Human-Centered Autonomy for UAS Target Search

TL;DR

A human-centered autonomous frame-work is presented that infers geospatial mission context through dynamic feature sets, which then guides a probabilistic target search planner that produces effective task mental model alignment and 15 times more efficient guidance plans then current operational methods.

Abstract

Current methods of deploying robots that operate in dynamic, uncertain environments, such as Uncrewed Aerial Systems in search \& rescue missions, require nearly continuous human supervision for vehicle guidance and operation. These methods do not consider high-level mission context resulting in cumbersome manual operation or inefficient exhaustive search patterns. We present a human-centered autonomous framework that infers geospatial mission context through dynamic feature sets, which then guides a probabilistic target search planner. Operators provide a set of diverse inputs, including priority definition, spatial semantic information about ad-hoc geographical areas, and reference waypoints, which are probabilistically fused with geographical database information and condensed into a geospatial distribution representing an operator's preferences over an area. An online, POMDP-based planner, optimized for target searching, is augmented with this reward map to generate an operator-constrained policy. Our results, simulated based on input from five professional rescuers, display effective task mental model alignment, 18\% more victim finds, and 15 times more efficient guidance plans then current operational methods.

Figures (6)

• Figure 1: Search & rescue incidents require operators to fuse multiple sources of information to direct an aircraft. Our human centered architecture fuses a variety of unstructured operator inputs to inform an optimal planning process that generates a policy for autonomous execution.
• Figure 2: Graphical model used in input fusion algorithm with observed variables (white) and unobserved variables (grey).
• Figure 3: POMDP observation model.
• Figure 4: Inputs from five rescuers overlay the resulting geospatial preference distribution. The lower row shows the progressive addition of rescuer 3's inputs and resulting concentration of reward around trails and streams within prioritized areas. Note that in execution, inputs are fused simultaneously.
• Figure 5: Example trajectories of our planner and the baseline simulated using rescuer 1 input data from Figure \ref{['fig: operator_data']}. Our approach finds the target in 329 timesteps, baseline finds the target in 858 timesteps. Trajectory opacity decreases with search time.