A decision support framework on simulation fidelity for transferable and autonomously optimised swarm behaviour
研究了群体机器人仿真中轻质点模型与真实差速驱动模型在优化速度和参数可迁移性之间的权衡，发现低保真模型虽快但迁移性差，并提供了基于任务复杂度和性能目标的仿真选择决策工具。
This paper presents a decision support framework for using autonomous tuning algorithms in swarm robotics applications. With the rapid emergence of swarm robotics, there is a growing need to automate parameter tuning for efficient transfer to diverse robotic platforms. Simulation-optimisation frameworks address this, but a key tradeoff exists between tuning speed and parameter transferability. This tradeoff is investigated using the Frontier-Led Swarming Simulation Optimisation Framework (FLS-SOF), and two simulation models are compared: a lightweight point-mass model and a realistic differential-drive model. Experiments span three tasks–coverage, navigation, and cooperative maze solving–executed across ground and aerial robot platforms. Results show that while the point-mass simulator enables a 5.4× speedup in optimisation and may be suitable for early-stage exploration or time-constrained development cycles, its tuned parameters exhibit limited transferability to complex tasks, incurring up to 98.7% more collisions. In contrast, the differential-drive simulator improves navigation success by 196.7% and maze exit rates by 125%, while maintaining safety and coordination. These insights are synthesised into a flowchart-based decision tool that guides simulator selection based on task complexity, platform dynamics, and performance objectives. This tool enables practitioners to make fidelity-aware choices for optimising transferable swarm behaviours in real-world robotics deployments.