Summary
Self-driving
Maps
Others
Notable Tweets
Self-Driving
最近Cruise的无人驾驶测试出了大事
事实却很好笑
April morning, around 4 am, a San Francisco Fire Department truck responding to a fire tried to pass a doubled-parked garbage truck by using the opposing lane. But a traveling autonomous vehicle, operated by the General Motor subsidiary Cruise without anyone inside, was blocking its path. While a human might have reversed to clear the lane, the Cruise car stayed put. The fire truck only passed the blockage when the garbage truck driver ran from their work to move their vehicle.
一个you don’t know what you don’t know的情况。
如果说Cruise和Waymo属于1.0的self-driving,现在有一批2.0的公司。区别在于
Here's the new idea. Instead of building a system with multiple neural networks and wiring these together by hand, Wayve, Waabi, and others are each building one large neural network that figures out the details by itself. Throw enough data at the AI and it learns to convert input (camera or lidar data about the road ahead) into output (turning the wheel or hitting the brakes), much like a kid learning to ride a bike.
这种end-to-end的做法还有不同的flavour
Like Wayve, Waabi is using end-to-end learning. It isn’t (yet) using real vehicles, however. It is developing its AI almost fully inside a super-realistic driving simulation, itself controlled by an AI driving instructor. Ghost also adopts an AI-first approach, building driverless tech that not only navigates roads but learns to react to other drivers.
Autobrains is betting on an end-to-end approach too, but does something different with it. Instead of training one large neural network to handle everything a car might encounter, it is training many smaller networks—hundreds of thousands, in fact—to handle a very specific scenario each.
好处是只用一个模型,培训起来很容易。复制性也很强。不像Waymo和Cruise没去一个城市都要花很多时间,因为需要高精度地图
But before Cruise drives in a new city, it first has to map its streets in centimeter-level detail. Most driverless car companies use these kinds of high-definition 3D maps. They provide extra information to the vehicle on top of the raw sensor data it gets on the go, typically including hints like the location of lane boundaries and traffic lights, or whether there are curbs on a particular stretch of street.
高精度地图需要不断的更新。如果路上有很多无人驾驶的车,那就容易了。但目前并没有。
These so-called HD maps are created by combining road data collected by cameras and lidar with satellite imagery. Hundreds of millions of miles of roads have been mapped in this way in the US, Europe, and Asia. But road layouts change every day, which means map-making is an endless process.
很好奇的是,如果采集数据的硬件变了,end to end的模型如何更新。1.0的公司就认为这是他们的优势之一
Cruise has built a vast virtual factory to support its software, with hundreds of engineers working on different parts of the pipeline. ElShenawy argues that the mainstream modular approach is an advantage because it lets the company swap in new tech as it comes along.
2.0的出现似乎在说明软件和GPU的进步超过了别的硬件的进步。
Maps
谈起地图,AR应用也需要一个地图,而且和普通的地图不一样。Niantic的CEO认为精度要求很高
To do AR in the way that I just described, the level of accuracy that you need is more than an order of magnitude greater that you need. So you think about three orders of magnitude more cost to map that in a brute force way, coupled with the fact that it’s not nearly as efficient, if you have to send humans around walking versus cars driving.
Niantic通过Pokemon Go游戏,已经积累了很多用户提交的数据。现在又开放了开发者平台Lightship来采集更多的数据,占有了先发优势。
对于还要进入这个领域的startup来说,这个barrier也许很大。不过,对于苹果或者谷歌来说,不觉得Niantic的这个优势会有多大。当然不要忘了Snap,也是一个有地图和软件的公司。
如果AR最后是一个平台的规模,关键还是在硬件上。
这周,还有一个collaborative的地图公司融资,Felt。
Felt is built to be the home for maps, where they can be shared and worked on together by teammates. But maps are often shared externally, to partners or customers or to the public. And as every new user joins Felt, the product benefits not just from the direct network effects embedded in the multiplayer experience of mapping, but from the data network effects that come from data layers and templates that will get built into the product, to make it more usable for more use cases over time. Every map can tell a story, and no one has enabled the map to be a canvas for storytelling until Felt.
Others
Ultima Genomics号称可以把测序成本降低90%。
Ultima says that its sequencing machine and software platform, the UG 100, can perform a complete sequencing of a human genome in about 20 hours, with precision comparable to existing options, but does so at a far lower cost per “gigabase,” which is to say per million base pairs of DNA analyzed.
There is considerable margin for improvement on the process, primarily in the size and density of the wafer and its surface, leading to improved throughput. This could push the price lower, but for now a 90% reduction is more than enough to go to market with.
日本的研究团队找到一种不错的carbon capture方法
Removes carbon dioxide directly from the atmosphere with unprecedented performance. Isophorone diamine (IPDA) in a “liquid-solid phase separation” system was found to remove carbon dioxide at the low concentrations contained in the atmosphere with 99% efficiency. The compound is reusable with minimal heating and at least twice as fast as existing systems, an exciting new development for direct air capture.
Notable Tweets
备注,我持有的股票包括MQ,PL和PTON。