Copyright © Adrian Bowyer 2000
Keywords: biomimetics, social insects, construction, self-assembly, robotics.
The purpose of this Technical Report is to describe some initial ideas
and experiments on the project. It is hoped that it will form the basis
of a research grant application.
.
The robots will be electrically powered, and will have re-charging stations
to which they will return when their batteries get low (see the section
on powering the robots below). These stations will also clean off any accumulated
building material that may be impeding the robot. (An appealing alternative
would be to have the robots groom each other on a chance encounter, but
we will not rely on this initially; see below.) Finally, the stations will
replenish the robots' supplies of the building material.
The robots will have on-board micro-controllers that will be sufficient
to decide their behaviour and to make them autonomous. They will all communicate
with a single data-logging computer using Bluetooth [2]
or by means of infra-red links. This computer will not be controlling the
individual actions of the robots---social-insect building (and other) behaviour
is an emergent property
of their individual autonomous actions [6]---but
it will be used to monitor what is going on, and to change strategy when
necessary.
Each robot will also have access to its position and orientation in space using infra-red/ultrasound triangulation and tilt sensors. The position information ought not to be needed if the behaviour is properly programmed, but it will still be useful for measuring and gathering statistics.
The software controlling each robot will be downloaded from the central computer using the link. This will make it simple to change the behaviour of each robot as construction progresses.
It is not intended that the robots will move very quickly. We wish to
demonstrate that our proposal represents a feasible method of construction
and we have no intention of competing on speed with conventional building
techniques initially. Slow robots will be more efficient in their use of
electrical energy, prolonging battery life. They will also be easier to
program, as it will be possible to ignore dynamic considerations in movement
and to concentrate on kinematics.
The foam is very light, is non-toxic, and expands on being deposited as the propellant driving it evaporates within it. This last fact means that a robot can have small foam and propellant tanks and still do a lot of building before they need to be replenished.
This project is intended to prove the concept of biomimetic robot building. There is no reason, once this has been established, why other materials could not be used as long as they can be supplied to---and carried by---the robots. Ordinary cement comes to mind as an obvious alternative for a follow-on project.
Figure 1: A small arch made by repeatedly depositing by hand lumps of diphenylmethane 4,4'-diisocyanate foam similar to the example on the right. In front is a foot rule.Figure 1 shows a small arch made by repeatedly depositing lumps of foam about the size of the isolated example piece on the right of the figure. The arch was made by creating a layer of lumps by hand, allowing them to set (which took about 15 minutes), and then putting another layer on top. The finished arch was strong enough to support a person standing on it. The addition of a small amount of water to the foam as it was deposited would speed up setting times.
The foam has a density slightly higher than that of expanded polystyrene.
times bigger
than the individual termites that build them. This ratio is only exceeded
by the largest human cities.
Considerable research has been done by a co-applicant of the author's [4] and others on the nest-building behaviors of termites and other social insects. In particular it has been shown that a single set of building rules can lead to a wide variety of useful structures if the build material emits a pheromone the strength of which depends on how recently the material was laid down, and which is carried by air currents.
At a more basic level, a multiply-arched structure can be achieved by the following simple algorithm, which was derived from observations of social insects:
within a range 
?
look for a pile or arch within a distance 
.
and 
, of course, but the
right values will generate self-supporting structures of considerable complexity.
We have written a
Figure 2: Three simulations of the simple termite building algorithm, showing how structures such as crenellations, arches, and flying buttresses emerge automatically.C++ simulation program for investigating such building behaviour. Figure 2 shows three results of running it on the above simple algorithm with different sets of input parameters.
If the robots can walk, climb, sense their immediate surroundings (probably
by touch and smell---see below), and deposit lumps of building material,
they can follow this procedure---and similar ones.
In either case, it is important that the protective oil or grease is not shed onto the structure being built, as that would obviously prevent adhesion where it was wanted.
However, a better scheme may be to use the lotus effect (named after the plant, Nelumbo nucifera) [1]. Lotus leaves are covered in a hydrophobic wax that, crucially, has a special microstructure to its surface roughness that not only repels water, but ensures that any water running over the surface carries away the maximum amount of contaminants (in the case of the plant, the principal concern is to remove infectious fungal spores). Figure 3 shows a lotus leaf being cleaned by the effect.
Figure 3: A lotus leaf that has been contaminated by sprinkling a fine red powder on it (Sudan-III pigment powder: 1 - 20m, Merck) cleaning itself using rain (after Barthlott and Neinhuis [1]).
Biomimetic research into this in Germany has already led to the
commercial development of a paint [5]
that could make a good basis for a robot covering.
The recharging station could also be equipped with rotating brushes (like a car wash) to remove any foam adhering to the robots.
The cleaning problem will be minimized by getting the robots to avoid areas where foam has been very recently laid down (see sensors below).
Figure 4: A crude claw made from sewing pins soldered to copper wire. This supported a load of over 1.5kg. Dragging it over the foam with no extra downward force was sufficient to lock it into the surface. It could be removed by reversing its direction; this required no discernible force at all.The solidified foam is strong, and its surface is quite hard. But it is easily punctured by a sharp object. If the robots' legs were to be equipped with claws (possibly modelled on squirrels' feet, which are perfectly attuned to climbing both up and down vertical surfaces---see Figure 4) it ought to be possible to have them climb. The damage done to the foam surface should not be significant. It may even be possible to make the robots cling underneath an overhang.
The robots would also be designed sufficiently robustly that occasional falls from a height would not harm them.
The liquid foam in its tank needs continuous agitation, which will probably be achieved using a disconnected magnetic stirrer like those used in chemical experiments. It also needs a gas propellant. This is usually butane, which obviously will need to be handled carefully and ventilated well. But it ought to be possible to design the foam nozzle such that the foam is in its pressurized liquid form right up to the point of application. An auxiliary line from the propellant tank could then be used to blow clean the very last section of the applicator after foam has been deposited from it.
Clearly this problem is related to the cleaning problem discussed above, and the lotus effect may well prove beneficial here too.
We propose to mimic the social insects' use of pheromones to control their building behaviour. Semiconductor gas sensors are readily available that will detect a variety of gases and vapours, and equipping the robots with these would allow them to detect recently deposited foam if it were laced with the appropriate vapour-emitting solvent. Three or four sensors at the extremities of each robot (possibly also on the legs) should allow concentration gradients to be roughly estimated. Ethanol or acetone would work quite well; Polypag have been consulted on this, and say that acetone has the advantage as it would not affect the foam chemically.
It may prove advantageous to have one chemical marker for the foam, and another as a trail left by the robots all the time to mark the paths they take. Though gas sensors are available for specific substances (like ethanol and acetone), the manufacturers [3] have also been consulted and say that there would be considerable cross-interference between two sensors nominally intended for two different organic solvents. However, a reducing-agent sensor would allow a clear discrimination to be made. Sensors are available for chlorine, which is obviously too toxic for this application. But those sensors may well also respond to iodine, the solid form of which emits small amounts of vapour under ordinary conditions, and which is comparatively benign. A suspension of small quantities of iodine in water may make a good second marker substance.
It would be possible to equip the robots with small CCD cameras, but we hope to manage without this. In particular, it should be recalled that termites are blind.
Figure 5: An experimental robot leg made from model aeroplane servos. The knee joint weighs only 6g. The microcontroller and three H bridges outlined in red on the left are all the electronics needed to control the leg.
We have conducted experiments on these, with the preliminary results
that, though the mechanics of such devices are excellent, the requirement
for them to be driven by a simple low-frequency PWM signal, which is an
inheritance from the early days of radio-control, makes them prone to jitter
and hunting. A good solution is to keep the servo mechanics and the feedback
potentiometer, to throw away the electronics, and to drive the servo motor
from a microcontroller with inbuilt A-to-D converters such as a PIC [8]
feeding into an H-bridge. One PIC can drive several servos, and thus can
be used to control an entire leg. This mimics a widespread biological phenomenon---that
of autonomic control of part of an organism by local nerves. PICs have
inbuilt UARTS, and so arranging a number of them in a token ring is a simple
low-wire-count method of data distribution and control if several are to
be used.
It may transpire that ordinary nickel metal-hydride batteries are sufficient
to power the robots. But walking robots are notoriously power-hungry, even
if power conservation is explicitly programmed into all their servo controllers.
An alternative that may be worth investigating is powering the robots with
a (well silenced) model aeroplane engine
connected either to a DC motor acting as a dynamo---which would be simple,
and could double as a starter motor---or to a small alternator---which
would be more efficient. These engines can easily generate several hundred
watts, and a volume of fuel the size of a battery pack would let them run
for quite a long time. They could also be refueled in seconds. Given
the falling costs and rising energy densities of fuel cells, a methanol
fuel cell might be another alternative[7].
A particularly strong aspect of the proposed project is that we have already written, as part of these initial investigations, a complete simulation program for robot/social-insect building behaviour. This will allow us to try out a wide range of building algorithms speedily before committing the robots to the much slower task of doing the actual construction. The ability to set the robots building using only pre-tested algorithms that thus have a good chance of success should save considerable time. (We also plan to use the software in pure social insect behaviour research independently of this project.)
The vast majority of existing co-operating robot work is either in two dimensions, or is done only by simulation. To the best of our knowledge this project will be the first to build real useful three-dimensional structures on a human scale using robots programmed from observations of the social insects.
