Steering, then
the whole car.
It started with the one system the driver touches every second of every lap — a steering rebuild 68% lighter and sharper in the hands. Then the question grew: if the linkage is honest, is the whole vehicle? That pulled the work from a single mechanism out to a full kinematic and compliance model of the car.
The Brief
Steering is the one system the driver touches every second of every lap. It has to be light, stiff, and honest — and the car it came on was none of those.
The previous design used a steel column and a chain of universal joints to route the wheel around the driver's legs. It worked, but it was heavy, it flexed under load, and the U-joints made the steering feel slightly different depending on where in the rotation you were. The goal was a system that disappeared — that let the driver feel the front tires directly, with as little mass and compliance in between as possible. Solving that mechanism honestly is what later opened the door to modeling the whole vehicle.
Steering geometry
Four decisions, each a weighted trade. Together they took most of the mass out and put the feel back in.
Steel column → carbon-fiber tube
The old column was a steel tube — heavy, and adding rotating inertia right where the driver feels it most. A multi-layer CFRP layup cut the mass dramatically while keeping torsional stiffness high, so feel got sharper, not vaguer. This single change drove most of the 68%.
Double-U-joint → printed bevel gearboxes
Routing the column around the driver's legs meant a chain of universal joints — compliance and non-constant velocity through the motion. I replaced them with compact metal SLS-printed bevel gearboxes: fewer parts, tighter packaging in the nose cone, a cleaner transmission of input.
KAZ rack → NARRco rack
Rack selection came down to a weighted decision matrix across mass, tolerance, ratio, and packaging. The lighter, tighter-tolerance NARRco rack won — less backlash at the wheel and a better fit within the bulkhead constraints.
Slight anti-Ackermann, on purpose
On a high-grip autocross car running real slip angles, perfect Ackermann isn't optimal — the heavily loaded outer tire wants a larger angle. The steering-arm geometry was tuned for a modest anti-Ackermann, validated against bump-steer through full suspension travel so toe stays honest over bumps.
↑ The steering field up top is that geometry, solved live — the same anti-Ackermann split, drawn from the linkage solver rather than described.
From a linkage to a vehicle
A linkage that behaves is only honest if the car around it does too. So the model grew outward — from four ball joints to the whole vehicle's balance.
I built a kinematic model of the full vehicle to optimize performance and stability — sweeping tie-rod and upright geometry to minimize bump steer, and balancing understeer / oversteer with the Milliken Moment Method.
The anti-Ackermann target stopped being a static number. With compliance modeled in, I tuned slip-angle behavior directly — then closed the loop against Python simulation and track feedback until the model and the car agreed.
A closed feedback loop: Python sim ⇄ track data
Each geometry change was predicted in the Python model, validated on the assembled chassis for Ackermann and bump-steer, then track-tested for effort, feel, and transient handling. Disagreements fed straight back into the compliance terms — so the model earned its trust instead of assuming it.
Compliance & FEM
Stiffness isn't infinite, and pretending it is hides the truth. The current work models exactly where the car gives.
I'm building a compliance model to simulate the stress of vehicle parts under load — extending the FEA already run on housings, shafts, and the CFRP column for stiffness and safety factors out to a whole-vehicle picture.
The next layer is collision detection through FEM — so the model doesn't just predict where parts deflect, but where they'd interfere or fail through the full envelope of motion and loading.
Bump-steer hides in the toe
The geometry that's perfect at ride height can lie through suspension travel: as the wheel moves up and down, the tie-rod arc drags the toe with it. A car that feels planted on a smooth skidpad can dart over a bump. The fix isn't a single number — it's co-tuning the tie-rod and upright geometry so toe stays honest across the full travel, then proving it in the model before it ever reaches the track.
What the rebuild bought
The carbon column and printed gearboxes took the mass and the slop out at once, the NARRco rack killed backlash at the wheel, and a deliberate anti-Ackermann split made the loaded outer tire happy. From there the work scaled up — a full-vehicle kinematic model and a compliance / FEM pass now in flight.
Want to talk vehicle dynamics?
Happy to walk the geometry, the FEA, or the modeling loop — from the linkage out to the whole car.