RANT: Ant-Inspired Multi-Robot Rainforest Exploration Using Particle Filter Localisation and Virtual Pheromone Coordination

TL;DR

This work tackles reliable multi-robot exploration and hotspot mapping in noisy, partial-observability settings. It introduces RANT, an ant-inspired framework that fuses particle-filter localisation, a gradient-guided hotspot exploitation controller, and virtual pheromone blocking implemented in Webots. Key contributions include a complete Webots-based pipeline with shared maps ($R_g$, $P$, $V$, $B$, $C$) and an extensive evaluation of team size, localisation fidelity, and coordination, demonstrating the necessity of PF localisation (e.g., $N=150$ particles) and the positive impact of blocking on coverage quality. The results show that larger teams speed hotspot discovery but encounter diminishing returns due to interference; localisation fidelity is structurally required for accurate mapping, and pheromone-based coordination improves coverage uniformity by reducing redundancy, with implications for field deployments in environmental monitoring.

Abstract

This paper presents RANT, an ant-inspired multi-robot exploration framework for noisy, uncertain environments. A team of differential-drive robots navigates a 10 x 10 m terrain, collects noisy probe measurements of a hidden richness field, and builds local probabilistic maps while the supervisor maintains a global evaluation. RANT combines particle-filter localisation, a behaviour-based controller with gradient-driven hotspot exploitation, and a lightweight no-revisit coordination mechanism based on virtual pheromone blocking. We experimentally analyse how team size, localisation fidelity, and coordination influence coverage, hotspot recall, and redundancy. Results show that particle filtering is essential for reliable hotspot engagement, coordination substantially reduces overlap, and increasing team size improves coverage but yields diminishing returns due to interference.

Figures (5)

• Figure 1: Supervisor visualisations: (left) ground‐truth richness $R_g$; (right) 3D true richness.
• Figure 2: Supervisor blob--vs--coverage maps before and after a sample $N{=}5$ run. Red cells denote covered hotspot regions; white cells are hotspot cells not visited; grey traces show robot trajectories; blue is background. The internal mask encodes each grid cell as $-1$ (visited background), $1$ (hotspot), or $2$ (covered hotspot).
• Figure 3: Visit–count heatmaps for team sizes $N=\{1,3,5\}$ (left to right). Brighter regions indicate cells visited more frequently. With more robots the explored area grows and hotspots are detected earlier, but redundant sampling becomes more localised around blob regions, reflecting increased interference at higher team sizes.
• Figure 4: Blob–vs–coverage maps for team sizes $N=\{1,3,5\}$ (left to right). In the $N=5$ case, the final blob appears only partially covered because the supervisor terminated the run immediately after all four hotspots were detected, before additional samples could accumulate around that region.
• Figure 5: Illustrative particle-filter behaviour during a five-robot run. The left panel shows live particle clouds; the right panel shows true robot poses overlaid on the richness map. Stable filters maintain tight clouds around true poses; unstable filters spread and drift, producing large localisation error.