Introduction
Content Coming! Eventually
We are a group of 3 students at UCSD looking to pursue an education in Data Science, and this is our journey at designing an autonomous vehicle
Building
To the left is the initial step we took towards builing
We had to take apart the chasis of an RC car set, replace it with a more powerful motor, and prepare it for adding new equipement
On the right is a custom PCB that was made for this project.
Developed by other ECE students, it provides the right power for the different equipment we will be using. Some of which include:
We had to put all the
equipment together on a custom cut acrylic board so it could fit on the RC car chasis. Here
you can see Jetson SBC on top in a custom case, the DC-DC converter below
it, and the lidar to the left, with the camera system mounted above it
After it was all mounted,
this is what the body looked like. By adding some standoffs in holes where things were previously
on the chasis, it was a clean and sturdy fit
Electrical- This is the underview of the
chaisis. Here are testing the voltage of the different ports of the PCB to ensure we don't blow up the
at least $2k worth of equipment that will be using it. Once it was tested the battery barely pictured
in the bottom left is popped in and all components are plugged in to the PCB and the top is folded
ever and the car is complete (pictured above)
Shocks- After that we changed
the shocks. After adding all that we did, we realized the car sunk way to low and would likely hit the
floor with almost any gain in speed. These shocks are absolutely ideal. They have adjustable springs so
we can change the tension as we see fit (really important if we add and remove features). They are also
just stronger at max value than the other ones we had.
(Unfortunately we had to change them again to a
another different kind but the alternatives were basically just as good)
Simulation
To get this car driving on it's own, we had to train an AI model. Using a driving simulator and hours of in it, we were able training data for the model to read. As you can see in the video below, the virtual track actually looks very similar to the real world track
We uploaded these models to the car and it was able to detect the same lines. Results were streamed on twitch
Tuning - Distracted Robot
A this point, the car worked, but had a lot of tweaking to do Our robot was programmed to follow the orange line, but as a side effect, it got very easily distracted by anything else orange , like the cone it crashed into.
This is a perspective of what we could tweak on the fly, and how we can change what the car can see. By adjusting filters and ranges for the camera's POV, we tuned how well the car could percieve what we wanted it to.
Final Demo
This was pretty much it, watch the vid and see how he runs :) I'm so proud of him