Multi-Agent Ergodic Exploration under Smoke-Based, Time-Varying Sensor Visibility Constraints

TL;DR

This work tackles multi-agent IPP under time-varying sensor visibility caused by smoke in wildfires. It adopts a multi-agent ergodic trajectory optimization (ETO) framework driven by a visibility-aware expected information distribution ($\\Phi$) that is constructed from a smoke density field and uncertainty maps; one variant uses Shannon entropy to form $\\Phi_2$. The results show that incorporating visibility into the EID and using the Shannon-entropy EID significantly improves ergodic convergence and information gain compared to baselines and simpler formulations, for both static and moving targets. The approach enables faster, more robust wildfire monitoring and search-and-rescue planning in environments with dynamic visibility, and can be extended to heterogeneous, real-world drone teams.

Abstract

In this work, we consider the problem of multi-agent informative path planning (IPP) for robots whose sensor visibility continuously changes as a consequence of a time-varying natural phenomenon. We leverage ergodic trajectory optimization (ETO), which generates paths such that the amount of time an agent spends in an area is proportional to the expected information in that area. We focus specifically on the problem of multi-agent drone search of a wildfire, where we use the time-varying environmental process of smoke diffusion to construct a sensor visibility model. This sensor visibility model is used to repeatedly calculate an expected information distribution (EID) to be used in the ETO algorithm. Our experiments show that our exploration method achieves improved information gathering over both baseline search methods and naive ergodic search formulations.

Figures (5)

• Figure 1: Multi-Agent Exploration with Varying Visibility Constraints. Three information peaks have varying visibility according to the concentration of smoke over the peaks. Given that less information can be captured when observing peaks with low visibility, the multi-drone team prioritizes exploring peaks with high visibility.
• Figure 2: Time-Varying, Visiblity-Aware Multi-Agent Ergodic Exploration. Shown is a flow chart of our visibility-aware, time-varying ergodic exploration process. To begin, the underlying uncertainty distribution and smoke model are used to calculate the EID. Using this distribution, multi-agent ETO is performed. The agents follow these generated trajectories, and the uncertainty map is updated using the agents' measurements.
• Figure 3: Stationary Targets: Expected Information Measure Performance Comparison. The average iterations required to minimize the ergodic metric (left) and the percent reduction in uncertainty (right) for multi-agent ETO are compared across three different approaches for calculating expected information.
• Figure 4: ETO Trajectory Evolution using Shannon Entropy. For a sample uncertainty map, the EID is calculated from Shannon entropy at each time step. Ergodic trajectories are then planned at each time step over the current EID. It can be seen that at uncertainty peaks with high instantaneous smoke density, the expected information yield is low, and so the drones spend little-to-no time near those peaks. Later, as the smoke moves, some previously obstructed peaks become unobstructed and drones then visit those peaks. Over time, the total uncertainty decreases due to measurement by the drones.
• Figure 5: Exploration Method Performance Comparison. The percent reduction in uncertainty of our ergodic search method is compared to that of lawnmower and greedy search methods over randomly-generated uncertainty maps for both static and moving targets.