A self-driving car has no steering wheel, so the whole cabin has to change. We invented a surface that reshapes itself to fit.

Autonomous vehicles remove the steering wheel and pedals. The cabin becomes open space, a place to work, rest, or socialize. At highway braking speeds, anything unsecured on a surface becomes a projectile.

BMW sponsored a nine-month collaboration between Stanford and TU Delft to prototype interior solutions for this problem. Seven people, four from TU Delft and three from Stanford, formed the team. It was the first project of its kind between the two institutions.

The project ran as Stanford's ME310. I led UX research, prototyping, and communications. Three pieces of work were mine end to end. The no-seatbelt restraint concept, from the build through testing to the decision on its future. An interactive head-tracking simulation for the A-pillar study, built solo in p5.js. And the kirigami experiments for the surface cover.

The final particle jamming prototype was a build split six ways. Isshita and Shu moulded the cells. Souparna handled cell assembly and the electronics. Isshita made the silicone top fabric. I worked on the silicone casting, and owned the external casing, the fabric stitching, and the assembly alongside Lotje. Willem and Alba produced the presentations and digital assets.

Coaching came from George Toye, Mark Cutkosky, Andrew J. Milne, Arjen Jansen, Dave Murray-Rust, Stefan Heijboer, and Brian Mok across the two schools.

The team generated six concepts, a see-through windshield, a cabin that reconfigures itself, a car that senses the driver's mood, a transparent mode for socializing, an interior that generates its own energy, and a car with no seatbelts.

We split into two teams. Each selected concepts to prototype within five days.

I owned this concept end to end, from the build through testing to the call on whether it went forward. The restraint was built from a real BMW seat, packaging material, and thread. We tested it across scenarios, sitting still, gentle braking, hard braking, sitting sideways, sleeping, and a full crash test with a dummy.

The seat-tilt mechanism had to activate at the moment of braking or earlier. Testing established the constraints precisely, covering sensor accuracy, refresh rate, and the margin between hard braking and a collision. At production scale, those margins set the ceiling on the concept.

I made the call to retire it once the test data was in. It delivered real freedom of movement, and the sensing it depended on was not going to clear production safety standards. Reading that early moved the whole team onto the surface concept, which became the final prototype.

I built an interactive head-tracking simulation on my own in p5.js with ml5.js FaceMesh, so people could feel what a see-through A-pillar would be like. It tracked keypoint 6, between the eyes, and rendered a three-layer parallax, with a background driving video at 1/15 speed and a foreground BMW interior image at 1/3 speed. Two additional tests rounded out the study, a taped-over racing game as a basic vision test and a driving simulation with a blocked sightline.

Three participants tested driving with and without the A-pillar visible. The view without pillars was rated as providing more freedom and more control, but drivers felt less safe. Structural elements need to remain visible to preserve a sense of enclosure.

A patent search turned up a Volvo patent that already covered see-through pillars, so we dropped the idea the same day.

The team built a life-scale particle jamming chair from a wooden frame and vacuum cleaner pumps. Users responded to the shape-conforming property, and the test established where the technology belongs, since jamming holds one position by design, which suits objects rather than a body that needs to shift and be supported. The particle volume a full seat would need pointed the same way. A visit to a BMW dealership in Mountain View identified the right component, the dashboard, a large surface of roughly 500mm × 300mm with simple geometry and an open opportunity for securing objects.

Three particle materials were tested. Coffee grounds gave the strongest jamming, with hardness and weight as the trade-off. Bird feed was softer and gave up shape retention. Polystyrene beads performed best on jamming at low weight, and sat outside BMW's sustainability constraint. Coffee grounds were selected for the final prototype, sustainable and fine-grained enough for good object conformity.

The first cell design used heat-sealed plastic bags with external bladders, and it set the requirements for the second, no gaps between cells and a stronger wall material. The second design placed the bladder inside the cell at 30% of total volume, with particles filling the rest. Cell walls switched from plastic to cast Ecoflex 00-30 silicone rubber, selected over BBDINO 30A for being softer, grippier, and more conforming. The final prototype was a 2×4 matrix of eight cells. Air removal took ~2.14 seconds. Air restoration took ~0.64 seconds. Vacuum pressure ran at −85 kPa.

The cover I stitched over the matrix was a silicone fabric composite, a knitted 82% polyester and 18% spandex blend, chosen for high friction, elasticity in every direction, and adhesion to the silicone beneath it. Air handling ran on two separate blowers, one for the particle cells and one for the bladders, feeding a vacuum chamber and pump.

The team specified how a person would operate the surface, with a two-second press to lock an object and a double tap to release it. The cell moves through three states in response, named default, morphable, and holding.

The working prototype grips objects by shaping around them and locking in place. It meets the design requirements, holding objects 3cm deep, delivering grip force of 5N, and returning to shape within one minute.

I led communications for the project and presented to more than 150 people at Stanford EXPE 2024, a formal exhibition and design expo. The audience included BMW's team and investors. One investor noted the technology has applications in spacecraft. The project was the first collaboration of its kind between TU Delft and Stanford.

After EXPE, the work was handed to BMW's central R&D team in Germany, where the technology is being developed further.