Deforming virtual reality, on purpose

It's a bit deceptive, but for the right reason. Restoring motor control after an accident can be a long, painstaking and often discouraging process. So for his Master’s project in computer science, Sidney Bovet studied a way to help patients stay motivated by amplifying their movements in virtual reality. With his system, the movements made by patients’ avatars are larger than the patients’ actual movements – which should help encourage patients to keep pressing ahead with their therapy. The challenge for Bovet was to make sure the avatars’ movements remain realistic and provide convincing haptic feedback.

EPFL researchers had already been studying movement, rehabilitation and motor control, but Bovet approached the subject from a whole new angle. He examined the way a user’s movements are transformed before being projected in a virtual reality headset. He looked specifically at the motion capture and tracking systems of virtual reality devices, as well as users’ perceptions of their own bodies.

A barely noticeable difference

Bovet’s research was overseen by Ronan Boulic, who heads EPFL’s Immersive Interaction Group. Bovet developed a distortion function and added it to the motion tracking software used for patients. He opted for a linear function to simplify the analysis. For instance, if a patient moved his arm ten centimeters, the avatar would move its arm by 20 centimeters, for a distortion factor of 2.

But that’s just the mathematical part. Bovet also wanted to know how patients perceived the distortion – and especially whether they felt comfortable with it. If the distortion was too obvious, it could become ineffective at keeping patients motivated. The researchers therefore tested different distortion factors on healthy volunteers and asked them if, for each factor, the movements their avatars made corresponded to the movements they actually made. “Our initial tests showed that as long as the distortion factor remained between 1.5 and 2.0, users rarely saw a difference. But beyond that, they quickly caught on,” says Bovet.

The more realistic, the greater the distortion

However, Bovet found that what mattered even more than the distortion factor was how accurate the contact reproduction was when patients touched their own bodies. “In virtual reality, if an avatar’s body shape is too different from the user’s own body shape, then the user will have a hard time identifying with the avatar,” explains Bovet. “For instance, if a user looks like Santa Claus but his avatar is slim, then when the user puts his hand on his belly, the avatar’s hand will be several centimeters in front of his stomach, not touching anything. So before we introduced the distortion, we had to make sure that self-contact by a patient would be correctly reproduced by his avatar. Because the more accurate that reproduction is, the more a patient will identify with his avatar, and the less likely it is that he will notice the distortion. That makes for a more immersive user experience.”

Further research is needed before the promising results of this project can be used in therapeutic applications. But that research will have to be done without Bovet, who has decided against pursuing a PhD. He will nevertheless stay on the EPFL campus, working on more mature applications at Logitech.