Heter-Sim: Heterogeneous Multi-Agent Systems Simulation by Interactive Data-Driven Optimization.

Area of Science:

• Computer Science
• Artificial Intelligence
• Virtual Reality

Background:

• Current multi-agent simulation algorithms require extensive parameter tuning for realistic behaviors.
• Simulating heterogeneous agents (e.g., crowds, traffic, vehicles) with varying dynamics presents significant challenges.

Purpose of the Study:

• To introduce Heter-Sim, a novel approach for simulating heterogeneous agents in virtual reality.
• To achieve plausible and interactive agent behaviors with reduced parameter tweaking.

Main Methods:

• Combined physics-based simulation with data-driven techniques using an optimization-based formulation.
• Estimated motion states from real-world datasets (position, velocity, control direction).
• Incorporated constraints like velocity continuity, collision avoidance, attraction, and direction control via a novel energy function.
• Reduced computational search space for collision avoidance and solution computation.

Main Results:

• Heter-Sim simulates tens to hundreds of agents at interactive rates.
• Accuracy validated against real-world datasets and prior algorithms.
• User studies confirmed the plausibility of generated behaviors in virtual reality.

Conclusions:

• Heter-Sim offers a general and efficient method for simulating heterogeneous agents.
• The approach significantly enhances the realism and interactivity of virtual reality simulations.
• Reduced parameter tuning makes the simulation method more accessible and practical.