Control >> Cooperative transport >> Integration of Self-Assembly and Transport Behaviors s-bots

Integration of Self-Assembly and Transport Behaviors s-bots

This section details the experimental work concerning the integration of the modules for self-assembly and transport (see here for an overview of the modular control scheme). Currently we consider only non-blind s-bots, that is, s-bots that are able to localize the transport target.

In the following we detail an experimental study on self-assembly in the context of a cooperative transport task. We demonstrate the ability of a group of six independent s-bots to localize, approach and transport an object (called the prey) from its initial position to a home zone.

Controller

The two basic control modules on self-assembly and cooperative transport have already be introduced. We modified the transport module such that a) it makes use of the camera (instead of the proximity sensors) to detect the target, and that b) a connected s-bot stops transporting if it perceives a teammate that is still in the approach and assembly state (blue LEDs). In this way, the s-bots first connect to the prey, and then start transporting the prey towards the target.

Experimental Setup

The prey is of cylindrical shape. It is equipped with a surrounding ring which allows for physical connection via the gripper device. The mass of the prey is 2310g. Three s-bots are incapable of moving it, while four s-bots, connected in star-formation around the prey, can move it with an average speed of about 1cm/s. We observed that the performance varies depending on the type of spatial arrangement used. However, once a group of six s-bots is pulling and pushing, the prey is moved with satisfactory speed (i.e. at least half of the speed of a single robot without any load) regardless of the spatial arrangement used.

    

Figure 1. Environmental setup: overview (left), and potential initial placements and orientations around the prey (right).

Figure 1 (left) shows the robots' environment. The prey is initially put in a fixed location. It has to be moved across a circular line delimiting a home zone around a light emitting beacon. If moved in a straight line, the distance covered by the prey to enter the home zone is 125cm.

The initial position of each robot is assigned randomly by uniformly sampling without replacement from a set of 16 specific starting points; independent of the position, the orientation is chosen randomly from a set of 4 specific directions. The 64 potential placements of a single robot are illustrated in Figure 1 (right).

Results

Figure 2. Illustration of the task: self-assembly and transport of a heavy object to a home zone.

The task is to let six s-bots self-assemble and transport the prey from its initial position to a home zone (see Figure 2. Once the prey is entirely inside the home zone, the trial is completed. If the prey does not reach the home zone within a fixed time period of 300s, the trial is aborted.

Figure 3. Results on integration: (a) Amount of s-bots successfully connected, and (b) Time period the group was busy self-assembling and transporting the prey inside the home zone.

In total, 30 trials have been performed and recorded on video tape. Figure 3 (a) shows for each trial, the number of s-bots which could successfully establish a connection. In 26 out of 30 trials, all six s-bots established a connection. Only in 5 out of 180 attempts of a single s-bot, no connection was established. According to the simple control strategy we use, connected s-bots start transporting the prey once they do not perceive any teammates which are in the approaching phase. Thus, the transport phase was not started in 4 out of 30 cases.

In the following, we study the 26 cases, in which all s-bots connected successfully in more detail. Figure 3 (b) illustrates the time period the group was busy self-assembling and transporting the prey to the home zone. The assembly time A is the time from the beginning until all six s-bots are successfully connected. The transport time T is the time until the prey arrived inside the home zone reduced by the assembly time A.

In 20 out of the 26 trials, the whole group could successfully self-assemble within 83 seconds, in the other trials self-assembly was successfully completed within 167 seconds.

Only in a single case, in which all s-bots assembled successfully, the prey could not be moved entirely inside the home zone. In this case, the prey got stuck just on the border line of the home zone.

In all other cases, the prey has been moved within a short time period to the target: the average transport performance of 8.20 cm per second is more than half of the speed a single s-bot can reach moving straight and without any load.

Since the prey is so heavy that it cannot be moved by a group of s-bots directly manipulating it using a simple pushing strategy (we observed that a group of three s-bots is incapable of moving the prey; there is not sufficient space available for a group of more than three s-bots pushing (i.e., not pulling) the prey towards the target), we have successfully demonstrated the utility of a self-assembling system in a cooperative transport task.

We have integrated the control modules for self-assembly and transportation. With an experimental study we have demonstrated the utility of the concept of self-assembly in the context of a transportation task. By self-assembling, our system could transport an object that was too heavy to be moved by robots that could directly manipulate it using a classical pushing strategy. Further experimentation with groups composed of blind and non-blind s-bots are planned to be carried out in near future.

Example movies:

• Self-Assembly and Cooperative Transport by 6 s-bots, heavy prey (2310g) (Quicktime, 6.5 MB).
• Self-Assembly and Cooperative Transport by 4 s-bots, heavy prey (2310g) (Quicktime, 13.2 MB).
• Self-Assembly and Cooperative Transport by 3 s-bots, light prey (700g), dynamic target (Quicktime, 7.7 MB).

References

• Groß R. and Dorigo M. Group Transport of an Object to a Target that Only Some Group Members May Sense, In Yao X., Burke E., Lozano J. A., Smith J., Merelo-Guervós J. J ., Bullinaria J. A., Rowe J., Tiňo P., Kabán A., and Schwefel H.-P., editors, Parallel Problem Solving from Nature - 8th International Conference, PPSN VIII, volume 3242 of Lecture Notes in Computer Science, pages 852-861. Springer Verlag, Berlin, Germany, 2004