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

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.

💡 Research Summary

The paper introduces RANT, an ant‑inspired multi‑robot exploration framework that integrates particle‑filter localisation, gradient‑driven hotspot exploitation, and a lightweight virtual‑pheromone blocking scheme for coordination. Implemented in the Webots simulator, a team of differential‑drive robots explores a 10 × 10 m arena containing a hidden “richness” field composed of four Gaussian blobs. Robots receive noisy GPS, IMU, and wheel‑odometry data, which are fused by a particle filter (150 particles, systematic resampling when the effective particle count falls below 95 % of the total). The filter provides pose estimates that drive a three‑state behaviour controller: EXPLORE (biased random walk with obstacle avoidance and dispersion bias), BLOB (local gradient ascent toward high‑richness samples), and RECOVER (straight‑line motion away from visited hotspots).

A central supervisor maintains global maps on a 50 × 50 grid: the true richness field, a pheromone map, a visited mask, a blocked mask, and a visit‑count map. When a robot detects a hotspot it sends a BULLSEYE message; the supervisor then inserts a circular blocked region into the blocked mask and increments the pheromone and visit‑count maps. Pheromone values decay each timestep (P(t+1)=0.995·P(t)), providing a simple stigmergic mechanism that discourages re‑visiting already sampled hotspots while allowing the blocks to evaporate over time.

Three experiments assess the impact of team size, localisation fidelity, and pheromone‑based coordination.

1. Team size (N = 1, 3, 5) – Coverage increases from 62 % to 84 % as robots are added, but the marginal gain diminishes due to inter‑robot interference (collision avoidance, communication drop‑outs). Hotspot recall improves similarly, while the redundancy metric (1 − unique visited cells / total samples) drops from 45 % (no blocking) to 18 % when virtual blocking is enabled.
2. Localisation fidelity (PF‑on vs. PF‑off) – With particle filtering active, robots achieve a mean localisation error of ~0.12 m, frequently enter BLOB mode, and generate dense sampling around hotspots (≈1,200 sample requests per run). Without the filter, error exceeds 0.8 m, BLOB transitions rarely occur, and sampling collapses to <200 requests, demonstrating that accurate pose estimation is essential for hotspot engagement.
3. Pheromone blocking (enabled vs. disabled, N = 5) – Enabling global blocking reduces redundant sampling by roughly half and slightly raises hotspot detection rate (0.92 → 0.97). The blocked zones act as soft repulsive obstacles; however, overly large block radii impair overall coverage, confirming the need for careful parameter tuning.

Key insights: (i) particle‑filter localisation is a prerequisite for reliable gradient‑based exploitation; (ii) virtual pheromone blocking provides a lightweight stigmergic coordination that markedly cuts overlap without centralised path planning; (iii) scaling robot numbers yields diminishing returns because physical interference grows non‑linearly. The authors note that the current work is confined to a modest indoor arena and that future extensions should address larger outdoor deployments, dense canopy environments where GPS is unavailable, and hierarchical coordination for tens or hundreds of agents. Overall, RANT offers a coherent, experimentally validated architecture for robust multi‑robot exploration in noisy, uncertain settings.