Technical Report #100023: Evaluating the Costs and Constraints Used to Make Footstep Decisions

The majority of steering algorithms model very few locomotion constraints or capabilities, and output only a force or velocity vector to motion synthesis. This simplistic interface between navigation and motion synthesis lacks control and introduces ambiguities. This paper considers how footsteps can be used to navigate in crowds of characters. We propose a novel locomotion model that represents a character’s stepping state and footstep actions using a 2D approximation of an inverted spherical pendulum model of bipedal locomotion. We use this model with a short-horizon planner to generate a timed sequence of footsteps that existing motion synthesis techniques can follow exactly. During planning, our locomotion model not only constrains characters to navigate with realistic locomotion, but more importantly, it also enables characters to control finer-grain subtle navigation behaviors that are possible with exact footsteps. Our approach can navigate crowds of hundreds of individual characters with collision-free, natural behaviors in real-time.