DNEG Transform crowd assets

With the goal of improving the iteration time of the creation of complex crowd shots, we set about the task of trying to provide more art directable tooling for our crowd artists. We wanted to be able to give our artists the tooling to allow them to produce the same level of complexity that traditional simulation techniques allow, whilst also not requiring a complete re-simulation for minor edits.

Taking inspiration from a technique called Synchronized Multi-Character Motion Editing (Kim et. al) we combined a laplacian path editor with a motion graph implementation. From here, we were able to allow users to define constraints, and solve for appropriate motion to satisfy those constraints. The video below shows an example where three spatio-temporal constraints (a position at a given time) were placed, and appropriate walking and turning animation are produced.

By maintaining the desired frame, but extending the distance between constraints, we can trigger different animations. In this example, we trigger a transition into a running clip simply by moving the third constraint away from the second one.

To ensure that artists are able to trigger certain animations at specific times, we provide the ability to label sections of animation before the motion graph is constructed. From there, artists can specify a label constraint which ensures that a piece of labelled animation will be solved for at specific times.

The framework developed can be extended to solve for other common use cases for crowd tasks such as collision avoidance. Normally this would be done in simulation by applying forces to agents which triggers a change in action. By using an iterative version of the laplacian solver to edit the trajectories, we can run a collision avoidance pass on a set of agents that are colliding. A slowed down, iteration by iteration example of an extreme case for collision avoidance is shown in the video below.

Whilst there are still a lot of kinks to iron out and refinement of tooling before this becomes a fully fledged replacement for simulation, it shows enormous promise and we’re keen to test its boundaries.

Find out more about DNEG and The Asset Pipeline