Shepherding is the problem in which one of more shepherd robots attempt to guide a flock to a goal. The problem is difficult because it is highly under-actuated, dynamic, and requires coordination of movement among the team of shepherds.

image Learning to Herd Agents Amongst Obstacles
We propose the first known learning-based method that can herd agents amongst obstacles. By using deep reinforcement learning techniques combined with the probabilistic roadmaps, we train a shepherding model using noisy but controlled environmental and behavioral parameters.
image Shepherding via Deformable Shapes
The shepherds use a deformable shape (a discretized "blob") to represent the flock. The abstraction helps to reduce the search space and to more accurately model the configuration space. This allowed us to handle much larger flocks and teams of shepherds.
image Medial-Axis Graph-based Shepherding Behavior
The shepherds use the medial axis as a roadmap to connect intermediate milestones toward which they move the flock. We compared this simple approach to techniques using high-level planning, including RRT, EST, and meta-graph.
image Interactive Shepherding
In this work we used created an interactive system which allows a user to guide the position of the shepherd using a laser pointer. The simulation is projected onto a wall or screen and a camera tracks the position of the laser pointer.

List of MASC Research Pages
Computer Science @ George Mason University