Robotics in real world

Self-driving cars. Waymo Part 2

Welcome to the second part of the series, I am glad you could join us. And if you were not with us in the first part do so now. But all the same welcome!

In the second part, we are going to talk in length about hardware behind the Waymo, not talking about the design of the car itself – we wouldn’t expect them to design from scratch, right?

Remember the sequence that is constantly being run to build the brain of the self-driving car?
Get data, determine right controls and send to the actuator to give correct control.

Get data from the environment

Self-driving cars don’t just come to be in a day, they learn how to drive, the rules of the road, the ethics of driving before they can be said to be self-driving etc.

The source of everything is the data gotten from the environment and sensors are responsible for encoding the data from the environment.

But the sensor although many are limited in this case (self-driving car) because they need to above everything else be: accurate and available in real-time.

No margin for error, else lives may be lost!

The iconic sensor that Waymo has chosen for its self-driving car is a 360 DEGREE LIDAR SENSOR, though they still use a Radar and a Camera.

The 360 Light Detection and Ranging (LIDAR) SENSOR

A lidar sensor uses light from a laser beam to measure distances. The differences in return time and wavelength gives a set of 3D coordinate system (x, y, z), called points cloud, and these points cloud can be used to construct 3D representation of the target.

built from a sets of point cloud read by a lidar sensor
built from a sets of point cloud read by a lidar sensor

A typical LIDAR contains: Laser, the source of light, detector, Optics and Rotating Mechanism: Needed to give a 360-degree view of the object, Timing Electronics, and the Computing system

For the Waymo….

Waymo indicated that they have dropped their previous LIDAR sensor (Velodyne HDL 64E) for a new one custom built for the purpose as the previous one cost $75,000 per one accumulating the cost their model in 2017 to $150,0000. But promised that the new custom built one cuts it by 90%!

But since the new LIDAR sensor has not being unveiled to the public. Let’s us gauge the previous one they dropped and foresee how powerful the present one would be.

The previous one is the VELODYNE HDL 64E LIDAR sensor.

Velodyne HDL 64E
Velodyne HDL 64E

And this sensor:

  1. Has 64 channels (source of light beams)
  2. Has 64 data lines (to help with the intricate details necessary)
  3. Has a range of 120 m (meaning it can draw a circle of radius 120m)
  4. Produces an estimate of 2.2 million points per second (like saying x1, y1, z1 2.2 million times in a second)
  5. Has a 360-degree rotating mechanism

To me that’s freaking powerful, don’t you think it is?

But Waymo custom-made lidar sensor that caused the rift between them and Uber is advertised to have a range of 3 football field… no comments

At the tail end, the LIDAR sensor’s ultimate aim is to produce a very accurate 3d coordinate system to help build the 3d image of the scanned area…a tiny slack can cost lives, and that’s why it needs to be almost a beast.

Radar and Camera

Radar

Though the LIDAR sensor is the boss, the google self-driving car still uses the Radio and Detection and Ranging (RADAR) for measuring how far objects are. The distance is measured by the time of return of radio waves when they bounce off obstacles.

A typical radar has a horizon (its reach) to be 19km at an altitude of 23 m.

Camera

The Waymo car also has a camera mounted on it to help to recognize visual signals which include but are not limited to traffic signals, human.
The camera is a crucial part of the computer vision system for the Waymo and any image recognition system.

Where do all the data go?

To the server.

The data that is collected needs to be store somewhere, and that place for the Waymo self-driving car is the trunk of the vehicle.

The server has its own cooling system as it has to store a data of about 1GB of sensor data per second and 2 million GB of data per year.

That’s huge!

But remember, we are trying to build a brain that can drive. Not a small feat.

Takeaways

The LIDAR sensor is the boss, and it is used to build the 3d representation of the target.

Though the LIDAR sensor is the boss, it still uses the radar to see afar and big things.

It uses the camera to recognize visual signals.

Conclusion

We have come to the end of the second section where we talked about main hardware used in the Waymo.

My focus has been on the vision part as I believe that is where the ability to drive lies.

Join me on the final part

please check out the underlying discussion about autonomous cars in part 1 or jump straight to the software discussion in part 3

Leave a Reply

Your email address will not be published. Required fields are marked *