Image Image Image Image Image Image Image Image Image Image

Bartlett School of Architecture, UCL

Scroll to top


No Comments

Solar Powered Robotics: Resource Constrained Interactivity

Solar Powered Robotics: Resource Constrained Interactivity

Since their introduction in the last 60 years, solar panels represented not only an opportunity to use renewable resources for general utility, but also a powerful tool to feed solar powered robots and artworks. Thanks to a direct conversion from solar radiation to electrical energy, solar powered robots become more autonomous and less dependent on more restricted sources of energy. Robots do not need to be stationary anymore, but they have the possibility to travel and migrate, trying to reach a condition of internal autopoiesis and a niche in its surrounding habitat. In a scenario of a self-sufficient robot, we may ask to ourselves how it can find a compromise between the available stored energy and the one the robot spends to move finding the way to prolong its activity.  Answers can be found by analysing the physical hardware components, understand the balance, the inputs and the outputs and propose software strategies for a behaviour focused on the energy conservation.


This study focuses on the relationship between renewable resource powered robots and energy consumption, investigating on strategies for actuators and behaviours that extend and prolong opportunities for object-to-subject interaction. Specifically, the research analyses the hardware components of a solar powered pneumatic system, driven by artificial muscles, and try to increase the energy conservation through different solutions. The energy can be guarantee only through solar panels, leading to a more autonomous system, trying to find a balance between the energy it collects, along several variables, and the energy it spends. Moreover, software strategies are examined to increase endurance and longevity given to a resource constrained mobile robot by introducing biologically inspired behaviours.


Aim of this study is to analyse the exact consumption of the hardware and propose energy-conservation behaviour strategies to understand, through projections, if a robot can modulate its interactivity and successfully reach a balance, a niche in its environment.


 Photovoltaic panels are widely used as a main tool for making a robot self-sufficient, in terms of energy source, by converting solar radiation into electrical energy. This technology represents one of the best and efficient ways for robots to harvest energy even though their efficiency is strictly connected to several variables, among them the geographical position, weather conditions and the site.

BEAM Robots

BEAM Robots, which stands for Biology, Electronics, Aesthetics and Mechanics, are simple robots built with analogue circuits, rather than a microcontroller. Simon Fraser has been creating a wide number of BEAM solar robots using extensively small photovoltaic panels as energy source.

Power Smart BEAM Robot, rotating to point towards the brightest part of the environment.

(Source: Simon Fraser).

Solar powered BEAM robots are strictly dependent on the environmental condition. For instance, if a shadow is casted, from clouds or from the local environment, on the solar panel, then the robot will stop to work. In addition, since BEAM robots are analogically programmed, it may be difficult to program them accordingly to specific data, such as sun vectors or coordinates. For that reason, a simple microcontroller can be more effective and flexible.



Sky Art is an important series of conference that seeks to challenge the relationship between technology, environment and art. The term was coined in 1969 by Otto Piene, denoting temporary artworks connected in the urban sky space. Later, in 1981, Sky Art became a series of four conferences, directed by Piene and took place for the first time at Massachusetts Institute of Technology (MIT), organised by the Center for Advanced Visual Studies (CAVS) where important artists, architects, engineers and scientist joined together exhibiting various projects connected to the sky and space. Among them, Mark Drela, Paolo Soleri, Rockne Krebs. Aim of these four conferences, which took place from 1981 to 1986 in various countries, consisted an attempt to document the emerging field of sky art and to discover and realize forms of art in space (Goldring 1987). Desert Sun/Desert Moon was the conference which was located in the natural landscape Alabama Hills, in California (USA), from June 8 to June 16, in 1986. Lowry Burgess, Joe Davis, Elizabeth Goldring, Otto Piene, Tom Van Sant, Christopher Janne were the main artists who displayed their work. Joe Davis exhibited the Desert Crawler, built from recycled materials and electronics. It consisted of a solar powered robot capable of drawing lines in the sand, while using the converted solar energy to move a few feet each day. Davis imagined his robots creating planetary graphics in the lunar soil and on the desert of Mars. For Joe Davis, using emergent technology such as the solar panel was an important breakthrough to realize a machine which could power itself. However, the goal of the Desert Crawler was not about prolonging the robot’s activity as long as it could.

Desert Crawler, Joe Davis. Recycled Steel and Aluminium, electric motor, solar panels, rechargeable batteries; 56 x 36 x 22 in, weight 200 pounds; 1986 (Goldring, 1987). (Source: Otto Piene).



In 2008 Smallwood built “The Sound Lanterns”, a series of devices strictly coupled to photovoltaic panels with no battery or voltage regulations. Each lantern consists of noise-making analogue circuit and 8 Ohm speakers fixed in a glass jar, connected to a 5W solar panel, which could provide from 3 to 16 volts (Cook and Smallwood, 2010). Due to the lack of an accumulator, to weather conditions variables, and to the direct connection to the solar panels used as a sensor, the behaviour of each lanterns was extremely reactive to the solar panels input, leading to significant changes. In comparison to the Desert Crawler, the performance of the Sound Lanterns was disadvantaged due to its high dependency on the presence of sunlight. Similar to the majority of BEAM robots, there is no gradual transition of the activity between the two conditions.

The Sound Lanterns, Smallwood S. 5W solar panels, analogue circuits, 8 Ohm speakers, glass. 2008. (Source: Cook P. and Smallwood S).


SOLA — Sustainable Orchestras of Laptops and Analog

A more recent experiment starts to focus on how to reduce energy consuming so that energy supply can be extended, providing not only conservation and ecological improvements but also alternative performance sites. Starting from 2005, Princeton Professors Perry Cook and Dan Trueman, founders of Princeton Laptop Orchestra, explored with their students how to integrate solar powered devices (computers and speakers) for music performance. Their goal was to investigate the minimum system and power to feed a PLOrk station (Princeton Laptop Orchestra work station), including an Apple laptop powering a multichannel interface (60 Watts) and a hemispherical speaker (48 Watts). The speaker is fed by a 12Volts 4-amp power supply, making its maximum power of 48 Watts, while the 2 GHz Apple Macbook required a maximum continuous power of 60 Watts.

PLOrk station, Cook P. and Smallwood S. , Solar panel, Macbook, hemispherical speaker, charge controller. (Source: Cook P. and Smallwood S).

After some tests, Professors discovered that the laptop was the most expensive device, due to the huge amount of energy from the screen. Analysing three different screen settings, they could reach the maximum efficiency by dimming the screen to black, providing about 250 minutes of autonomy, in comparison to a Macbook with the screen on and software running (Processing) which guaranteed the half of autonomy. In addition, another dissipation of energy was observed while converting from DC to AC (140 Watts Inverter) and to AC to DC again (MacBook power supply).

Internal Battery Performances. (Source: Cook P. and Smallwood S).

As a result, the musicians were able to play music from anywhere between 2 to 6 hours depending on the two groups of the constraints: devices (computer screens and external hard drive) and the site and weather conditions. Their goal was to optimize the energy resource by understanding and testing which devices can be modified in order to consume less energy, suggesting using no visual feedback onscreen. This poses the question: what if a dynamic approach could be set? What if screen or speakers (volume and switch) could be proportionally balanced according to the available energy?



In 1995, a group of artist and engineers called SolArt Global Network with Jürgen Clause as founder of the organization, contributed to underline the importance of renewable energy in the Solar Age (Rifkin and Howards, 1985), such as the solar radiation, by applying the “biospheric art” (Claus, 1995). The aim of SolArt Global Network was to introduce artistic works based on scientific researches in order to promote ecological awareness. A range of devices such as photovoltaics panels, mirrors, lights has been used to build the installations. The major artists who took part in this festival were Peter Erskine, Alex and Martha Nicoloff, Jürgen and Nora Claus. One example is the Solar Crystal. Inspired by Buckminster Fuller’s concept of the tetrahedron as the “minimum and simplest structural system in the Universe”, Jürgen and Nora Claus, with the collaboration of engineer Wolfgang Krug, designed and built a steel frame supporting coloured Plexiglas tetrahedrons, an array of photovoltaic panels (Siemens panel type M55, 1300x330mm), connected to 6 batteries and a low-energy 25watt lamp (Claus, 1985).

Solar Crystal, Solar installation by Jürgen Claus and Nora Claus with Wolfgang Krug. Plexiglas, steel, photovoltaic panels and various materials. Jülich, Germany. (Source: MIT Press).

The kinetic structure was capable to collect and convert solar radiation into chemical energy, completing charging the battery during the day and lighting up in colours during the night. As a result, the duration of the internal illumination of the sculpture is rigorously dependent on the amount of solar energy the photovoltaic panels have converted during the day, giving every time different outcomes according to the dynamic environment. Although Solar Crystal is a significant example of “biospheric” art, aiming to make people aware of the energy consumption, the system is not able to adapt itself to the different scenarios and the environment.



Plantas Nomadas, or Nomad Plants in English, is an artwork by Gilberto Esparza, a Mexican artist who imagines creative solutions and experiments for the environmental issue, through art and biology. The robot is designed to explore the contaminated riverbank in search of resources to sustain itself. It has two main sources of renewable power: microbial fuel cells capable of converting the pollutants of the river into energy, cleaning water and feeding the plants accommodating its back.  Additionally, small panels are installed on the top to harvest solar radiation, as a supplementary resource of energy. Fuel cells are not enough to produce sufficient amount and constant power since the robot has to seek and extract water, and further processing it. The procedure requires a sufficient amount of energy in advance in order, for the robot to convert the contaminated water. In addition, it is problematic in this scenario that solar panels are not oriented towards the sun and the surface used may be too small to produce enough energy.

Nomad Plants, Robots by Gilberto Esparza. Mexico (Source: Plantas Nomadas).




Nature is offering us the most sophisticated examples of all kinds of structures, materials and different types of responses from organism to adapt themselves and survive in their environment. During the last century engineers started to deeply analyse these phenomena and reproduce nature’s principles. Biomimetic (Biotecnics/biotecnique) consist in the observation and analysis of organisms’ behaviour/patterns in order to apply it into the design principles. This discipline explores the characteristics of all kind of organisms, from cellular to the animal world. On the other side, interesting approaches, mostly regarding the behaviour, can be taken from the ethological and botanical world, which studies the animal behaviour in its own environment: nature. When an animal is reaching a balance in its environment, when it can find food, recognise and escape from the enemies and interact with its community, it finds its niche, a stable status in its habitat. Obviously, due to the complex and constantly dynamic nature of the environment, a successful animal should be able to adapt to the new conditions to survive. Observing some pattern of behaviours in the animal and botanical world can be useful to imagine a niche for a robot or an artwork which can negotiate with the energy collected and succeed to keep enough of its energy until an energy source, such as the sunlight, will be available again. To understand the correlations, in the animal behaviour, between stimuli and response, we need to define three behavioural classes:


  1. Reflexes: rapid, automatic involuntary responses triggered by some certain environmental stimuli, where usually duration and strength of response is strictly dependent on the stimulus one.


  1. Taxes: Behavioural responses that orient the organism towards or away from a stimulus. In this case we can assist, for example, to a positive or negative tropism. For example, wood louses use their compound eyes to orient themselves towards the light as well as


  1. Fixed-Action Patterns: In this case the correlation in terms duration, speed and strength between stimulus and response is not direct. In fact, a stimulus can trigger a response patter that can last longer, even after the stimulus disappear and can vary according to the internal state of the animal.

Plants’ behaviour is influenced by the daily rhythms, known as Circadian Rhythms. According to the environment, leaves can occupy different position during day and the night, with the so-called nyktinastic movements (sleep movements). During natural conditions, nyktinastic movements are synchronized by the shift day-night. However, if the plant is moved into a different environment characterised by continuous darkness for several days, the movements starts to change their patterns, as shown in the following figure.

Continuation of petal movements in Kalanchoe blossfeldiana in constant darkness (shaded area). Ordinate: rising values = opening, falling values = closing of flowers (Valenzeno, Pottier, Mathis, Douglas 1990). (Source: Valenzeno, Pottier).

A different response connected to the energy shortage and drastic changes from the environment can lead to a long-lasting phenomenon of hibernation. After an animal finds or makes a living space (hibernaculum) that protects it from winter weather and predators, the animal’s metabolism slows dramatically, utilizing its body’s energy stores to survive.  When spring weather arrives, the animal’s body changes again its metabolism and leaves its hibernaculum, starting again the research of food.Some frogs, like the wood frog (Rana Sylvatica) and the spring peeper (Hyla crucifer), have the ability to literally freeze in the ice, stopping heart beating and breathing. It can appear dead, but, actually, but a high concentration of glucose in the frog’s vital organs prevents freezing.


The current examples examined lacked dynamic improvement for energy conservation. Robots and installations could only work until the energy accumulated during the day has been exhausted. Availability of Sun is limited, not constant and it can vary during the day according to a multitude of variables. BEAM robots, which usually uses capacitors to temporary store the energy accumulated, are strongly affected by the environment and few versatile when it comes to programming. For most applications, involving for examples computers, microcontrollers and devices which needs a smooth and constant source of energy, a raw system of solar panels is not recommended. Regarding the world of ethology and botany, the examined examples can suggest different strategies for energy-conservation behaviour. In fact, hibernation represents a way to limit the activity of a body, protracting the complete shortage of energy. On the other side, a behaviour based on cycles like the Circadian Rhythms, can represent an opportunity to regulate the use of energy in near time. By so doing the robot can calculate the availability of energy in the next future and regulate the one that has been already accumulated.


With the objective of prolonging the activity of a system, it is possible to operate not only on the less energy-consuming actuators choice but on the behaviour of the robot as well.  Therefore, the purpose of the following experiments is to analyse the energy consumption of different actuators and provide, with a given scenario, example of behaviours for constraining interactivity and protracting energy availability.


According to the conservation energy principle, energy cannot be created nor destroyed but it can only transform to one form to another. Thus, the total amount of energy remains constant. Transformation, as the second law of thermodynamic states, can only be transformed in one way, dissipating, therefore, decreasing its quality. This phenomenon is known as entropy, which represent the quantity of unavailable energy present is a system.

The energy balance is expressed by the formula:


ΔE = Ein — Eout                                                                                       


 where ΔE — energy variation    [Joule]

Ein — energy input           [Joule]

Eout — energy output       [Joule]


It means that the change (internal, kinetic, potential…) in the energy content of a system is equal to the difference between the energy input and the energy output by heat, work and mass (Çengel and Boles, 2015). The system, in order to be self-sufficient, and must satisfy the following equation:


Qi > Qo                                                               (2)


where Qi – energy in     [Joule]

Qo — energy out     [Joule]

Qi represents the energy (light energy from the Sun) converted and accumulated in the battery (chemical energy) and Qu represents the energy required from output actuators such as speakers, lights, motors, solenoids, microcontrollers.


Scheme of main components and variables of a general solar powered device



Solar energy has been used by humans from centuries, since the 7th century B.C. to light fires with magnifying glass materials or in the 3rd century B.C., when the Greeks and Romans used to harness solar power with mirrors to light torches for religious ceremonies. In 1839 Alexandre Edmond Becquerel made discovered and analysed the photovoltaic (PV) effect through an electrode immersed in a conductive solution and exposed to the light. Since then, a lot of researches and experiments have been done to develop a practical system capable of converting sunlight into electricity, until the 1954 when, in the Bell Labs, the scientists Pearson, Chapin and Fuller succeed to produce the first silicon solar cell with an efficiency of 8% (Fraas, 2014). In the last 15 years photovoltaic panels rapidly increased in the market, becoming more and more efficient and cheap, even though the extraction and production of the main material, silicon, requires a consistent amount of energy, leading to the use of new PV technologies such as devices based on molecular absorbers (Nelson and Emmott, 2013). Commercial PV panels belongs to different typologies with a different efficiency which determines the capability of a solar panels to convert solar radiation into electricity. Precisely, the conversion efficiency of a solar cell is defined as the ratio of its electric power output to the incoming light intensity that strikes the cell, in standard test condition (Grätzel, 2007). The most used PV typologies are based on mono-crystalline or poly-crystalline silicon cells, along the thin-film (Amorphous Silicon Solar Panels) and Concentrated PV Cell. Monocrystalline Solar Panels (Mono-SI) have an efficiency of approximately 20%, optimized for the commercial use. Polycrystalline Solar panels (p-SI) and Thin-Films have a lower efficiency, respectively around 15% and 8%, but less expensive and durable then the p-SI. Concentrated PV Cells are very efficient (40%) but very expensive. They required a sun tracker device to reach a high efficiency rate.

How to orient a Solar Panel?

An important factor is represented by the solar panels orientation towards the sun. An inadequate orientation can compromise the energy conversion capabilities of a PV panel. In the simplest way, the PV panel can be fixed to a certain angle according to the site where it is installed. As an alternative, solar panels can rotate, through different axis, in order to follow the Sun’s movement throughout the day, keeping the angle between the sunrays and the normal PV surface as minimum as possible. Such device is known as sun tracker. There are two groups of sun trackers which are the single-axis and dual-axis trackers (Assaly, 2012). Single-Axis sun tracker can pan or tilt, since it can rotate through only one axis on a horizontal or vertical pivot, adjusting according to the azimuth (θ) or to the solar elevation angle (α). The 2-axis sun tracker can tilt and pan at the same time so that any location in the upward hemisphere can be pointed.

Photovoltaic Panel orientation typologies 1. Fixed angle 2. Single axis Sun tracker    3. Two-axis Sun tracker.


Except for the fixed panel, the movement of Sun trackers are usually defined by computers in different ways, usually controlled by a computer or microcontroller. The simplest way is to provide the system of several light dependent resistors (LDR) detecting the difference of sunlight hitting each sensor. The system does not use any information about the position of the sun, the day and time, the geographical position but it only senses the amount of light through several sensors with a BOTTOM-UP approach. On the other side a most sophisticated system can orient the panel by using the sun data and other important information such as the geographical position through a digital compass, day and time through a digital clock connected to a microcontroller (if a microprocessor is not available) with a TOP-DOWN approach. Unfortunately, the last system, although more accurate, cannot detect if the sky is overcast or a shadow from a nearby object is casted onto the panel, unless a light sensor is integrated in the system.

Photovoltaic Panel integrated into a basic 2-axis Sun-tracker. Panel’s characteristics: 60mm*90mm*2mm, 6V, 100mA.   



Solar energy is available in the form of Electromagnetic Radiation. Solar rays are assumed parallel and uniform. Due to reflection, scattering and absorption by gases and aerosol in the atmosphere, only a part of solar radiation reaches the Earth soil. The amount of light energy from the solar disk hitting a square meter of terrestrial soil is known as solar irradiance [W/m2]. The total radiation (IT), which is the main contributor of energy on the Earth (Garner, 2017) is estimated as a sum of 3 values: the direct radiation flux (ID), diffuse-scattered radiation flux (IDS) and reflected radiation flux on the surface (IDR) (Mlynski, 2016).

IT = ID + IDS + IDR                                                        (3)


where IT  -  Total radiation                           [ W/m2]

ID — direct radiation flux                               [ W/m2]

IDS – diffuse scattered radiation flux           [ W/m2]

IDR  – reflected radiation flux                        [ W/m2]


On the Earth, the apparent position can be calculated by two main factors: the azimuth, and the angle of solar elevation. The azimuth is the local angle (θ) between the direction of due North and that of the perpendicular projection of the Sun down onto the horizon line (measured clockwise), while the angle of solar elevation (α) represents the angle between the direction of the geometric centre of the Sun and the horizon.

Azimuth and solar elevation angle.


Apart from the site and local factors, other variables which can interfere with the correct solar energy collection are the geographical position, elevation, the date and the time. In addition, to make a system self-sufficient by collecting solar radiation and convert it into electrical energy, it is necessary to understand the environment where the tests will be taken since buildings, trees or other entities cast shadows which can decrease the energy conversion by solar panels. To obtain the amount of solar radiation received on the Earth at standard conditions, it is necessary to know the geographical position (51°32’47” N,0°1’22” W), the inclination of our solar panels (it can be at a fixed angle of a sun tracker), the days (December, 21st, March, 21st June 21st ). The following Irradiance/time graphs provides the amount of solar radiation hitting one square meter of sun-tracking surface under two values: Global real sky and Global clear sky. Since the global clear sky, as a cloudless sky at ground level, corresponds to an optimal condition, values from the Global real sky have been considered for the further tests.

Graph Irradiance/time on March 21st received by a 2-axis sun tracker.  51°32’47’North, 0°1’22”West. (Data Source: PVGIS).

Graph Irradiance/time on June 21st received by a 2-axis sun tracker.  51°32’47’North, 0°1’22”West. (Data Source: PVGIS).

Graph Irradiance/time on December 21st received by a 2-axis sun tracker.  51°32’47’North, 0°1’22”West. (Data Source: PVGIS).



Since an energy balance analysis require specific and certain data to analyse, such as the exact electrical energy requested from a certain actuator, there could be different scenarios according to the outputs used. The following chart summarises the most common outputs used in arts and robotics.

Scheme of possible channels for certain outputs.



In order to convert a decent amount of solar radiation, the size of a PV panel should be fair enough for actuations. In addition, the structure or the robot should be robust to accommodate the solar panel. For this reason, and for the relative costs of actuators, a pneumatic system has been considered for the following experiments. Compressed air will be used to inflate and deflate Pneumatic Artificial Muscles (PAM), biologically inspired complaint actuators using the property of a sleeve mesh which is contracting when a body (a balloon or a silicon hose) is inflating inside of it. Air Muscles have been widely used in robotics to mainly actuate walking machines or movements imitating human and animal motion with a mechanism optimized by the evolutionary process over the last millenniums. (Dillmann, Albiez, Gaßmann, Kerscher, Zöllner, 2007). The amount of air flowing inward and outward of the muscles can be regulated by the use of solenoids or servomotors coupled with air flow valve. More complex and precise mechanism can be used for controlling the air-flow, but very expensive and far from the purpose of this thesis. Each PAM can be inflated by opening a solenoid or a servomotor for a certain interval of time, depending on the size of muscle and the available pressure in the air reservoir. The higher is the pressure, the less time a muscle needs to be inflated. Once a muscle is contracted, the other solenoid or servo can be activated to deflate the PAM and release the exhaust air into the environment. The solenoids are more rapid and responsive, and usually use 12V voltage, whilst the servomotors are slower and use 5V voltage, affording a more dynamic control on the air flow.

Pneumatic Artificial Muscles and possible air control systems: solenoids and servomotors coupled with air flow valves.



Solenoids can be controlled with microcontroller by using mosfets or relays. Solenoids can be programmed to completely open the valve or close. The opening of a valve cannot happen gradually. Even if it is possible to regulate the time to switch on-off a solenoid, the muscle inflates rapidly and intermittently.

Devices scheme using solenoids actuated PAMs



Servomotor is a rotary actuator allowing precise rotation. It consists mainly of a motor coupled to a sensor for position feedback. This type of actuator can be continuous or have a fixed angle, usually ranging from 0 to 180 degrees. It can be attached to a manual air flow valve through gears in order to progressively adjust the valve’s opening. The result is a smooth inflation of the artificial muscles which can provide a wider variety of movements in comparison to the use of solenoids.

Servomotors used for airflow controlling, HITEC HS-422

Servomotors used for airflow controlling


In comparison to solenoids, servomotors do not need mosfets or relays to operate. It is sufficient to connect their signal wires to the microcontroller pins and provide a continuous 5 Voltage. Alike the solenoids, each artificial muscle can be activated using a couple of servos, for inflating and deflating. When using 12DC (i.e. from SLA battery), it is necessary to use a 12V to 5V converter to feed the servomotors.

Devices scheme using servo actuated PAMs



Artificial muscles require a constant availability of compressed air, which is usually provided by using an electric motor coupled with a piston that compressed air, stored in an air reservoir. For small robots or installations, Dc motors can be enough for compress the amount of air needed, but different typologies can be used for this purpose.  Therefore, additional tests have been made analysing the energy consumption of different small air compressors. Three types of pumps have been adopted for the tests, as showed in figure 23. They include: a 12VDC pump AIRPO C2028 (A), a prototype using a 12DC High Torque Motor (40RPM) coupled with a bike pump (B), a 12V digital Tyre Inflator Tacklife ACP1C (C).

Pumps used for analysing energy consumption: A. 12V DC motor pump AIRPO C2028. B. Prototype of a pump using a 12V DC High Torque Motor (40 RPM), bike pump (300mm length, 8mm diameter) C. 12V Digital Tyre Inflator Tacklife ACP1C

Comparative diagram between the air compressors used for the tests.


The three devices run with 12V and mainly differ by size, weight, noise, speed and current. Speed and current represents the two main factors to identify the best solution, in order to achieve a good compromise between energy conservation and available compressed air for the actuations.



Generally, the word behaviour refers to a reaction to a stimulus. A robot/system behaviour can be reactive and deliberative. In reactive robot action/response and perception/stimulus are tightly coupled, without the use of abstract representation of the world (Arkin, 1998). On the other side, a more deliberative robot requires relatively complete knowledge about the world and uses this knowledge to predict the outcome of its actions, optimizing its performance relative to its model of the world. Moreover 3 different approaches can be taken:

  1. Ethologically guided/constrained design.

Inspiration taken from animal behaviour

  1. Situated activity-based design.

A robot’s actions are predicated upon the situations in which it finds itself. Perception problem = recognize the situation and choose an action.

  1. Experimentally driven design (bottom-up).

Endow a robot of a limited set of capabilities, run experiments in the real world, see what works and what does not, debug imperfect behaviour and add new ones through iterations until a satisfactory condition is reached. No simulation technologies are required.


The purpose of the following tests is to calculate the quantity of energy necessary for a single artificial muscle to operate. Therefore, tests concentrate the main outputs to activate the muscles: solenoids, servomotors and air compressor. Each PAM used in the experiment has an approximate length of 350mm and can be actuated by two solenoids or two servomotors regulating the inward and outward air. An air compressor is activated to maintain a constant air pressure around 2.5 bars in a 0.75 litre air reservoir, providing approximately 8 cycles (inflate and deflate one artificial muscle). Energy consumption of solenoids and servos have been tested through three tests each one, then, after getting average values, it has been necessary to calculate the area from the graphs to have an approximate estimation of the energy required for one actuation cycle. For avoiding any interference between the current reading and the actuations during the tests, it has been necessary to split the two functions by using two different Arduino Unos connected by an optosisolator TPL 521. One Arduino UNO is connected can start the actuation through a button. When it is pressed, solenoids or servos start to operate, opening the valve and let the air pass through. At the same time an impulse is sent to the optoisolator, communicating to the second Arduino UNO to start the recording from a current sensor (ADAFRUIT INA219) every 100 milliseconds.


Solenoids used for this test are common normally-closed 12V solenoid valves, connected to the Arduino UNO with two mosfets IRF520. When button is pressed the first solenoid lets the air inflate the muscle, then a second solenoid deflates the muscle by releasing the exhausted air into the environment.

Schematic summarizing the component used for testing the solenoid energy consumption for a cycle inflate/deflate


. Current consumption of solenoids actuated PAMs, average from 3 tests.


From the outcomes, the solenoids are consuming approximately 0.2A when switched on. Calculating the area subtended by the graph, the total amount of energy spent is 0.381 Ampere*second, for a complete cycle (inflate and deflate a PAM once). Converting into Ampere/hour, the solenoids will consume 0,00011 Ah for one actuation. In addition, when the solenoids are not operating, they consume a slight amount of energy (approximately 0.7 mA, or 3.52 mA*s) which corresponds to 0.92% of the energy used by the solenoids when active.

Non-operative solenoids current consumption. Comparison between operative and not operative solenoids  



Analogous tests have been done by replacing the mosfets and solenoids with two servomotors (Hitec- S422) coupled with air flow connector to regulate the air flow. Voltage has to be dropped from 12 Volts to 5 Volts. Two Arduino UNOs has been used as the previous system, connected by the same optoisolator.

Schematic summarizing the component used for testing the servomotors energy consumption for a cycle inflate/deflate


Once the button is pressed one servo opens the valve and let the air come into the air muscle, then it closes. After 4 seconds, the second Servo is opening for deflate the air muscle. During this experiment, a precautionary operation angle has been set avoiding excessive energy consumption. In fact, the servos have an initial angle of θ=10°, while the maximum angle is set at θ=170°, instead of the originally θ=0° and θ=180°.


Servo actuated PAM Graph, average from 3 tests.


The outcomes reveal the discontinuous amount of energy consumed by servo motors while operating, passing from 0.017 A when not in use to maximum peaks of 0,45A. Calculating the area subtended by the graph, the total amount of energy spent is 0.539 Ampere*second (or 0.539 C since Ampere*second= Coulomb), for a complete cycle (inflate and deflate one artificial muscle once). Converting into Ampere/hour, the solenoids will consume 0,00015 Ah for one actuation.

Concerning the use of energy when the servo motors are not operating, a more consistent amount of current has been recorded (approximately 0.017 A, or 0.107 A*s), which corresponds to 20% of the current absorbed by the servo motors when active.

Non-operative Servo current consumption. Comparison between operative and not operative servos.


As a result, solenoids reveal to be more efficient than servo motors. A discrete amount of energy is wasted when servomotors are on but not operating. Therefore, further tests have been carried out trying to reduce the energy consumption in the particular state of “idle”. To achieve this result, a linear voltage regulator LDO (Linear Technology LT1129IT, Low Drop-Out Regulator), which can regulate the output voltage even when the supply is very close to the output voltage, has been attached to each servo. Furthermore, every time a servo is not working, the voltage regulator can drop down the current.

Setup for Servos actuated PAM and Voltage Regulator tests.

 Servo actuated PAM Graph using Voltage Regulator, average from 3 tests

Servo actuated PAM Graph, average from 3 tests


After using the LDO, the amount of energy consumed by servo motors has discreetly decreased. Calculating the area subtended by the graph, the total amount of energy spent is 0.344 Ampere*second, for a complete cycle (inflate and deflate). Converting into Ampere/hour, the servos will consume 0,00010 Ah for one actuation. If we want to convert into Joule, considering that the Voltage is 5V, the amount of energy for each actuation is J = 1,72. Moreover, when servos are not rotating, the current consumption recorded is about 0.7 mA instead of 17mA, according to the previous test (Figure 30 Graph). It represents a crucial improvement since almost the 20% of energy can be saved.

. Comparative graph between the results from actuators tests.


Tests has been useful to identify the most efficient actuator and to understand the energy needed for each actuation. Servo motors combined with voltage regulator represent a suitable solution ensuring reduced energy consumption and an additional dynamic control of the artificial muscles. Values obtained from the examined tests can be used to compare the converted energy from the solar radiation and the possible number of actuations that a robot or artwork can realize.



The tests focused on the amount of energy required to compress air in a 0.75 litre air reservoir with a pressure of 2.5 bars, which can be used to actuate a 350mm length artificial muscle for eight times. From the following graph showed in figure 35, we can compare the energy consumption of the three pumps. The most suitable air compressor can be identified in the sample A, which took 27 seconds for compress air to 2.5 bars by consuming approximately 27.837 Ampere*second. The other solutions are extremely expensive in terms of current (sample C = 89.952 Amp*s) or too slow (sample B = 68 seconds) to fill the air reservoir with the needed amount of compressed air. As a matter of fact, the 12V DC pump (sample A) absorbs 27.837A*s, which converted in Joule is 334.044 J. Since :

Energy = Current * Potential * time                                                (4)

[ Joule = Ampere * Volts * seconds ]

 every time the sample A is active and compressing air, the amount of energy spent is J = 27.837A*s*12V = 334.044 Joule.

Current consumption from 3 different pumps, compressing air from 0 to 2.5 bars.

Comparative graph between the results from air compressor tests.




A basic scenario is set to understand the performances of a robot on March 21st , June 21st , December 21st . According to the irradiance/time graph and the supply, such as a 25W solar panel with 20% efficiency and 0.175 m2 surface, SLA battery 12V 7Ah, 6 air muscles and 12 servos, one 12 DC motor pump and a microcontroller.

Energy consumption from the main output components

Scenario with a solar powered robot, using 6 artificial muscles.


The robot behaviour here considered for the tests it consists of a simple wandering attitude. Therefore, artificial muscles are always working, and the DC Air pump is constantly keeping the pressure at 2.5 bars. In details, an energy consumption of 417 Joule for every 30 seconds has been set. From the outcomes, it is clear that the robot will not succeed to reach the next sunrise. Moreover, during the winter solstice, the robot will be not capable to completely charge its own battery, which needs an amount of 302400 Joules to be completely charged. ( a 12V battery with a 7Ah storage capacity can accumulate a maximum of 12V*7A*3600s =302400 Joule).

Graph Energy/time showing robot’s performance on March 21th

Graph Energy/time showing robot’s performance on June 21th

Graph Energy/time showing robot’s performance on December 21th


A fast prototypes built, according to the model proposed in the scenario, using Rapid-Kinetic toolkit. 




Circadian-Based Behaviour

As for the explored examples in the botanical fields, a behaviour which can simulate a Circadian Rhythms can lead to some improvements in terms of energy conservation. The idea is to connect a digital clock, informing the robot of the sunrise of each day so that it can redistribute the actuations according to the available energy and the sunrise time, when robot could have the possibility to convert and accumulate more energy from sola radiation.

For example, in the graph showed in the following figure, the clock updates the system at 19:00, recalibrating the frequency of air compression every 90 seconds, instead of the initial 30 seconds.

Graph Energy/time showing robot’s performance on March 21st — 22nd. Adding a digital clock as input can help recalibrating and distributing the actuations until the next sunrise.




The research analysed and improved the hardware for a solar powered pneumatic robot, focusing on the balance between the energy converted from a solar panel and then used to actuate pneumatic artificial muscles driven by servo motors and a 12DV air pump. Moreover, simple biologically-inspired behaviour, such as the circadian rhythm, has been considered in the test, providing positive outcomes.




Arkin, R.C. (1998) Behavior-based Robotics. Cambridge, Massachusetts 02142: MIT Press

Assaly, J. A. (2012) Dual Axis Solar Tracker. Faculty of Engineering II, Lebanese University

Clause, J. (1995) The SolArt Global Network ’95: Artworks for the Solar Age, Leonardo, Vol. 28, No. 2, pp. 143-144. Cambridge, Massachusetts 02142: MIT Press

Çengel, Y. and Boles, M. (2015) Thermodynamics. New York: McGraw Hill Education

Cook, P. R., Smallwood, S. (2010) Sustainable Orchestras of Laptops and Analog.Leonardo Music Journal, Vol. 20, Improvisation (2010), pp. 89-95. Cambridge, Massachusetts 02142: MIT Press

Dillmann, R., Albiez, J., Gaßmann, B., Kerscher, T., Zöllner, M. (2007) Biologically Inspired Walking Machines: Design, Control and Perception. Philosophical Transactions: Mathematical, Physical and Engineering Sciences, Vol. 365, No. 1850, Walking Machines, pp. 133-151. Royal Society

Fraas, L. (2014) History of Solar Cell Development. Low Cost Solar Electric Power. Edition No.1, Chapter 1. Springer. University of Washington Seattle, USA

Garner, R. (2017) Solar Irradiance. NASA

Goldring, E. (1987) Desert Sun/Desert Moon, Leonardo, Vol. 20, No. 4, pp. 339-348. Cambridge, Massachusetts 02142: MIT Press

Grätzel, M. (2007) Photovoltaic and Photoelectrochemical Conversion of Solar Energy. Philosophical Transactions: Mathematical, Physical and Engineering Sciences, Vol. 365, No. 1853, Energy for the Future, pp. 993-1005. Royal Society

Mlynski, S., (2016) Sunlight, optimisation, reflection, growth. London, UCL — Bartlett School of Architecture

Nelson, J., Emmott M. (2013) Can solar power deliver? Philosophical Transactions: Mathematical, Physical and Engineering Sciences, Vol. 371, pp. 1-8. Royal Society. Department of Physics, Imperial College London, London, UK

Pfeifer, R., Scheier, C., (1999) Understanding Intelligence. Boston: Massachusetts Institute of Technology

Rifkin, J. and Howard, T. (1985) Entropy. London: Paladin

Scientific American (1997) How do frogs survive winter? Why don’t they freeze to death?

 Song, W., Paik, J. (2015) Sensory-Coevolution in New Media Art. Chung-Ang University. Seoul, Korea

Thompson, R., Mukhopadhyay, T. P. (2016) Microbial nanoids: electronic arts in the face of Mexico’s megadiversity crisis. Department of Art and Enterprise, University of Guanajuato Salamanca, Mexico

Valenzeno, D.P., Pottier, R., Mathis, P., Douglas, R. (1990) Photobiological tecniques, Series A: Life Sciences, Vol. 216, pp 279-282, Springer Science+Business Media, LLC, New York, USA



Submit a Comment