Blog · · 5 min read
AI industrial design under hard limits: four renderings of a solar-roof e-bike
How an AI design study turns hard rules into design: four deliberately different directions for a solar-roof e-bike in Singapore, renderings built from measurable 3D models, a compliance check for each, and a vote before CAD starts.
Ask an image generator for "a futuristic e-bike with a solar roof" and you get something beautiful in seconds. You also get something you cannot build: the roof floats, the frame changes from one view to the next, and nobody knows how wide or how heavy it is. Fine for a mood board, useless for a product that has to pass a regulation.
This article shows a different approach, using a public project on makemachine.ai as the example: Rooficle, a light e-bike with a solar and rain roof that has to be legal on Singapore's roads. Its design study produced four deliberately different directions, each as a set of renderings that can be measured, checked and compared before any mechanical CAD starts.
Start with the rules, not with the sketch
In Singapore, an e-bike is a power-assisted bicycle, and the Land Transport Authority sets clear limits. According to LTA's requirements, it must comply with EN 15194, the motor may deliver at most 250 W continuously, assistance has to fade out and stop at 25 km/h, the motor may only help while the rider pedals, and the unladen weight may not exceed 20 kg. The weight includes everything bolted or screwed onto the bike, but not accessories that can easily be removed (LTA). The regulations also limit the width to 700 mm.
A roof is the opposite of light and narrow. So before anyone draws a line, the design brief turns these rules into a table of hard limits with their sources, next to the wishes: shade and rain protection, solar charging, and a bike that still looks like a bicycle.
In the concept phase, a version with every add-on fixed to the frame (windscreen, rear box, phone console) came out at about 21 kg in the estimates, over the limit. Making the rear box quick-release brought one variant back under it. One rule about removable accessories changed the architecture of the product.
Four directions, deliberately different
"Different" is the hard part. Four color variants of one idea are not four directions. In this study, each direction has its own silhouette and its own idea of what a roofed bike is:
- A · Fastback: sporty, one long sweeping line from the top of the roof down to the rear wheel.
- B · Halo: a slim ring behind the rider carries the roof, and the sides stay open.
- C · Monoline: one slender arch forms the frame and carries a thin roof, like a single drawn line.
- D · Glider: a wide, flat, wing-shaped roof on one central mast, open from every side.
All four carry the same technical package: three flexible 30 W solar modules on a roof field of 1.07 × 0.57 m. That keeps the comparison fair. The choice is about form, not about which direction happened to get the bigger battery.
Renderings built from 3D models
The images are not generated pixels. Each direction is a 3D model written as a script and rendered with a physically based path tracer. Because the model exists, the main dimensions come from the brief in meters, and the renderings show what the model actually is.
Each direction gets the same set of images: a front three-quarter hero view and a rear three-quarter view in a Singapore-style street setting, plus a side silhouette so the shapes can be compared directly. Each direction also defines its materials and colors, what industrial designers call CMF: color, material, finish. Every image is labeled as a rendering of a design, never as a photo of a real product.
The script also makes the study repeatable. When a direction changes, the whole set is rendered again, and all views still show the same bike.
A compliance card for every design
A beautiful direction that breaks the rules is wasted time. So each direction gets a compliance card with values measured on the model: width, length, roof height, headroom, roof angle and windscreen, plus an estimated mass. All four passed the hard limits in the model.
The margins, however, differ a lot. The Halo comes out at an estimated 18.97 kg, about one kilogram under the limit. The Fastback has only about 0.8 kg left, and the Monoline about 0.5 kg. Those are estimates, not weighed parts, and the study says so. Still, a margin of half a kilogram tells you which direction will hurt when the first real screws and cables are added.
Physics shows up as well. A roof catches wind. The concept estimates showed that a parked bike with a roof tips over in gusts of roughly 7.5 to 9.8 m/s, against about 16 m/s without one. Every direction therefore needs a parking solution, which is the kind of detail a pretty picture hides.
A recommendation, then a vote
The study scores every direction on impact, everyday use and ergonomics, function, weight, manufacturing, cost and risk. Here, the Halo scored 42 of 50 points, ahead of the Glider (36), the Monoline (35) and the Fastback (33). The recommendation was the Halo, with the Fastback as the emotional alternative.
But taste is not a score. For a public project, the question goes to a vote on the project page: all renderings to swipe through, the comparison board first, then one button to vote. If nobody answers in time, the study takes its own recommendation and says so in the result.
The chosen direction then becomes a design brief for CAD: main dimensions with what is fixed and what CAD may adjust, key lines and surfaces, the split into parts, a CMF table and a 3D reference model.
What AI does well here, and what it does not
AI is good at the wide part of design: producing several genuinely different options quickly, checking every one of them against the same rules, keeping views and variants consistent, and writing down every assumption.
It is not good at the parts that need a body and a workshop. Mass is estimated until parts are weighed. Ergonomics needs a person on a real bike. Whether a roof still counts as "similar to a conventional bicycle" is a question for the authority, not for a model. And the final call on what looks right belongs to people, which is exactly why the vote exists.
To see the four directions and everything that led to them, open the Rooficle project. For the bigger picture, read what AI can and cannot do when you build a hardware product, or how the next step works in CAD as code.
Watch hardware get built by AI. Follow public projects on makemachine.ai, read every result in full and get an email when there is something new.
More from the blog
- Will my electronics enclosure overheat? A five-minute heat estimate for a sealed box
- Joining 3D printed parts: screws, heat-set inserts, snap fits or glue?
- Free or paid AI 3D modeling? What you really pay for, and a 30-minute test to compare tools
Open-hardware projects built with AI
- DIY Automatic Plant Watering System: An open-hardware watering system for indoor plants: a small pump, a refillable tank and a soil moisture sensor in each pot keep up to six…
- DIY BBQ Temperature Controller: A temperature controller for charcoal grills and smokers that makers can build themselves. A pit probe, two meat probes and a small fan on…
- ZählerSchau: ZählerSchau is a small, battery-powered unit that clips onto an electricity meter in Germany (water meters via a clamp adapter), takes a…
- Roadstoy: Roadstoy is a remote-controlled toy version of the new Tesla Roadster that can briefly jump and flip using compressed air. It drives on an…
- The forever Washing machine: The forever washing machine is a washing machine offered as a service instead of a product. Customers do not buy the machine but pay for…
- Bike Light: I want to build a bike light that needs no battery. Magnets sit in the spokes close to the hub, and a coil on the fork picks up their…
- Rooficle: I want to develop a light, efficient and affordable e-bike that can be used legally on public roads in Singapore. Its core is an…
- BBQ Modular Grill: I want to design a modular grill. Instead of one sealed block, it is made of separate parts that can each be bought on their own online…
- Go FANUC Go: A mobile base that moves an existing FANUC M-710iC industrial robot (about 560 kg, with a milling spindle at the wrist) out of a room and…
- Jump & Run Robot: I want to build a small, low-cost toy humanoid robot that people can put together themselves. It should be about 30 cm tall, or as small…