Researchers at MIT have introduced a new technique for artists that could revolutionise the way animated characters are brought to life in movies and games. The technique is based on barycentric coordinates, a mathematical function that defines how 2D and 3D shapes can move and bend. This is a significant advance on existing techniques which limit artists to a single set of barycentric coordinate functions for each character, with any changes necessitating a completely new approach.
The new technique offers animation artists far greater flexibility, allowing them to manipulate motions and appearances to achieve their intended look. In addition to its applications in the arts, the research could also benefit medical imaging, architecture, virtual reality and even help robots better understand how objects move in the real world.
The technique was developed through a unique approach, using a special type of neural network to model the barycentric coordinate functions. This network architecture can output barycentric coordinate functions that meet all the required constraints, allowing artists to focus on their creativity without needing to worry about the underlying mathematical challenges.
The researchers drew on principles dating back to the 19th century, using the triangular barycentric coordinates conceptualised by German mathematician August Möbius. Though these principles work well with simple shapes, modern cages used in animation are far more complex. To overcome this, the researchers overlaid the shape with virtual triangles that connected points on the cage’s exterior. The neural network was then used to predict how to combine these triangles to create a complex but smooth function.
Using this approach, artists can experiment with different functions, tweaking the movement until they achieve an animation that matches their vision. The technique was successfully demonstrated, showing the potential for more natural-looking animations that won’t appear rigid or inflexible. Future plans include improving the speed of the neural network and developing an interactive interface that allows artists to iterate on animations in real time.
Unlike traditional methods, this new technique honours what art is truly about: the artist’s creativity and vision. Rather than hindering them with a rigid and inflexible mathematical framework, it gives them the freedom to experiment and bring their creations to life in a way that no previous technique has allowed. In turn, this could lead to a revolution in the way animated works are produced, with a noticeable enhancement in the quality and realism of animated motion. The research was funded by several organisations, including the U.S Army Research Office, the CSAIL Systems that Learn Program, and the MIT-IBM Watson AI Lab.
