Scout-based Route Validation and Re-Planning in Unknown Environments
Abstract: Autonomous multi-robot missions often rely on a larger, capable worker robot to execute a task along a pre-planned route through an environment that has not yet been mapped. Sending the worker directly is risky. The route may be blocked, too narrow for its footprint, or otherwise infeasible, and discovering this on the worker itself is expensive and sometimes unrecoverable. This thesis studies scout-based route validation: deploying a smaller, agile scout robot to determine, ahead of time, which parts of the worker's candidate route are traversable, which require detours, and which are unreachable.
The central difficulty is a fundamental information gap. The scout operates in unknown space, while worker feasibility can only be evaluated in known space. The scout has a different footprint and different collision constraints than the worker, so traversability for the scout does not imply traversability for the worker. The scout must therefore actively plan viewpoints that observe the route well enough to reason about worker feasibility, while itself navigating the unknown environment safely.
We address this with a two-phase scout exploration framework built on a graph-based representation of free space. The first phase biases viewpoint selection toward observing the reference route. When parts of the route turn out to be infeasible for the worker, the second phase shifts the scout's attention to investigating detours through unobserved regions adjacent to the gaps, supporting replanning of the worker's route from observed free space. At termination, the framework reports a per-segment traversability assessment, the best replanned worker-feasible route it can construct, and an explicit characterization of the remaining uncertainty grounded in how exploration terminated.
Advisor: Professor Constantinos Chamzas
Committee: Professor Kevin Leahy, Professor Jing Xiao
Zoom link: https://wpi.zoom.us/j/99246538958