Name: 让我们备份一下猫穿越（Let’s Back Up: Cat Crossing）
Uploaded: 2021-05-24T02:51:14.000Z
Duration: 2 min 8 s
Description: 【看影片學英語】數萬部 YouTube 影片，搭配英漢字典即點即查，輕鬆掌握單字發音與用法，長久累積看電影不必再看字幕。

Having a Waymo Driver that can respond to things in the

情態副詞

real world as quickly as possible is really important for

safety critical situations. Hey, my name is Jonathan. I'm

a Product Manager at Waymo on the Perception team. O

work on machine learning models that allow the Waymo

Driver to see, hear, and understand the world around it.

Today, I'm going to be taking you through a case where the

Waymo Driver is driving through the Phoenix area and a

cat darts out from underneath the vehicle on the right.

Let's back up and see what went into it. The Perception

team's responsible for taking in raw data from things

like camera images, lidar point clouds, and radar

returns and turns them into logical concepts. Like there's

a person here at the crosswalk or there's a vehicle in

this lane traveling at 22 miles per hour. What we can see

on this camera view is the Waymo Driver cruising through

suburban Phoenix and on the right, all of a sudden, a cat

jumps out from underneath the vehicle. There's actually a

few challenging pieces about this case. One is that the

object is very small. Another is that it's moving at quite a

high speed and a third is that it's quite low to the ground

The key thing I want to highlight here is actually how

little time the Waymo Driver needs to react. In this case,

that would mean the time between the cat appearing up from

underneath the vehicle to the time at which we started

applying brakes for it. It's especially challenging

because we have an occlusion. There are lots of challenges

that come with building a low latency perception system

like this. One way that we try to do this is by

reporting objects as quickly as possible. Even when our

detectors aren't quite ready to say what it is yet. By

doing this, we allow downstream components like planner to

know that there's an object there earlier instead of

waiting for a more detailed object confirmation. In this

example, we're able to react in less than half a second

which thankfully was fast enough to allow the cat to

safely cross. Part of building the safest driver means that

we want to build a reliable, consistent, and responsive

driving system. How do I feel about cats? I'm slightly