet had the ability to taste this at scale, but it’s clear that the stress, the inefficiency, the risk and safety challenges that we see in driving, these are real value-adds for society, and it’s an imperative we get this technology out quickly.
Is regulation moving quickly enough?
Regulation is moving well. It’s awesome to see 网页链接{the bill} in the House of Commons in the UK at the moment and weeks away from coming through, so we’re really pleased to see that. In general, the world is moving quickly to enable this technology, because it brings such a transformational impact to society.