Traversability-Aware Cooperative Path Planning for Human-UGV Casualty Evacuation

Abstract

Heterogeneous multi-robot path planning is a well-studied problem in which agents with disparate kinematic and dynamic models must coordinate to achieve shared objectives. These formulations, however, treat all agents as robotic-their cost models are mechanical and their traversability is sensor-derived. In human-robot teaming, the human partner remains relegated to command and supervisory roles rather than being modeled as a physical co-navigator with distinct mobility constraints and dynamic energy reserves. This work investigates joint path planning for a two-agent human-UGV team in search-and-rescue casualty retrieval scenarios. We model the human agent using the Pandolf-Santee metabolic cost model with fatigue-modulated speed, and the UGV using a rolling-resistance energy model with terrain-dependent speed limits. By exploiting the complementary traversability of each agent-the human's ability to traverse dense vegetation and shallow water versus the UGV's superior speed on open terrain and roads-we optimize casualty transfer locations, termed switch points, to minimize total mission time. Evaluated across multiple synthetic 1km2 environments with procedurally generated elevation and land-cover data, the optimized strategy reduces mean mission time by 5.3% relative to a human-only baseline and by 7.0% relative to a naive human-UGV strategy without switch point optimization, while reducing human energy expenditure by 17.8% relative to baseline. Notably, the naive strategy reduces human energy expenditure by a larger margin (22.4%) but incurs a 2% increase in mission time relative to baseline, illustrating that switch point optimization is necessary to realize time savings from human-UGV teaming.