CL2 Group

F1TENTH

F1TENTH is a 1/10th scale racing car that serves as a platform to research Head-to-Head Autonomous Racing. I got access to this vehicle by doing a URA under professor Yash Pant since September 2022.

I implemented the entire software stack onto this car (mapping, localization, raceline generation, pure pursuit, obstacle avoidance) using ROS2. During this process, I learned so much about autonomy and realized just how much effort it takes to simply getting Robotics to work.

The F1TENTH is worth $3000+ and equipped with:

• Nvidia Jetson NX (Compute Platform)
• Hokuyo 10LX (LiDAR)
• VESC 6 MkV (Motor Controller)
• Intel RealSense D345i (Depth Camera) (OPTIONAL)

Engineering Focus (DONE)

Got really good at implementing algorithms on ROS2, including:

• Perception: SLAM (Mapping), Particle Filter (Localization), Occupancy Grids from LiDAR scans
• Planning: Generate Optimal Racelines with Convex Optimization, RRT* for Obstacle Avoidance
• Control: Pure Pursuit and Stanley Controller (better) for line tracking

Competed at F1TENTH @ CPS-IoT Week 2023 for May 2023. Got fastest qualifying lap. Crashed out for race day due to Localization issues with changing track conditions.

Some things I still want to look into:

• Model Predictive Control
• Linear Quadratic Regulator

Research Focus (In Progress)

Check out Head-to-Head Autonomous Racing and F1TENTH Research Proposal. Very much still in the early phases of ideation. This stuff is hard…

Through the research, I’ve learned about concepts such as

• Multi-Agent RL
• Vehicle Dynamics
• Differential Game
• Markov Game
• Autocurricula
• Nash Equilibrium
• Proximal Policy Optimization (PPO)
• Deep Q-Learning (DQN)

External Links

• https://f1tenth.org/
• F1TENTH Foxy Documentation, pdf version here
• Bill of Materials
• F1TENTH Syllabus 2023 (Google Sheets, includes lecture videos)
• F1TENTH permanent syllabus
• F1TENTH Code by CL2-UWaterloo (GitHub)
• Autonomous Racing Literature

Concepts Learned from Lab Assignments

• Automatic Emergency Braking (AEB) $\to$ Lab2
• PID Control for Wall Following $\to$ Lab3
• Obstacle Avoidance $\to$ Lab4
• Scan Matching $\to$ Lab5
• Particle Filter $\to$ Use particle_filter
• SLAM $\to$ Lab6, use slam_toolbox
• Motion Planning using RRT $\to$ Lab7
• Perception $\to$ Lab8

F1TENTH Notes

• F1TENTH Field Usage
• F1TENTH Gym
• F1TENTH ROS
• F1TENTH Competition
• F1TENTH Blog

Concepts

• Rigid-Body Transformations
• Ackermann Steering
• Model Predictive Control
• Raceline Optimization using Optimal Control
• Teleoperation
• Head-to-Head Autonomous Racing
• F1TENTH Research Proposal

Links to Projects

• https://docs.google.com/document/d/1wCbG4vm1M94S3QlmazuQ56hELHVv8O91VZqQY8iDxaI/edit?usp=sharing

• https://github.com/MichaelBosello/f1tenth-RL

About the F1TENTH Car

One of the coolest features that was presented is this custom ESC, the VESC - Vedder ESC which has so many COOL features, including built-in IMU chips and Regenerative Braking.

Links

• Building the F1/10 car , with slides

It uses a Velineon 3500 Brushless Motor Motor (BLDC)

• RPM/volt: 3500, so we operate at 10-12V,
• Max RPM: 50,000 (this is ridiculously high) → It can go up to 30 miles per hour
• Wire size: 12-gauge

Full voltge: ~12.2V Low Voltage: Around 11V is where it is start to get really low.

Other random things I learned:

• Slipper Clutch
  • This is a safety feature

Engineering Lessons

• SLAM doesn’t work when Odometry is inverted
• Car doesn’t drive well / squeaking, then it is most likely a hardware issue. In my case, it was because the pinion and the spur gear didn’t make a solid connection