Energy-efficient mid-term strategies for collision avoidance in crowd simulation

Abstract

Figure 1: In some situations, humans can see some kind of tunnel inside a flow that can facilitate the crossing of the flow. The left figure shows a first person view point where the tunnel is clearly visible. The right figure shows a top view of the situation as well as the space of colliding and collision free velocities. Regular local avoidance models try to find a collision free speed toward the goal. This means the red agent will choose a velocity from the green part of the space of velocities, making him slow down and miss the tunnel. AbstractWhen navigating in crowds, humans are able to move efficiently between people. They look ahead to know which path would reduce the complexity of their interactions with others. Current navigation systems for virtual agents consider the long-term planning to find a path in the static environment and the short term reaction to avoid collision with close obstacles. Recently some mid-term considerations have been added to avoid high density areas. However, there is no mid-term planning among static and dynamic obstacles that would enable the agent to look ahead and avoid difficult paths or find easy ones as human do. In this paper we present a system for such mid-term planning. This system is added to the navigation process between the path finding and the local avoidance to improve the navigation of virtual agents. We show the capacities of such system on several case studies. Finally we use an energy criterion to compare trajectories computed with and without the mid-term planning.

ProactiveCrowd: Modelling Proactive Steering Behaviours for Agent‐Based Crowd Simulation

“…Golas et al [GNCL14] proposed a long-range collision avoidance model which extends the local RVO model by exploring future states of agents with lookahead times. Bruneau and Pettré [BP15] introduced a mid-term planning model which also explores future agent interactions within the next n time steps and devises optimal avoidance strategy based on an energy cost function. Best et al [BNCM14,NBCM15] proposed a DenseSense (DS) model which uses a density-dependent filter to correct the preferred velocity from global planner.…”

“…Research on proactive steering behaviour modelling has emerged as a promising direction in crowd simulation. To realize proactive steering behaviours, existing work [PPD07, GNCL14, BP15] usually adopts a prediction‐based approach, which infers proactive steering motions based on the prediction of future states of agents. Such an approach has to estimate future interactions among agents which could be complex when the number of agents increases.…”

“…It may also require some cost functions (e.g. energy consumption function in [BP15]) to determine the optimal velocity of agents. However, in real‐life situations, a person may not adopt such optimization techniques (e.g.…”

Position-based multi-agent dynamics for real-time crowd simulation

“…Guy et al [2010] simulated large-scale crowds by optimization based on the Principle of Least E ort. Bruneau and Pettré [2015] presented a mid-term planning system to ll in the gap between long-term planning and short-term collision avoidance.…”

Modeling Gap Seeking Behaviors for Agent-based Crowd Simulation

“…More recently, Reciprocal Velocity Obstacles (RVO) model [21] was introduced to generate collision-free locomotion of agent based on velocity space sampling. Mid-term planning is a new kind of steering method proposed by Bruneau and Pettré [3]. The method models the navigation process between global navigation and local steering.…”