In this article we present our implementation of intelligent system developed for a LEGO robot. We have built a robot which is using sensors to trace the line and follow it. The model of tracing is based on fuzzy rules, which have been done to follow the shape of a black line on the white background. The system was implemented using C# language for LEGO Mindstorm elements. Results of our experiments show that the robot is able to follow various shapes of black lines. lect such libraries and optimize them for robotic constructions. Also some approaches are based on frameIn many fields of industry and production various ro- works, as proposed in [5] that frameworks are also botics are applied to help in situations where the work possible for other constructions based on arduino elmay be dangerous for humans or there is required high ements, which provide similar possibilities of robotic precision at work. To start the research in such field constructions. All these what we want to create dewe have selected LEGO mindstorm elements. This rob- pends on our ability to develop the construction and ot model is easy to develop just by using simple bricks than to implement the control software. An interestin many shapes from the set. The set also provides ing discussion was given by [6, 7]. a programmable electronics with a variety of sensors. The other aspect of robotic models is to select a proOn the other hand such approach is widely presented per decision support model. In many IoT constructions in other literature, what gave us an inspiration to de- we can find various approaches. In [8] was proposed velop our idea. how to combine neural networks with rules in a form In [1] a model of LEGO blocks was used to work of soft set table. While in [9] the fuzzy rule system with human gestures, where sensors were used to read was implemented with neural networks to work in a gestures and therefore take actions in mindstorm ele- smart house environment. It is also very popular to ments. In [2] was presented how to use such LEGO develop fuzzy rules working on images, as discussed models to help in first contact with machine intelli- in [10]. Mechanical vehicle constructions also very gence. Authors describe good practices when using often benefit from artificial intelligence, both at conLEGO Mindstorm as a platform for programming ar- struction and simulation level, as proposed in [11, 12, tificial intelligence systems. As presented in [3] these 13, 14]. models also support creative thinking, especially when Our approach is using C# language to implement we develop new algorithms that must be programmed control system based on fuzzy rules to first detect and in a special way accepted by the LEGO Mindstorm than follow the line. The LEGO Mindstorm robot is platform. There are various approaches to use pro- using two servo-motors as accelerators of wheels congramming languages in developing such robots. trolled by our developed artificial intelligence to folHowever the most necessary is tu use an optimal low the line. our experiment shows that our construclibrary, which will support all necessary function that tion is well defined and can follow even the complex provide optimal connection and configuration between lines on the board. elements of the robot. In [4] was discussed how to se-
© 2020 Copyright for this paper by its authors. Use permitted under Creative CPWrEooUrckReshdoinpgs IhStpN:/c1e6u1r3-w-0s.o7r3g CCoEmUmoRns WLiceonrsekAsthtriobuptioPnr4o.0cIneteerdnaitniognasl ((CCC EBYU4R.0)-.WS.org)
In this section we will describe how we have developed the model both from programming side and thinking model.
The aim of this project was to build an artificial intelligence system based on fuzzy logic, which would steer the vehicle used in LEGO Line-follower competition. Based on readings, the system must decide how to change the speed of each engine, so that: • The vehicle does not go of the road (A moment when a black line is not between vehicle’s wheels is considered going of the road) = 100 (2) • The vehicle goes as fast as possible −
Where and are accordingly, maximum and
The robot is connected to computer via Bluetooth. minimum reading value of left sensor. and The robot sends raw readings and data to computer are maximum and minimum reading values of right where all necessary calculations are being performed. sensor. These values are obtained during setup conThen, based on calculation results appropriate com- figuration. Reading output is determined using the folmands from the computer are sent to robot. lowing formula:
= |(
−
) ∗ − 100|
(
Firstly, we must normalize the readings from color
sensors to negate the hardware diferences. To do that we
must first calculate parameters and accordingly
as:
=
100
−
(1)
=
−
10
(
To determine grade of membership Bell shaped and Linear functions are used. The formulas are as follows:
{ =
1
1+| − |2
0
{
, ⩽ ⩽
, <
∨ >
= 01 ,, <⩾00 (
Then, the resultant vehicle speed composed from = separate engines speeds is being calculated. In order to + + determine it we calculate each engine’s grade of mem- where = 10, = 50 and = 90 are coeficients of bership for all of speed membership functions and ac- speed rules. cordingly to fuzzy rules base shown in Tab. 1 we conduct necessary calculations.
The results are being grouped by the rules they correspond with and for each of rule we choose the biggest value. Lets assume that the biggest value for rule slow is , for medium - and for fast - . The final resultant speed is obtained by using center of gravity method defined as:
(
The next step is to calculate the inclination, which con- 4 sists of resultant speed and resultant reading values. where is the inclination value calculated before. In order to determine it we calculate resultant speed In case of Easy Left and Easy Right if the value is grade of membership for all of speed membership func- lesser than -50 or greater than 50, then it is set to 50. tions and resultant reading grade of membership for all of reading membership functions. Then, accordingly to fuzzy rules base shown in Tab. 2 we conduct 4. Implementation necessary calculations. The results are being grouped by the rules they correspond with and for each of rule Let us now present the software we have done. we choose the biggest value. Lets assume that the biggest value for rule HL is 1, for EL - 2, for S - 3, for ER 4.1. Application - 4 and for HR - 5. The final inclination is obtained by using center of gravity method defined as: ℎ = (19) = 1 ∗ 1 + 2 ∗ 2 + 3 ∗ 3 + 4 ∗ 4 + 5 ∗ 5
1 + 2 + 3 + 4 + 5
arewchoeerficeie n1ts=o−f 1in0c0l,in2at=io−n5r0u,le3s a=cc0o,rd4in=g5ly0,fo5r H=L1,00 4.2. Program code
EL, S, ER and HR. The program is written in #. In addition to the
stanof Fmineamllbye,rtshheipruvlaeluweitihs tchheosgernea.tDesetpienncdliinnagtioonn igt,rathdee ldiabrrdar.y for cloimbrmaruiensi,cathtieonprboegtwraemenutshees ctohme p.uter and3
engines speeds are calculated and set. For Hard Left the robot. The program has been divided into classes
(
Based on the mathematical model described above, we have created a robot control application. The application has a very simple graphical user interface, which makes it user-friendly. In Fig. 5 and Fig. 6 application window can be seen. ℎ ℎ ℎ ℎ 4 = = = =
Reading method This method is responsible for the normalization of color sensor data. This method also serve as a precaution against loosing the route by a robot. Method code can be seen in Fig. 7. (12) HowTheRouteRuns method This method is a frag(13) ment of the fuzzy system. It combines the route incli
nation with the current vehicle speed and decides how (14) to react based on that data. A piece of the method code (15) is in Fig. 8.
Robot class This is the class which object repre(17) sents vehicle instances in the program. The most important class fields and properties representing the state of the object can be seen in Fig. 9. class object, which belongs to the . 3 library represents the LEGO EV3 brick. property represents the state of connection between the program and the robot. Fields and ℎ store values obtained from color sensors. Fields _ and _ ℎ store current motors speeds. The method is responsible for choosing the direction in which the robot should go. The code for this method is shown in Fig. 11.
Go method This method belongs to the class. TurnHard method This method also belongs to the The method is responsible for the robot’s movement. class. It is responsible for updating the variables It contains loop in which the fuzzy system makes cal- representing the current motors power. The full code culations, decides about the next move and finally ma- of the method can be seen in Fig. 12. The class ckoedsethoafttmheovmeebthyosdetctianngbtehefopuonwdeirnoFfimg.o1t0o.rs. The full aolpseorahtaiosn s are analoganouds tℎo the methomdsetwhohdo.se ChooseTurn method This method belongs to the class. It is called in the loop described above.
In order to test how the vehicle performs we created a test route (Fig. 13). The route is a 19-millimeter-wide black line on a white background. It is quite complicated due to a lot of 90◦ angle turns, it does not contain crossroads though. The vehicle runs through the route precisely but with little speed. It is worth mentioning that the system is incredibly sensitive to any changes in lightning and the non-uniformity of it. A part of a test ride can be seen in Fig. 14.
Because of poor functioning of some of . 3 library functions that our project was based on, the results were not as good as we had imagined. Commands must have been initialized for a specified time amount which led to delays and unstable movement of vehicle. What is more, in order to work properly the program had to use function (), which stops the robot suddenly for few milliseconds.
Considering that . 3 library has not been updated for 7 years and it’s author oficially abandoned the project, rewriting the whole project to language seems to be the best way to improve program’s performance, allowing the program to run on EV3 brick itself. Apart from solving problems mentioned above, that approach would also terminate problems connected with Bluetooth connection latency. rule-based systems, Journal of Artificial Intelligence and Soft Computing Research 10 (2020) 271–285. [12] M. Woźniak, D. Połap, Hybrid neuro-heuristic methodology for simulation and control of dynamic systems over time interval, Neural Networks 93 (2017) 45–56. [13] G. Capizzi, F. Bonanno, C. Napoli, Hybrid neural networks architectures for soc and voltage prediction of new generation batteries storage, 2011, pp. 341–344. [14] F. Bonanno, G. Capizzi, C. Napoli, Some remarks on the application of rnn and prnn for the chargedischarge simulation of advanced lithium-ions battery energy storage, 2012, pp. 941–945.