How the vehicles score

These are the vehicles’ results rather than mine. I am one engineer on a crash test team and a rating is the output of hundreds of people, but it is what the testing is measured against.

• Cybertruck. Five star overall from NHTSA, with five stars in side impact for both rows, and the lowest overall probability of injury and lowest rollover risk of any pickup NHTSA has tested. IIHS Top Safety Pick+, with a rating of Good in all three crashworthiness tests, for trucks built from April 2025 onward. Not rated by Euro NCAP, since it is not currently sold in Europe.
• Model 3. Five star overall from NHTSA with five stars in every individual category. Named Euro NCAP Best in Class Large Family Car for 2025, at five stars, scoring 90% for adult occupant protection and 93% for child occupant protection.
• Model Y. Named Euro NCAP Best in Class Small SUV for 2025. IIHS Top Safety Pick+ for the refreshed car. A 7.9% rollover risk in NHTSA testing, among the lowest ever recorded for an SUV.

The barriers

Each protocol specifies its own barrier, so a lab running several protocols needs several, all of them movable and repeatably aligned.

Euro NCAP full width deformable barrier, Spring 2026. The FWDB fixture for full width frontal testing, where the whole front of the vehicle strikes a deformable face at a fixed offset from the floor.

IIHS small overlap transportable barrier, 2023 internship. I led the mechanical design and fabrication of an 8,000 lb transportable barrier for small overlap IIHS crash testing at 40 kph.

Cybertruck safety

Tesla published a film on how Cybertruck is engineered for crash performance. I appear in it positioning a dummy before a test, and in the stills on Tesla’s safety page.

Engineering Cybertruck for safety
Tesla, Engineered for Safety

Phase I: Requirements

An adjustable cart that reproduces a vehicle’s photographic reference points, so photo technicians could frame and focus without occupying the crash test vehicle.

• Height, width, length, mid tire, sill, window, roof, dash, steering center, dummy shoulder
• Tolerance within ±1.06 in, and adherence to the internal guide
• In focus
• Easy to store, move around and assemble

Phase II: Prototyping

A cart with a coordinate system. I used CATIA to obtain all coordinates of interest from the vehicle models and build the cart around them, to a tolerance of plus or minus 1.06 in. The Cybertruck set the outer envelope for length and width.

Phase III: Procurement and build

All 8020 and compatible parts through McMaster: T-slotted framing, pivots, locking rail glides, linear bearings, corner brackets, and casters rated for 265 lbs. Total, $2,400.

Band saw to cut material down to the height, width and length-wise pieces. Tap M12 holes in the height pieces to fasten via bolts to the pivots. Drill press to create holes in the 8020 for fastening components.

Problem: pivot unstable

The height pieces did not stay in the 90 degree locked position, so the cart would not be a true representation of height.

The solution was a GD&T CAD drawing of two 0.25 in steel plates. The red plate bolts into the bearing. The blue plate is T-shaped on both ends to lock on both ends. The plates held the 8020 at 90 degrees when the pivot-locking force could not.

Problem: long rails bending

When extended to Cybertruck length, the 8020 beams bend due to gravity.

Bottom rail: add a caster in the center. Side rails: add additional bolts to the linear bearings and use corner brackets to attach to the 8020.

Phase IV: Final build

Lock width, height and length to the model’s color coded dimensions. Look for color coded markings for a specific point of interest, then slide the sliders to those points.

Phase V: Analysis

The cart removed a one to two hour setup bottleneck per test. It moved the photo timeline up by 1.5 hours, meant less overtime from technicians, and gave a 30% increase in productivity.

My takeaways were questions and communication, and CATIA assemblies.

Electrical design #1: Bluetooth

PS3 remote, Arduino Uno with code uploaded, USB host shield, Bluetooth dongle, motor shield. I learned Arduino without the hardware setup first, mapping joystick integers to PWM values and setting motor directions based on the plus and minus terminals.

The remote did not connect to the Arduino host shield. I tried with other PS remotes. The reason was that the host shield had been removed from Arduino’s library.

Electrical design #2: Wifi

PS3 remote and an ESP32, which is less heavy duty than a Raspberry Pi, with a motor driver. The Arduino code is compatible with the ESP32. I connected to the PS3 with a physical MAC address and wired the components from the ESP32 to the motor driver, which needed an additional power source. We got motion.

Two concerns. Only brushed DC motors are compatible, and we wanted brushless because they are more effective at handling torque and are lighter. And wifi operates at 2.4 GHz, so if another team uses wifi it could interfere.

Continuing this route would require more research into using the ESP32 with an ESC. There is limited documentation on this method, and we had time constraints.

Electrical design #3: radio frequency

A Flysky radio transmitter and receiver driving ESCs, with no Arduino. There is ample documentation on controlling three motors with the Flysky radio, including tank drive. It was equal in price to the Bluetooth and ESP32 route and less heavy. Ready for use one week after.

The ESCs must be compatible with the current rating on the motor, which is 30 A. The battery must supply three times the ESC current, so 90 A. Battery max current is 50 C multiplied by 2.2 Ah, which is 110 A.

Physical compatibility

I checked that the electrical designs were compatible with the mechanical throughout the process: four motors, the battery, the ESP32 motor driver, and wiring fitment.

Pulleys

The little pulleys originally required a motor shaft adapter to elongate the shaft. I made a single double pulley which could be printed as one piece, so no motor shaft adapter was required.

On the big pulleys I experimented with 36T, 40T, 44T and 48T. 40T was the largest size we could fit in the bot. I changed flange height to balance belt concerns against the size constraints of the bot. The original design included a mounting block, which we removed to conserve space, then press fit the pulleys and reinforced with two part epoxy.

3D printed body structure

We initially planned to use PLA for prototyping only. There were several iterations before the final print. The first iteration was much larger than the final product, and the bearing and motor mounts were breaking at 7% infill.

On the next one I raised the height of the supports on the motor mounts and centered the motor mounts between the bearing mounts for locomotion. We split the body in two because print time was too long, around 30 hours, then had to readjust the connection style on the halves and reprint. The weapon was originally planned in AR500 steel for hardness, and we used 6061 aluminum instead for machining.

Experimentation

Three alterable values: weight of the bot, max power to the weapon, and max power to the forward controls. The last two are programmable on the transmitter and affect the sensitivity of the controls.

A factorial experiment with three factors, two settings and eight trials, with two responses: the time it takes to traverse the arena and back, and ease of control on a one to ten scale. Main effects analysis showed the highest correlation between sensitivity and time. Drive sensitivity dominated traverse time, taking it from around 15 seconds to around 8, while ease of control went from 3 to 8.

The fight

Won one round. Lost one round. The failure mode was that one of the wheel hubs came off, so the bot lost motion.

Reflections

• Arduino time sink, so concurrently explore other connectivity
• Bulky wiring, so more physical check ins
• Difficulty handling, so explore using the ESP32 with the ESC instead of an RF receiver

What I took from it: Arduino applications, engineering decision making, team integration, analyzing design pitfalls, adhering to budgets and timing, and using resources effectively.

Chassis and calculations

A team member mocked up the chassis dimensions. My contribution was the static and dynamic force hand calculations. I used FEA in SolidWorks to model impact with a 50 kg wall, then used max deformation and conservation of energy to determine the impact force, which came out at 1377.1 N. The simulation also showed a safety factor of around 10 for the chassis.

We used quarter inch plywood because we wanted a material that is easy to manufacture. It is lightweight compared to metal or aluminum, cheap, and gave us flexibility to change the design if we needed to. We laser cut it at Texas InventionWorks.

Drivetrain

Two gears, a 10 tooth pinion and a 76 tooth spur, to reduce possible sources of error if there were more than two gears. I performed hand calculations to determine the gear ratio and the max velocity that would achieve a desired top speed of 5 m/s.

The top speed calculations were carried out twice. During the first iteration the speed came out at 11.02 m/s, which is significantly over what we wanted. The second iteration involved increasing the weight on the chassis by 2 kg to increase friction, so that the motor torque increases and its speed decreases. We did not want to change the gear ratio, since the diametral and teeth sizes were determined by our wheel sizes. The resulting efficiency comes out to 40%, and maximum acceleration from rest is 10.81 m/s².

The rear axle spins on bearing blocks, and I designed press fit hex inserts to attach the wheel to the axle.

Steering

It was originally calculated from the initial CAD mockup that the turning radius was larger than 32 in, which seemed like a problem. By making small changes to the long bar and steering columns, and adding cutouts to the wheel wells in the front half of the car, we decreased the minimum theoretical turn radius to 9.88 in. The wheelbase length from CAD was 6.175 in, which gave a maximum turning angle of 32 degrees.

The main hurdle for building was improving the steering system, which originally relied on a paperclip connection between the servo motor and steering rod. This did not provide a large enough angle of turn because of the paperclip’s flexibility. We fixed it by adding a more rigid setup including a paperclip and screw.

Test

Our car ended up winning the races for our section, but not the final race between sections. Our wheel popped off because permanent glue did not hold it well enough.

Going forward I would switch our gears to metal instead of 3D prints to avoid the wear from extended use, and create adapters to hold the outer rings of the bearings to the front wheels and the inner rings to the axle.

The final race

Characterize the manual process

The method of stretching, what is considered acceptable, and outer diameter data, to quantify the variation in outer diameter measurements of the part introduced by hand stretching.

Run studies

Set speed, temperature and weight ranges. I ran two design of experiment studies, three factors, two settings, eight parts each, measuring outer diameter and inspecting each part by hand.

Analyze the data

Excel and Minitab to gather trends and a final equation. I determined that speed of the chamber was the most important factor. I left for the school year, and another engineer carried out the next set of studies focused on speed.

What I learned: how to structure a DOE, new equipment interfaces in the heating tower and laser micrometer, and data driven decision making.