Robotics Research Lab
USC Computer Science
USC Engineering
/ Research / Projects / Collective Construction with Multiple Robots

Objective Controller Robot Experiments Simulation Experiments Support


The objective of this research is to design a multi-robot controller based on minimalist communication principles that can effectively achieve the construction of simple 2 dimensional structures. This work does not exhibit a system that performs construction as fast or as precisely as possible. It rather explores different coordination strategies.


The controller is behavior-based and does only use local communication and no global map or other global information at all.

Figure 1: The Controller
Robot Experiments

The robot experiments were performed on a Pioneer DX2 equipped with:

The following pictures show a robot attaching a block to the end of the barrier.
Robot Robot
Robot Robot
Figure 2: Attaching a block to the barrier

One of the barrier the robot build during the experiments is show in the next pictures. In the back one can see the laser beacon which is used to determine the starting point and the orientation of the barrier.
Barrier Barrier
Figure 3: Barrier of 10 blocks build by a single robot


Contents: Single robot building a 10 block wall
Clip length: 3 minutes
Speedup: 10
Download: [MPG 30.4 MB]
Simulation Experiments

Two simulation experiments were performed so far. For the first experiment the robot controller as used in the robot experiments was used. In the second simulation the controller was extended by minimalist communication.

No Memory About the Last Block

For this set of experiments the controller was taken as used in the robot experiments. So the robots have basic construction capabilities and in order to lower the interference at the construction site they ask for permission to enter the construction area.
The robots have no knowledge about the color of the initial puck underneath the laser beacon. They do neither know where the beacon is nor where the pucks are located. Also they do not have any information about their own position, the position of other robots or the layout of the environment.
When a robot is powered up the goal controller randomly chooses a block type to focus on. Note this is different from the real world experiments. This change was necessary to prevent that all the robots start with the same block type. After the initial choice the goal controller alternates the brick type every time the robot sets a block down, successfully or not.
The following image shows the average barrier growth over time for 1, 2, 4, 6 and 8 robots.

Figure 4: Average barrier growth over time (without information about the last block)

Minimalist Communication

In the basic controller the robots only use communication to achieve clearance for entering the construction corridor. This kind of communication is necessary to keep interference between the beacon approaching and the attaching robot low. It does not provided any kind of useful information about the puck sequence of the barrier nor does it help the robots to organize themselves in terms of in which order they should attach the pucks.
For this set of experiments minimalistic communication is added. The robot that performs the build behavior checks for the last puck in the barrier before attaching a new puck in order to guarantee the pucks will stick together. This may result in either attaching the puck or turning away and dropping the puck. In either case the robot knows the type of the last puck in the barrier and broadcasts this information. Other robots that are waiting in the waiting zone around the laser beacon receive this message and now decided to either keep on waiting since their puck is of the wrong type or start asking for permission to enter the construction corridor. Robots that are not waiting but approach the waiting zone in the near future make the same considerations. This technique give the robots knowledge about the puck sequence in the barrier and the question is if this knowledge lets the robots perform better.
The following figure shows the average barrier growth over time for 2, 4, 6, and 8 robots using minimalist communication.

Figure 5: Average barrier growth over time (with minimalist communication)

Performance Comparison


This work is supported by DARPA Grant DABT63-99-1-0015, and by DURIP Grant N00014-00-1-0638.