Earth-Shaping with Heterogeneous Robotic Teams: From Sim to Real

This work presents an overview of autonomous earth-shaping and aggregate-forming tasks, which we address with a heterogeneous team of Unmanned Ground Vehicles equipped with shovels. Firstly, we introduce Sheperd, a Hardware In the Loop development tool integrated with ROS and physical engines like Gazebo and FlexHopper. We further address the problem of computational cost in these kinds of simulators, providing preliminary results to replace them with a probabilistic model for aggregate motion. We proceed by addressing the following major challenges: localizing the team of robots and mapping the aggregates, as well as efficient motion planning strategies. We address UGV localization and material mapping, utilizing vision-based UAV-UGV collaboration and optimization techniques for robustness. Then, we discuss motion planning, providing solutions to shape on-site material exclusively using pushing actions rather than more standard pick-and-place solutions.