ABU ROBOCON 17 TECHNICAL REPORT TEAM PRATYUNMIS

Transcription

1 ABU ROBOCON 17 TECHNICAL REPORT TEAM PRATYUNMIS BIRLA INSTITUTE OF TECHNOLOGY, RANCHI

2 INTRODUCTION The Asia-Pacific Robot Contest (ABU Robocon) is an Asian Oceanian College Robotic Competition, founded in 2002 by Asia-Pacific Broadcasting Union. In the competition multiple robots compete to complete a task within a set period of time. The contest aims to create friendship among young people with similar interests who will lead their countries in the 21st century, as well as help advance Robotics Culture in the region. The event is broadcasted in many countries through ABU member broadcasters. Each year the competition has a different problem statement to tackle, but generally speaking two or more robots must be used to complete the tasks. One of the robots is manually controlled while the other is automatic. The best robots usually weight more than 10kg and span in one square meter area. To build the robots, contestants, who are restricted to be undergraduate students, must possess rich knowledge in programming, mechanical design and electronic circuit design 1

3 This Year s Theme: Asobi: The Landing Disc The theme revolves around the word asobi (play), which is also a fundamental philosophy behind Robocon. In asobi, playful, unique, original show of skills is often more important than winning or losing, as everyone friend and foe alike can applaud and enjoy them. The two teams(red and blue) needed to build a robot capable of Throwing a Frisbee so that it can land on the poles of different height and at different distances having Platforms. In total, 107 teams participated in the event. The participation included teams from IITs, NITs and other local colleges of Maharashtra and other parts of the country. The national round of the event was organized at Shree Shiv Chhatrapati Sports Complex, Balewadi, Pune and was hosted by MIT, Pune in association with ROBOCON India. The event was also telecasted live on DD SPORTS as ABU ROBOCON organizes the competition in association with DOORDARSHAN. 2

4 ROBOCON 2017 Game Field 3

5 CONTENTS The contents of this technical report can be summarized as Our Robot Image Processing ROBOT MECHANICAL ELECTRONICS IMAGE PROCESSING 4

6 MECHANICAL REPORT The bot comprises of mainly 3 parts, the base, the Frisbee throwing mechanism and the angle change mechanism. Base According to problem statement of ROBOCON 2017 the locomotion drive of bot should be as mobile as possible and the Holonomic drive clearly surpassed the need of Differential Drive. Differential drives - A differential wheeled robot is a mobile robot whose movement is based on two separately driven wheels placed on either side of the robot body. It can thus change its direction by varying the relative rate of rotation of its wheels and hence does not require an additional steering motion 5

7 Holonomic drives - A holonomic (or Omni) drive has motion with 3 degrees of freedom, unlike differential drives, which have only 2 degrees of freedom. As a result, a holonomic drive can shift from side to side or strafe diagonally without changing the direction of its wheels. It uses Omni wheels to minimize drag. Two common configurations are the Plus-Holonomic Drive and X-Holonomic Drive, named to describe their axes of motion. 6

8 . COMPONENTS OF THE ROBOT 1. Frame of the base 2. Motors used for the drives 3. Wheels used for the drives Frame of the base- The whole frame was made up of aluminum because of its light weight and high stress bearing quality. Mainly we had used 1 X1 inch square channel and 2 X1 L shaped channel for the manufacturing of the base frame. The whole frame is joined by welding as it reduces the weight of the frame and also didn t produce any stress concentration on the frame which is generally caused due to hole made on frame to join through nuts and bolts. 7

9 The whole base was designed in such way, that it has enough space to mount wheel and other electronics component and also have sufficient space to perform pole climbing process. The whole base is symmetric about the center line, so we didn t have to switched left or right irrespective of the arena. Motors used for the drives The motors used for the drives were very compact and had very high torque. Due to its high torque it can easily bear a load of 40 kg which was the maximum limit of weight in the robocon 17. It also allowed smooth drive of bot along all turns and corners. The basic information about the motor that we used in the drives is as follows: - Weight of the motor-756gm Rpm of the motor-250 Name of the company of the motor- Maxon Torque of the motor-200kgcm 8

10 Wheel used for the drives - We had implemented mecanum wheel for the locomotion of the bot. Mecanum wheels were part of holonomic drive in which rollers were aligned at an angle of 45 degrees. Mecanum wheels consist of 3 DOF (Degree of Freedom) i.e. it has 2 linear motion and one rotation motion. Linear motion helps in the straight and sideways movement without change in the orientation of the bot. For the straight alignment of the wheel a degree power transfer between shafts of motor and shafts on which wheels are mounted was needed, which could be easily done by the use of bevel gears but, it also had a chance of slip between the gears and moreover installation was also difficult. So this method was discarded rather the wheel was turned by degrees about itself. This reduced the difficulty as well as the material cost. 9

11 The basic specifications of the wheel we used are : - Diameter of the wheel-10cm Width of the wheel-4cm Number of rollers in the wheel-9 Name of the company of the wheel-robu.in THROWING MECHANISM The mechanism which was finalized after discarding of 2 or 3 other Mechanisms consists of a wheel connected to high RPM motor and a wooden platform and frisbee is in between wheel and circular curvature wooden wall which keeps the frisbee in track and as soon as the frisbee comes into contact of wheel circumference then due to Frictional force and angular velocity of wheel the circumference of frisbee gets his own angular velocity which drives it to a smooth flying journey. 10

12 Fig: Throwing Prototype LOADING MECHANISM The loading mechanism consisted of a hollow cylinder made up of aluminium sheet where the Frisbee would be loaded and a pneumatic piston of stroke length 62cm under a common base with a slot so that a servo motor attached to the piston would move freely and feed the Frisbee towards the throwing mechanism by pushing the Frisbee with a servo coupler on servo motor during reverse stroke of the piston. 11

13 Angle Change For landing of frisbee to the desired location there was need of changing the orientation of platform which is then achieved by using leadscrew arrangement between platform and structure of bot as shown in figure down. Fig. Angle change 12

14 Electronics Report The Electronics part of the robot has 3 applications on the bot Robot s Drive Frisbee throwing Platform orientation using lead screw Controller interface rendering it manual or line following Overview - The bot was controlled using the xbox controller and the analog stick were used for movement. The xbox button was used to change modes between mecanum drive and line following mode. The L2 AND R2 buttons were used to change the platform angle and the buttons were used to increase and decrease the speed of the motor throwing the Frisbee. 13

15 Robot s drive The robot s drive was designed to be used as a multi purpose drive. 1. Line following mode 2. Manual drive mode Line following mode of the drive was used when we needed the bot to travel in the main area and in accordance with the lines made on the game field from the start zone to the loading zone and it did so in ease. 14

16 Manual drive mode was used so that the operator could change or manipulate the bot s position and orientation at a point where it was needed. Another reason for the manual mode was that the circumstances of the factors affecting the Frisbee were unpredictable and changed according to the environment it was and this required the operator to judge the motion and make changes to the various parameters on the bot which were orientation distance from the pole Speed of the motor Angle of the platform Feed back mechanism on the drive There were individual encoders for each of the wheels on the drive. This gave us feedback in counts and we used them to calculate and match the rpm of all the motors on the drive. 15

17 Line following mechanism The line following mechanism on the bot was used to quiuckly get from the start zone to the loading zone of the game field and also to avoid delay while positioning the bot to throw the Frisbee as close evaluation of the game field tells us that the lines can also be used as a guide or a marker to tell us where to throw the Frisbee from This was done using the Cytron LSA08 sensor. It has 2 modes to be used 1. Analog value mode 2. Digital value mode We used the analog value mode which returns the analog value in the range 0-70 depending on where the line is starting from 0 at the extreme left to the extreme right at 70. PID Was tuned based on this value for the bot to follow the line 16

18 PROBLEMS FACED WITH THE DRIVE We faced numerous problems with the drive and most of them were due to unevenness in feed back so we had to write the code such that the unevenness was eliminated. The problems we faced were 1. Uneven drifting of the bot 2. Rough velocity profile 3. Mapping and thresholding of the controller s analog values HOW WE OVERCAME THEM The uneven drifting of the bot was due to the difference in rpm s of the individual motors even though they were being given a same signal. This drifting was eliminated when we mounted the encoders on each of the wheels individually and calculated the rpm. Once after calculating the rpm we took the average and then made necessary changes in signal to lower or increase the rpm of the bot. The rough velocity profile was made even with the use of acceleration when the bot speed up, we made sure it sped up evenly and not in a jerk and acceleration code was written to increment only at max a certain amount od speed In a loop 17

19 FRISBEE THROWING MECHANISM & Platform angle change The Frisbee throwing mechanism consisted of an e-bike motor which had 3000rpm on 24v. We mounted the ebike motor and coupled it to the wheel which constrained and threw the Frisbee. The rpm of the wheel was calculated from a feedback encoder coupled to its shaft. The rpm was set by the controller and can be changed by using the buttons on the controller. The platform angle change was done using lead screw The pitch of the lead screw was very less So we used a heavy duty 450 rpm Motor to rotate it and the screw was Hinged at one end and then the screw Was extended to be mounted on the Platform so as to change the angle The problem with this was that the load Was on the couple which caused the screw to loosen out. This was solved by hinging it and then placing the motor underneath and due to it the motor had all the load. 18

20 Controller interface We initially used the ps4 controller and the analog sticks returned the values of 0 to 255 in each direction but the problem with ps4 controllers was that they were wired and the wireless module we bought was not able to be paired with the controller. So keeping this in mind we switched to the xbox controller with cam with its own receiver kit which could be directly attached to the usb host on the Arduino. The xbox controller returns values from 0 to in each direction of the analog stick. This gives us a better resolution in the mapping so that we can get the drive to be more sensitive to the control. 19

21 MICROCONTROLLERS Initially we were using the Arduino Mega ADK as we needed 5 interrupt pins for freedback from the various encoders we used 4 encoders in the drive 1 on the Frisbee throwing motor The problem tha tarised was the due to continuous use of the usb port the adk after a certain level of usage was unable to read the data from the usb port so we swiched to the mega and used a usb host shield from REES52 instead which was more durable. Using this gave us an advantage to Use all the pins and also not have To buy and other Arduino Mega Adk Which is very expensive comapritively 20

22 DIGITAL IMAGE PROCESSING The aim of the team is to automate the robot by keeping the track of the Frisbee and feedback the parameters like its final position (the real time coordinates where it landed), its velocity and its acceleration at different instants of time so that by the use of machine learning algorithms the robot could be trained. OVERVIEW The algorithm is developed such that it tracks the Frisbee throughout its flight and gives us a reliable data (its X, Y and Z (depth) coordinate in real time) of at most 5-6 points. We are able to deduce the velocity and acceleration of the Frisbee at these 5-6 points in each of its flight. Also by using ellipse fitting technique we were able to get the angle of the Frisbee, the length of major and minor axis of the blob (ellipse shaped) at the instant it was first detected. 21

23 Tracking Process The tracking process involves four parts: 1. Frisbee Identification 2. Ellipse fitting 3. Motion tracking (upto 5 m) 4. Curve fitting 1. Frisbee Identification The first task is to identify the object we want to track (our Frisbee in this case). For this purpose, the method of frame differencing is used to identify our blob using contour detection method. For this a certain region of interest is set and a threshold range of area is set. Any blob moving away from the Kinect sensor which falls in this area range will be treated as our Frisbee and the command will get transferred to ellipse fitting code. 22

24 In this method certain cases are handled like if a very large moving body (a human) comes in front of the sensor it would not be treated as our blob. Also a blob would be treated as our Frisbee only if it is moving away from the sensor (i.e. its area is decreasing). 23

25 2. Ellipse Fitting The fitting is done for the first frame detected after frame differencing which has occurred in the ROI. Till now even if a non Frisbee object appears in the ROI it was considered as Frisbee. This mainly occurs when a human body appears before the Xbox Kinect sensor. For this we had to do ellipse fitting because the threshold image of the Frisbee was approximately ellipse. The ellipse fitting is done only once the first frame detected. In ellipse fitting, the contour of the threshold image is taken. After that the convex hull of the contour is found. By using contour and convex hull, convexity defects of the contour is found. As from the above threshold image we see there is no convexity defect (almost none), we distinguish between the images between the one's having large convexity defect and the one having almost no defects. If there are defects beyond the threshold defect depth, then the image (first frame) is dropped and not continued for tracking. If it passes then a fitted rotated ellipse is drawn around the contour of the image (frame) which returns us the approximated length of major and minor axis and the angle of rotation of the major axis with the horizontal, with which we can draw the ellipse. 24

26 Another problem arises that any convex contour may pass for ellipse, which would give us wrong ellipse fitting. For this we did area matching. We matched the area of the contour and the area of the ellipse having the dimensions returned after ellipse fitting. If the areas are approximately equal then, the detection is good and the frame proceeds for tracking. 3. Motion Tracking After identification of our required blob the command comes to the tracking part. In this portion the centroid of the blob is calculated in each frame and the velocity of the centroid is calculated in 2D plane using the previous coordinates. Using this velocity and centroid the upper left corner of the next ROI is decided which would become our region of interest for the next frame in which all these processes would get repeated. The size of the next ROI is kept large enough such that our blob will get detected despite any direction it moves. While the tracking process the value of depth (distance of blob from sensor), its x, y and z coordinates in real time are calculated using the mapper function of the Kinectand these values are used to calculate the velocity and acceleration of the Frisbee at different instants of time. 25

27 Some Problems that occurred: There arrived some cases where the blob movement was so fast that it was not detected in some frames. In that case the algorithm will make the ROI in set direction so that it can detect the blob once again. Since to get the depth data we are using the Xbox Kinect sensor there are some restrictions associated with the sensor. 26

28 1. The sensor has a pyramidal field of view with the sensor as the tip of pyramid and the view as an opening pyramid away from it. Because of which it is impossible for the sensor to detect the Frisbee if it is at high altitude at very close distance. This problem was solved by placing the Kinect just below the throwing end of the throwing mechanism so that the Frisbee can get detected as soon as possible. 2. Another problem is noise. Due to a lot of vibration of the system Kinect will also vibrate which will disturb the frame differencing process and tracking will not be possible. 3. Since the range for which Kinect sensor can give reliable data is from 40cm to 5m only. Tracking the Frisbee or calculating its landing position which may be beyond 5m is not feasible with the Kinect sensor. For this condition the method of position prediction using Kalman Filter was used but the equations used in this method were totally dependent on the instantaneous angle which was not possible to calculate in each frame also the addition of this process to our working code made the processing too slow. 27

29 Another method to solve the problem that is used is curve fitting. Since we are able to deduce the acceleration and position at some points (good points), by the use of a curve can be plotted using those good points which can be used to predict the end point of the curve (the landing point of Frisbee). After the tracking for a particular flight is completed (i.e. no more detection of blob is there in that flight) the command returns back to the first time detection. 4. Curve Fitting Curve fitting is a technique of creating a curve or formulating a mathematical formula or a function that will have a best fit to series of data points. The idea behind using this technique is to determine the possible end point of trajectory of the Frisbee. Also this technique was to determine the deviation of the Frisbee from the platform where it was supposed to land. Having this data was useful to calculate the necessary changes that should be made in orientation of bot or the initial push given to Frisbee so that the probability of its landing can be maximized. 28

30 The curve AB was determined by joining all the lines possible by adjacent data points. To determine the deviation from the pole, the top most point of the pole i.e. the centre of the platform was to be determined. This was achieved by limiting the range of sensor between 4.5m to 5.5 m and by finding the topmost point of the largest blob present Advantages: Using the proposed curve it was possible to find the position of Frisbee in the frame of pole. Using this data the deviation from the top most point was determined. Disadvantages: Since the actual motion of the Frisbee was not exactly similar to this curve as it was formed by joining lines formed the deviation calculated was not with hundred percent accuracy. 29

31 OVERALL PROCESS Bot reaches to location Frisbee is now on it s journey Platform at an angle Feeding of Frisbee Driving wheel starts rotating 30

32 FINAL ROBOT 31