LONDON: Scientists are developing sophisticated virtual stunt artists that could eventually put genuine stuntmen out of a job.
Thanks to an array of sensors the new computer-generated two- and three-dimensional human models respond to the physics of the real world unlike previous versions which had to be laboriously animated frame for frame.
Petros Faloutsos based at the University of California in Los Angeles said computer stuntmen could eventually replace the real thing since they allow directors to plan daring acrobatic feats without risking anyone’s lives, New Scientist magazine reports from London.
The results of his PhD thesis on everyday motions and stunts can be seen on a website which includes an intriguing sequence showing a robot-like figure sitting down on a toilet before arising again and being hit on the head by a red ball.
The virtual stuntman is a jointed skeleton figure that responds to the forces produced by gravity, friction and impact with other objects in its virtual environment.
“A benefit of this is that you don’t necessarily know what’s going to happen to a character until they’ve been hit,” said Faloutsos. The skeleton can be dressed up to resemble any film character.
A set of programmes called controllers governs the movement of these 3D figures, be it running or jumping over a wall. Combining the movements to create realism is the problem and that is where Faloutsos and his colleagues Michiel van de Panne and Demetri Terzopoulus at the University of Tornot come in. They make the individual controllers work in concert. The sensors keep track of the figure’s sense of gravity, its joint movements and points of contact between it and the environment.
For instance when the character has lost its balance, dive and fall behaviours take over from the running controller. Faloutsos developed the idea in conjunction with the Vancouver-based computer animation company Motion Playground while he was at the University of Toronto and the system can be seen in action on the following website: www.toronto.edu/:pfal. – dpa