# Jetson RACECAR Part 4 – Discussing Platform Mounting and Sensor/Component Selection

In the fourth part of our Jetson RACECAR build, we are going to discuss platform mounting where the computer, sensors and other components will be placed. Also, we’re going to talk about which sensors we might use in building the project. Looky here:

## Background

There are a few areas that we are going to explore on this project. The first, as we talked about earlier, is to recreate the MIT RACECAR that has been constructed for the 2016 season. In the coming weeks, MIT will be releasing the drawings, documentation and software for their robot. The bill of materials (BOM) for the parts is ~ $3500 USD give or take which includes a few of custom laser cut and 3D printed parts. The second area is an exploration of how to take the overall design of the RACECAR and reduce it to more minimal experience. The purpose of this is two fold, a curiosity as to determine which sensors, mechanisms and the amount of computational capability are actually necessary to get good performance from the car. The other part is to put a different budget constraint on the project, let’s say around$1500 USD, to make it more accessible.

Another exploration is to figure out what is actually needed to be able to build one of these robots. Ideally the robot would be assembled with a minimum of fuss with the right parts as a weekend project. The actual mechanism itself isn’t that complicated, so things should be fairly straight forward.

Since we’re a little ahead of the MIT release, this is a good time to examine some different alternatives of how we might put together a minimalist Jetson RACECAR of our own. With that said, I do have some information about the 2016 MIT RACECAR, so that should be useful.

## Platform Mounting

Using the TRAXXAS Rally as a base, a platform is mounted above the chassis to hold the computational, electronic and sensing devices of the robot. Here are three ways to accomplish that task:

• Use standoffs. There are several M3 machine screws which hold the Nerf bars to the chassis. These screw holes would act as mounting points for 4 standoffs spread around the car, with the platform then attached to the standoffs. This is how the 2015 MIT RACECAR was assembled.
• Mount the platform on modified TRAXXAS body mounting posts. Many parts on TRAXXAS cars are interchangeable and are used between different models. The platform could be attached to carefully selected body mounts with little modification.
• Screw the platform directly into the cars suspension towers. This is probably the easiest, but requires relatively accurate aligned holes between the chassis and the platform.

For initial exploration, let’s mount the platform onto TRAXXAS body mounting posts. In the video, a 3/16″ thick, 9.5″ x 17.5″ impact resistant acrylic platform was used, mainly because it was left over from another project. As is typical in most of these types of projects, we’ll build a quick mock up using available resources before investing in the final design elements.

As you have probably already guessed, one of the important aspects of this design is impact resistance. This is because of the inevitable crashes that occur when testing out new steering and path planning algorithms. The electronics on the platform are expensive, and need protection when things go a little wrong. One of the findings from last years MIT car was that using 1/4″ ABS provided better impact resistance than the more brittle acrylic. Both are relatively easy to machine. We will be using 1/4″ ABS when we get past the mock up stage.

In the video, two Traxxas 6815R Body Mounts were used along with some Hex Socket Button Head Screws, 3x8mm. A body mount was attached to the front suspension tower, the other was attached to the rear suspension tower after removing the stock body mounts.

As noted in the video, the platform is not level. We’ll plan on adding some spacers to the front body mount to level the platform. Holes in the body mount will serve as mounting points for the platform. We’ll also add a second platform on top of the first using standoffs to act as a roll cage structure in case things get a little upside down.

## Computers and Power

Either a NVIDIA Jetson TK1 Development Kit or a NVIDIA Jetson TX1 Development Kit can be used. There are slight variations that will be needed for wiring and mechanicals depending on the Jetson being used. In this series, we’ll probably make up a couple of different configurations of platforms for comparison, but will use a Jetson TK1 for the first mock up which will probably get beaten up a little. Platform mounting of the Jetsons will be on standoffs.

For power, a 3S Lipo battery will be used to power the Jetson. In the first, simplest version of the Jetson RACECAR we will power the Jetson with the battery, and any peripherals will draw power from the Jetson. This should be adequate for one USB camera and interfacing with few I2C devices.

## Sensors

On the 2016 MIT RACECAR there are three main sensors, an Stereolabs ZED camera, a Occipital Structure Sensor, and a Hokuyo LIDAR. As mentioned in the video, the Hokuyo LIDAR is relatively expensive; that part alone doubles the price of the car. With that in mind, one of the interesting experiments to perform is to determine how much better the robot car performs with the LIDAR installed versus without.

Remembering that the MIT RACECAR challenges are held indoors, outdoor performance of the sensors is an unknown. Typically IR based sensors like the Structure Sensor have issues with sunlight (the IR emitted by the sun tends to blind the sensor), so the ZED feels like a natural candidate for outdoor use. I happen to live in Southern California during a drought, and there is a park nearby which looks like a good place for working with the RACECAR. That being the case, the ZED will be the first sensor with which we will experiment.

One issue is that the car is basically blind to objects less than a foot or two away from the car. In the video I proposed using a LIDAR-Lite sensor. There is an issue with this selection in that Garmin acquired the company that is manufacturing the device. During the acquisition, the LIDAR-Lite became unavailable, but should become available soon. Ultrasonic sensors, or a TeraRanger One may be worth exploring as a replacement.

## Electronic Speed Control (ESC)

On the MIT RACECAR, an open source ESC is being used which allows better control of the car motor. As the course that the MIT RACECAR is used in is about control theory, it makes sense that a different ESC is being used. My current thought on the stock ESC is that it’s a major pain in the ass. Even if you gear the car down by physically swapping gears in the differential, and get the car going slower, you ultimately don’t have the full range of control over it that you want. Every movie I’ve seen ends up with the robots being out of control and either killing everyone or taking over the world, except for Star Wars where the robots are pussies. To start the project with the robot car out of control seems like a bad idea. With that said, the first mockup will attempt to interface with the stock ESC and steering servo with a PWM driver.

## Conclusion

The next bit here is to actually assemble the parts, load some software on the car, and get some action going.

1. As nice as the MIT competition is, the Europeans have been holding a field robot competition for a number of years now. Slower than speed trials in the tunnels, but still interesting.

One thing that struck me about the entrants this year. The best performers all use LIDAR.

Perhaps near the end of 2016 or the beginning of 2017, the automotive MEMS LIDAR units will drive LIDAR costs down for price sensitive applications to gain some traction.

• Sure, there are several competitions out there that have been doing interesting work over the last 5 years at least. Sparkfun has their annual race, and there are several different international competitions. The interesting thing for this website in particular is the inclusion in the MIT RACECAR of the Jetson, that is, replacement of a small PC with an capable embedded processor. Also, the University of Pennsylvania has a very similar Jetson based, TRAXXAS Rally robot that they use in their classes. There was even talk this year of replacing the human drivers in Formula E with autonomous behavior, which certainly would have been technically interesting (though I don’t know why people would watch it after the novelty wore off).

LIDAR is a well proven performer in the field, that’s what the DARPA Challenge showed. In that camp is the Google self driving car. Unfortunately the LIDAR on the Google car costs about the same as the rest of the car combined. There is a competing camp, companies like Audi, Tesla and NVIDIA think that using multiple cameras along with radar is a viable approach. I believe that the thinking is that the advancement in computational power and algorithms will will outpace the current advantage that LIDAR enjoys.
People have been talking about getting the MEMS LIDAR chips out at an inexpensive price point (for less that \$1K in quantity 100,000) for a couple of years now, I think a lot of manufacturers are taking a wait and see approach at this point before placing their bets. This also does not rule out using multiple sensor types in the same vehicle, though the accountants aren’t going to be happy with more hardware.

2. Hello,
Thanks for sharing all this stuff !

You plan to power TK1/TX1 directly from a 3S LiPo ?
The spec says 5.5V – 19.6V so it probably work. But 12V is needed for PCIe. Do you know if the card step-up to 12v even if the lipo provide less voltage ? (A 3S Lipo provide 12,6 V fully charged and only 10V discharged)
I’m thinking of using a 4S LiPo with a step-down circuit like this http://www.dx.com/p/dc-dc-adjustable-step-down-heatsink-power-module-blue-5a-319570 , what do you think of it ?

Don’t forget to use a Lipo Alarm (http://www.dx.com/p/2-in-1-1-8s-lipo-battery-low-voltage-buzzer-alarm-for-rc-helicopter-white-black-180468) to avoid damaging your battery or even burn your car…

• It should be interesting to watch. There’s been so much hype about this area for several years now it will be interesting to watch the adoption rate when affordable sensors and computing power is available.

3. Always thank you useful information providing for all.
I’m very interested in this race car projcet. So I hope to know that how to use both the lidar lite v2 and imu module for this project. Because these sensons need to connect at i2c of jetson tk1 using 3.3v. But jetson tk1 have only one 3.3 i2c port. So could you please explain how to handle it if I want to use two sensors in this project.

4. Always thank you useful information providing for all.
I’m very interested in this race car projcet. So I hope to know that how to use both the lidar lite v2 and imu module for this project. Because these sensons need to connect at i2c of jetson tk1 using 3.3v. And also for servo and esc controlling, i2c 3.3v port is needed. But jetson tk1 have only one 3.3 i2c port. So could you please explain how to handle it if I want to use all module in this project?

• You can connect them both to the same port. They will have two different device addresses.

• Hi, Kangalow. Thank you very much. I understood your comment. I forgot what i2c is. 🙂 In my case, I will try to configure between slave devices and master of jetson tk1 3.3v i2c connecting.