ScrubBot: Autonomous Field Cleaning

a robotic solution for cleaning a field where obstacles present

About

This is a project in collaboration with the industry. The objective is to cover a field as fast as possible, where static and dynamic obstacles may present. We built a robotic solution with modules including perception, global path planning, local motion planning and control and state machine. The software is wrapped in ROS framework and tested in simulation and on a real Husky robot. We used only a top-view camera as the sensor for localization and mapping. The global shortest path planning is done by solving travelling salesman problems (TSP) combined with A* algorithm. For local motion planning and control, a non-linear MPC is deployed.

The project was along with the course RO47007 Multi-disciplinary Project at Delft University of Technology. The final grade for this project was 9.7/10 (top 1 in the class).

husky_obstacle

husky_simulation

Report & Code

Due to the non-disclosure agreement (NDA), the full-text report cannot be published online. But the package for the motion control module will be available soon.

Supplementary Video