WarpDriver: Context-Aware Probabilistic Motion Prediction for Crowd Simulation

摘要

Microscopic crowd simulators rely on models of local interaction(e.g. collision avoidance) to synthesize the individual motion ofeach virtual agent. The quality of the resulting motions heavilydepends on this component, which has significantly improved inthe past few years. Recent advances have been in particular dueto the introduction of a short-horizon motion prediction strategythat enables anticipated motion adaptation during local interactionsamong agents. However, the simplicity of prediction techniques ofexisting models somewhat limits their domain of validity. In thispaper, our key objective is to significantly improve the quality ofsimulations by expanding the applicable range of motion predictions.To this end, we present a novel local interaction algorithmwith a new context-aware, probabilistic motion prediction model.By context-aware, we mean that this approach allows crowd simulatorsto account for many factors, such as the influence of environmentlayouts or in-progress interactions among agents, andhas the ability to simultaneously maintain several possible alternatescenarios for future motions and to cope with uncertainties on sensingand other agent’s motions. Technically, this model introduces“collision probability fields” between agents, efficiently computedthrough the cumulative application of Warp Operators on a sourceIntrinsic Field. We demonstrate how this model significantly improvesthe quality of simulated motions in challenging scenarios,such as dense crowds and complex environments.