Multi-Agent Warehouse Coverage & Cleaning
Efficient coverage and cleaning of unknown terrains using multiple ground robots — 2nd place at the AIITRA Robotics Challenge 2021.
The main objective of this project was to build an efficient multi-agent algorithm for cleaning an unknown terrain, for which we built Vox-Bot, a ROS package for a multi-robot platform. This was our solution submitted for the AIITRA Robotics Challenge 2021, where we secured second position among all the participating IITs and other prestigious colleges of India.
Description
Vox-Bot is a four-wheeled robot with omni wheels for maximum agility and mobility. The primary purpose of the robot is to perform vacuum cleaning autonomously in an unknown terrain. It houses a LiDAR for mapping and an SBC for all computational needs, with one high-power brushless motor for vacuum generation and four brushed motors for the exhausts. For the vacuum system, we conducted multiple studies on the airflow, which can be found in the Airflow section. Below, we explain our solution to the problem statement.
Pipeline
We modified the vanilla navigation stack offered by move_base in ROS, using move_base_flex to implement the architecture of the navigation stack below.
- We used multirobot_map_merge with rrt_exploration for mapping the environment.
- We used polygon_planner for planning the Boustrophedon path.
- The rest is custom-implemented, with Fortune’s algorithm used for computing Voronoi diagrams. More details can be found in our proposal.
Mapping
We used RRT exploration instead of standard frontier exploration to make it more efficient.
Optimal Coverage
We used Voronoi diagrams and a weighted-centroid algorithm to distribute the task between individual robots.
Boustrophedon Path
Generating Boustrophedon paths for the polygon given by the Voronoi cell each robot resides in.
Airflow Study
The vacuum works on the principle of the lower fan creating a pressure difference to suck in air, while the exhaust pushes out the air from the upper compartment for efficient vacuum generation. The fan is placed low for efficient cleaning, with a ground clearance of approximately less than half the wheel radius.
We analyzed the vacuum mechanism using the SolidWorks Flow Simulation tool to understand the behavior of the vacuum during actual operation. The simulation required us to bound our rotating regions with circular bodies to define the rotation boundary. We defined the inlet and outlet velocities as 0.6 m/s and 0.15 m/s, below the bot and at the exhausts respectively. As the simulation was internal, the image only shows the flow inside the bot, but the trajectory of the arrows makes it evident that, in real-world scenarios, the Vox would be an efficient vacuum design.
