About
I help engineers — whether still in school or already working in industry — who are stuck at tutorials and priced out of proper hardware, to building and deploying a high-speed autonomous racing robot from scratch — running full stereo visual-inertial odometry, 3D SLAM, and racing algorithms including behavioral cloning — using production-level ROS 2 on a budget anyone can afford, so they build a portfolio project that proves they can engineer real autonomous systems.
The Transformation
| Where they start | Where they end up |
|---|---|
| Knows ROS 2 basics but has never built a production pipeline | Writes real-time, hardware-aware ROS 2 nodes with proper TF trees, REP-105 conventions, and production launch files |
| Has done SLAM tutorials but never tuned a real VIO stack | Runs a full stereo visual-inertial odometry pipeline — stereo cameras + IMU fusion — with calibrated intrinsics, extrinsics, and time offsets using Kalibr |
| Can't afford F1TENTH (~$4,000) or university lab access | Builds a high-speed 1/12-scale autonomous racer with the same perception-mapping-control pipeline at a fraction of the cost |
| Doesn't understand sensor fusion — just calls a library | Understands IMU preintegration, timestamp correction, EKF state estimation, and why USB jitter kills your odometry |
| Has never deployed a racing algorithm on real hardware | Deploys Gap Following, Pure Pursuit, MPC, and behavioral cloning-based racing on a real platform under real-time constraints |
| Has only seen behavioral cloning as a demo concept | Collects expert driving demonstrations and trains an end-to-end imitation learning policy that runs on-robot at inference time |
Before
Knows ROS 2 basics but has never built a production pipeline
After
Writes real-time, hardware-aware ROS 2 nodes with proper TF trees, REP-105 conventions, and production launch files
Before
Has done SLAM tutorials but never tuned a real VIO stack
After
Runs a full stereo visual-inertial odometry pipeline — stereo cameras + IMU fusion — with calibrated intrinsics, extrinsics, and time offsets using Kalibr
Before
Can't afford F1TENTH (~$4,000) or university lab access
After
Builds a high-speed 1/12-scale autonomous racer with the same perception-mapping-control pipeline at a fraction of the cost
Before
Doesn't understand sensor fusion — just calls a library
After
Understands IMU preintegration, timestamp correction, EKF state estimation, and why USB jitter kills your odometry
Before
Has never deployed a racing algorithm on real hardware
After
Deploys Gap Following, Pure Pursuit, MPC, and behavioral cloning-based racing on a real platform under real-time constraints
Before
Has only seen behavioral cloning as a demo concept
After
Collects expert driving demonstrations and trains an end-to-end imitation learning policy that runs on-robot at inference time
Education
Current Work
Designing and analyzing communication systems with a focus on signal integrity, channel modeling, and system performance. The signal processing background from Johns Hopkins directly informs the sensor fusion and EKF work in the robotics platform.