Dynamically Simulated Characters in Virtual Environments
Dynamically Simulated Characters in Virtual Environments home page
Movies from our 1997 Siggraph Animation Sketch
More about our current virtual environment hardware setup
Slide 20 of 21
Listed in orange are two contributions we've made with this research. In our environments we have autonomous interaction between multiple dynamically simulated characters. However, it is still difficult to create the high-level behaviors for these agents. In the case of the bicyclists we completed a first pass at automating some of the high-level control by automatically generating a path that cuts through the race course while avoiding the edges of the road. A realistic racing agent would attempt to improve on this safe center of the road path so as to improve his race time. It is our goal to continue in the direction of automatically creating high-level behaviors for the agents.
Also, we would like to add some additional behaviors to the capabilities of our racing agents. By adding a simple energy function to each racer, we will provide a limitation on the energy expended by a racer. Each racer will have to optimize their performance given their energy function. Furthermore, we would like to add the physical benefits obtained by drafting (or riding closely behind another rider). Drafting will provide an incentive for groups of bicyclists to ride together, thus providing future research in the area of cooperative multiagent robotics.
The motion displayed by the characters in our environments is physically generated and results in motion that looks natural. However the physical capabilities also make creating virtual environments with dynamic simulations difficult. The physical capabilities of an agent constrain the agent's motion and complicate the high-level control. Furthermore, the computational requirements of dynamic simulations prevent arbitrarily large groups of agents from being placed in an environment.
Slide 20 of 21