AR Race Car

By 2018 we had built augmented reality prototypes with most of the major toolkits, including ARKit, ARCore and Vuforia. Each one taught us something about tracking, surfaces and how people react when digital objects appear in their space. None of them had shipped. A prototype can lean on a developer who knows its quirks. A product has to work for someone who has never seen it.

We wanted one complete AR app in the App Store: polished enough to hand to a client in a meeting, and complex enough to make us solve the problems that only show up once real people use it. We chose cars. They need no character animation, they make sense at a glance, and everyone wants to drive one.

Scale was the main technical risk. People would want a toy car on the desk and a full-size car in the driveway, but resizing a physics-driven car changes how heavy it feels, how fast it seems and how it handles. ARKit was also less than a year old, so there were few established patterns to follow.

Placing a car in the real world

AR Race Car is built in Unity, with Apple's ARKit for world tracking and surface detection. When the app starts, it looks for a tabletop, floor or road and shows a scanning prompt until it finds one. The player then places the car where they're pointing. It always faces away from the camera, ready to drive, stays level, and stays on the surface even if the phone drifts toward a wall. A short particle effect marks the landing, and soft shadows ground the car on the real surface.

Scaling the world, not the car

Instead of resizing the car, we scale the AR camera rig around the spot where the car was placed. The car keeps its true physical size inside the simulation, so its weight, grip and speed never change. What changes is how big the real world looks to it. A single slider takes the car from Tiny to full size across a 66× range, and it handles the same at every size, whether it's on a desk, the living room floor or a parking lot.

Driving feel, sound and smoke

Every car runs on a full vehicle physics simulation with suspension, tire grip and weight transfer, in three setups: Street, Race and Drift. Stat bars for acceleration, top speed, handling and traction show how the cars differ before you choose. Engine sound follows the throttle, tires squeal when they lose grip, and hard driving lays skid marks and tire smoke on the real surface. We licensed high-quality vintage race car and sports car models and gave each one new paint and liveries, which kept our effort on placement, scaling, controls and feel.

Everything a prototype skips

Shipping meant finishing the parts a prototype never needs: animated onboarding that teaches scanning, placing, scaling and driving, a pause menu, haptic feedback, a swipe-to-browse garage, iPhone X safe-area layout and a full test cycle. We enlarged the on-screen controls several times after testing so they work for thumbs on a phone held up to frame the scene. We also built in recording with ReplayKit from the start, because AR is something people want to show off.

AR Race Car launched as a free App Store app about five weeks after we started, built by one developer, and found its audience without a marketing budget. It also became our best AR sales tool. In client meetings we set a car on the table, hand over the phone, and within a minute the client is scaling it up and doing donuts in front of everyone.

• Downloads: more than 41,000
• App Store rating: 4.4 out of 5
• Time to first release: about 5 weeks (April to May 2018)
• Scale range: 66×, with the same handling at every size
• Team: 1 developer, with updates for Unity and Apple platform changes through 2020

The problems we solved here apply to almost any AR brief: getting a digital object to sit convincingly on a real surface, letting people resize it without breaking it, and designing controls for a phone held at arm's length. That experience now goes into our AR and VR development for brands and agencies. You can see more of the cars on the AR Race Car page.

Unity, Apple ARKit, ReplayKit, iOS haptics, vehicle physics simulation, DOTween, TextMesh Pro