Control >> Moving on rough terrain

Moving on rough terrain

This section describes the experiments carried out to assess the capability of a swarm-bot to move coordinately on rough terrain. For this purpose, we use the controllers developed for the coordinated motion on flat terrain. The strategy evolved for coordinated motion proved to be robust enough to cope with uneven terrains when tested in simulation. Here, we present the results obtained with a real swarm-bot.

Experimental setup

In order to test the capacity of the controllers evolved for flat terrain motion to generalize their ability on rough terrain, we devised the following experiments:

• Climbing slopes A swarm-bot is positioned at the center of a square arena that has slopes of varying inclination on the four sides (see figure 1 left). The arena is made up by a flat central surface shaped as a square of 60 cm per side. On each side there is a slope 20 cm long whose inclination can vary. The swarm-bot should prove capable of coordinately moving and climbing the slopes. The difficulty of these experiments resides in the fact that the weight of the turret may create a traction force when the s-bot is moving up a slope, due to gravity. This force may therefore influence the evolved controller.
• Climbing steps A swarm-bot is positioned at the center of a square arena whose edges are steps high enough to prevent a single s-bot to climb (see figure 1 right). The square's edges are 110 cm long and 2 cm high. Here, a swarm-bot should prove capable of coordinately moving in order to climb the steps. This test is intended to show how the swarm-bot can cooperatively face navigation problems that would block a single s-bot.

Figure 1: Experimental setup for experiments in coordinated motion on rough terrain. Left: Climbing slopes. Right: Climbing steps.

Results

The controller evolved in simulation can be used with real s-bots, provided that a front inversion mechanism is implemented to cope with the hardware limit in the turret rotation. We tested the evolved behaviors in the environments shown in figure 1.

Climbing slopes

When tested in an environment presenting slopes, a swarm-bot can generally overcome them exploiting the cooperation among the s-bots. However, the strategies evolved for coordinated motion do not necessarily generalize in such environments because they were evolved on the perception of the traction forces between turret and chassis of an s-bot. In fact, when an s-bot is positioned on a slope, it will perceive a traction force pointing down the slope, the steeper the slope, the stronger the force. This perceived force may influence the behavior of an s-bot, which may try to cancel this pulling force turning in the direction of the slope. In a similar condition, the coordinated motion behavior may be disturbed by the presence of a slope.

We tested the coordinated motion behavior with slopes of different steepness: 25% and 50% respectively. The s-bots have to coordinate first, choosing a common direction of motion, and then they have to move up the slopes in order to cover the maximum possible distance. We observed that a swarm-bot has no difficulties in climbing a slope with 25% of steepness. We tested a linear swarm-bot configuration and a square formation, both composed of 4 s-bots. Tests with 50% steepness slopes were unsuccessful: the swarm-bots are not able to overcome the slope. The 50% inclination is too steep to allow coordinated motion to function because, as mentioned before, the steepness of the slope generates a traction force that leads the swarm-bots toward the center of the arena. We observed that every time the swarm-bots approached a slope, they changed direction of motion, and normally ended up in circuiting around the center of the arena at the boundary between flat area and the slopes.

Example movies:
• 4 s-bots in linear formation confront with 25% steep slopes.
• 6 s-bots in a square formation confront with 25% steep slopes.

Climbing Steps

The slopes we used to test the coordinated motion behavior in principle can be also overcome by a solitary s-bot. The experiments presented above give us an indication on how the coordinated motion behavior generalize in such environments, but they do not show a situation in which the swarm-bot can collectively perform a task that a single robot cannot. Therefore, we designed an experimental setup in order to test such ability. A single s-bot cannot climb even a small step, because it topples down when confronting with it. A swarm-bot, on the contrary, has the possibility to climb high steps, exploiting the cooperation of the s-bots composing it. However, the coordinated motion behavior developed for motion on flat terrain could get disturbed by the presence of steps, as a collision with a step could cause a traction force perceived by the s-bots assembled in the swarm-bot. Following this traction force, the swarm-bot might change direction of motion.

We tested the coordinated motion behavior with a linear and a square swarm-bot. We observed that, exploiting cooperation, the swarm-bots are able to overcome the step without any difficulty. The linear swarm-bot may topple down if all the s-bots approach the step simultaneously. This problem is not present when the swarm-bot uses the square formation, which is therefore more suitable for this particular task.

Example movies:
• 4 s-bots in linear formation in an arena with steps
• 4 s-bots in a square formation in an arena with steps.

More examples

We also performed tests of motion on rough terrain using two different material to obtain a rough terrain:

• Brown plastic isolation foils for a regular rough terrain. See the movie: 4 s-bots in linear formation.
• White plaster bricks, for a very rough terrain. See the movie: 4 s-bots in linear formation.

Notwithstanding the characteristic of the terrain, the coordination of the s-bots in the initial phase was always successfull, suggesting that the evolved strategies are robust enough to cope with many types of rough terrains.

Control >> Moving on rough terrain