WarpDriver: Context-Aware Probabilistic Motion Prediction for Crowd Simulation

摘要

Microscopic crowd simulators rely on models of local interactionrn(e.g. collision avoidance) to synthesize the individual motion ofrneach virtual agent. The quality of the resulting motions heavilyrndepends on this component, which has significantly improved inrnthe past few years. Recent advances have been in particular duernto the introduction of a short-horizon motion prediction strategyrnthat enables anticipated motion adaptation during local interactionsrnamong agents. However, the simplicity of prediction techniques ofrnexisting models somewhat limits their domain of validity. In thisrnpaper, our key objective is to significantly improve the quality ofrnsimulations by expanding the applicable range of motion predictions.rnTo this end, we present a novel local interaction algorithmrnwith a new context-aware, probabilistic motion prediction model.rnBy context-aware, we mean that this approach allows crowd simulatorsrnto account for many factors, such as the influence of environmentrnlayouts or in-progress interactions among agents, andrnhas the ability to simultaneously maintain several possible alternaternscenarios for future motions and to cope with uncertainties on sensingrnand other agent’s motions. Technically, this model introducesrn“collision probability fields” between agents, efficiently computedrnthrough the cumulative application of Warp Operators on a sourcernIntrinsic Field. We demonstrate how this model significantly improvesrnthe quality of simulated motions in challenging scenarios,rnsuch as dense crowds and complex environments.