Control >> Passing over a hole

Passing over a hole

One of the main features a swarm-bot can exhibit is the ability to assemble in physical structures that can solve problems a single individual cannot cope with. One example of such a problem is passing over a trough that would block the navigation of a single robot. In similar situations, physical connections serve as support for those s-bots that are suspended over the gap, so that the swarm-bot as a whole can continue moving.

We designed a set of experiments in order to test the ability of a swarm-bot to cover a gap of varying size. In a first instance, we used the controller developed for coordinated motion to test the ability of a swarm-bot in passing over small furrows. In a further set of experiments, we employed the controllers developed for hole avoidance in order to let a swarm-bot move coordinately and pass over small gaps while avoiding to confront with holes too large to be passed. This test is intended to demonstrate how the simple controllers developed for hole avoidance (see here) generalise to a collective discrimination mechanism between situations that can be faced by a swarm-bot from situations that could be too hazardous even for a large connected structure (see Figure 2). In this section, we describe the results obtained in this experimental activity.

In the scenario depicted in Figure 1, a swarm-bot is placed at the centre of an arena presenting a square trough and open borders. We use here the same controllers evolved for hole avoidance, described in here. Therefore, the controller takes as input the traction force perceived by the s-bot and the readings coming from the four ground sensor. Recall that ground sensors are merely proximity sensors pointing to the ground, suitably filtered to extract the information about the presence of an hole. This sensor can therefore be used also the estimate the depth of a hole or the width of a nearby trough, as they have an inclination of 30 degrees with respect to an horizontal plane.

Figure 1. Trajectory of a group of 9 s-bots that passed over a trough and subsequently avoided to fall out of the borders of the arena.

If a gap is small enough to be bridged, the swarm-bot should pass over it exploiting the physical support through mutual interconnections. If the trough to be passed is too wide, the action to be taken is avoiding it. Therefore, it is necessary to provide the swarm-bot with the ability to estimate the width of the gap that has to be passed. In order to do so, the ground sensors used for the hole avoidance task can be used. As mentioned before, when near a trough, an s-bot can roughly estimate its width using the ground sensors, because if the trough is not too wide an s-bot near the border would perceive the opposite edge, having different perceptions with varying width. Starting from the fact that small gaps will not be perceived or will have a small influence on the overall motion of the swarm-bot, we want to show that the evolved strategies for hole avoidance can generalise to the ``passing over a trough'' behaviour.

A qualitative analysis of the behaviour produced by the controllers evolved for hole avoidance when used in an arena presenting small holes reveals that: (i) if the width of the gap is small enough (2-4 cm), an individual s-bot does not perceive it as an hazard---the activation of the ground sensors is rather low---and therefore the swarm-bot can pass over the trough in the same way as described above. (ii) If the width of the gap is bigger, the individual s-bot perceive the trough via the ground sensors and reacts consequently. However, the s-bot may be pushed out of the borders by the actions of the remaining s-bots in the formation. In this case, it may reach the opposite side of the trough, bridging the gap and letting also other s-bots pass. (iii) If the gap cannot be bridged by the swarm-bot, a normal hole avoidance behaviour is performed and the swarm-bot will move away from the hole (see Figure 1).

Example movies:
• A swarm-bot composed of 4 s-bots passes over a gap 8 cm wide (MOV, 3.1 MB).
• A swarm-bot composed of 9 s-bots passes over a gap 12 cm wide (MOV, 5 MB).
• A swarm-bot composed of 16 s-bots passes over a gap 20 cm wide (MOV, 8.9 MB).

The collective behaviour of passing over a trough relies on a delicate balance between the forces exerted by the s-bots that touch the ground and the missing influence of those s-bots that are suspended over the gap. The suspended s-bots cannot influence the behavior of the group and the dynamics of the swarm-bot are governed by fewer s-bots. Every s-bot that perceive a hole will react trying to change its direction of motion and trying to influence the behaviour of the whole group by exerting a traction force. However, the bigger the size of the swarm-bot, the bigger the inertia of the physical structure. Once the swarm-bot reaches an edge, its inertia will cause some s-bots to be pushed out, over the gap. In fact, few s-bots have a small effect on the overall behaviour of the group. When a sufficient number of s-bots is suspended out of the arena, the forces exerted by those s-bots that reach the edge can be percieved by the whole group, and they will trigger a change in the direction of motion of the swarm-bot in order to avoid falling. If some of the suspended s-bots reach the other side of the trough, they start again to have an influence on the rest of the group. First they align with the current direction of motion, and afterwards they contribute to the gap passing behaviour pulling the whole structure on the other side of the gap. This emergent behaviour can be considered self-organised, as it depends on the numerous interaction among individuals and on clear feedback loops: the conformist tendency of the s-bots in following the average direction of the group constitute a positive feedback, while the tendency to avoid a hole of the individual s-bots and the missing influence of the suspended s-bots constitute the negative feedback.

Control >> Passing over a hole