Trinity Autonomous Firefighting Robot Contest

Transcription

1 Trinity Autonomous Firefighting Robot Contest University of Connecticut Senior Design Team Members: Katherine Drogalis, Electrical Engineering Zachariah Sutton, Electrical Engineering Chutian Zhang, Engineering Physics Faculty Advisor: Professor John Ayers

2 Abstract The aim of this project is to create a fully autonomous robot that can navigate a model home in search of a fire, in the form of a burning candle, and then extinguish it. This project is in conjunction with the Trinity International Robot Contest which takes place April and is a not for profit event that promotes innovation and creativity in the STEM field. Ideally, the principles we use in the design could be extended to a more robust system that could be used to combat actual fires in residential or commercial settings. Background & Rules The main requirement of this project is to create a robot that is fully autonomous. By that, it means that once the robot is started by the user, it navigates, searches, and extinguishes the fire on its own, with no assistance or input from the user. The judge will place the robot in the maze and press the start button. The robot will then listen for the sound (a certain frequency) signaling to begin. Each robot must meet these requirements in order to compete in the competition. There are also strict rules about each robot having a carrying handle, a start button, a flame detect LED, a mic, and a kill power plug. Finally, the robot must also accomplish the goal in the allotted time, which varies depending on the maze level. The arena that our robot must navigate is an 8*8 foot plywood square partitioned by walls to mimic rooms and hallways. The robot will start at any arbitrary position in the maze and when prompted, will navigate the maze to find and extinguish the candle. The flame it must search for will be in the form of a candle placed in a random room in the arena. Our robot must find and extinguish the candle within a set amount of time. Quicker execution times will be given higher scores. There are three levels of competition with increasing levels of complexity and difficulty. Our robot must successfully complete each level before advancing to the next. We have five trials total to get through as many levels or get as short of a trial time as possible. A robot doesn t need to complete all three levels in order to place in the contest. We are planning to compete only in the first two levels since the third involves finding, lifting and carrying an object which will require additional systems. Since time is our main project limitation, we don t want to increase the scope of our project if it isn t absolutely necessary. In level one, there is a single arena with a set layout as shown in Figure 1. The walls of the rooms and hallways are smooth and painted white. We know the hallway and room locations to a reasonable degree of accuracy. Our robot must finish this course within 3 minutes of being started. In level 2, there are four possible arena layouts (Configuration C is shown in Figure 2), and our robot has 4 minutes to complete the trial. Level 2 will also incorporate some furnishings and decorations to make the arena more realistic. There will be rugs placed in some or all of the rooms and hallways, as shown in Figure 2. Wall decorations and wallpaper will also be used in this level, such as tapestries, mirrors, pictures, and paper of all different patterns and colors. Mirrors will not be used in the room where the candle is located. The contest is also separated into two categories, customized robots and unique robots. Customized robots are built from a kit while unique robots are designed and created by the user. Whether or not a robot is unique is determined by the contest judges. Our robot will be unique since we are ordering individual parts and designing the entire system.

3 Figure 1. Level 1 Arena With Dimensions. Figure 2. Level 2 Potential Rug Locations. Solution Physical Design The dimensional requirements for our robot are that it must not exceed 31*31*27 centimeters (Length*Width*Height). The material we chose for the platforms of the robot is polycarbonate "Lexan" sheet plastic due to its electrical insulating property, strength and resistance to cracking. It is also relatively cheap and easy to cut and fasten. The main shape of the platform are designed to be round in order to utilize the limited space. There are three main levels and a fourth sub level of platform. The bottom level is where the wheels and their driving motors are mounted. There will also be our power supply (rechargeable batteries) at the back of that level. At the front of the level, there will be a circuit for controlling the drive motors. The second level will have the main range sensor (laser, discussed later) in the front. We discussed placing this sensor high on the robot, but decided to set it lower so that it senses obstacles close to the ground. The required calculations for navigation will be simplified if the center of the scanner is directly above the center of the line between the drive wheels. The second level will also have the main microcontroller toward the back. The third level will hold the sensors and devices necessary for flame detection and extinguishing. These must also be positioned toward the front of the robot. However, we want to avoid blocking the view of the scanner as much as possible so supporting the third level will be challenging. The most important sensor arc to keep unobstructed is the 180 degrees in front of the robot. We intend to support the third level entirely from behind the scanner and have an arm that extends out to the front of the robot. A smaller fourth level will be set toward the back of the top of our robot to hold the start button, LED, microphone and kill power plug (contest required devices to be discussed later). We will also have a handle, as required by the contest, for easy maneuvering of our robot. Power Supply

4 We decided to use rechargeable batteries for the power supply to reduce expenses. Two packs of the batteries will be needed to prevent the situation (if we only have one pack) where we have to stop our test on the robot and recharge the battery. We did some rough analysis on how much total power is needed and decided that we want the battery to be about 5000 mah and at least 12 Volts. We decided to have two pack of rechargeable Li Polymer batteries with 14.8 Volts and 5500 ma. The one we chose is comparably economical they other products. We are planning to build a power distributing circuit so that we only need one power supply for the entire robot. We can regulate the battery voltage down to voltages that are useful for different devices on the robot. Motion Control We are making a differentially steered robot. This means that there will be 2 independently driven wheels and 1 or more free steering castors for support. The angular velocity of the robot is then a function of the difference between the rotational velocities of the two drive wheels. We decided on this control scheme because it is simple to implement and allows for maximum steering ability. If necessary, the robot will be able to rotate around its own center. There will be two separate motors driving two wheels independently. The separate motors enable our wheels turning different angles at the same time. We intend to implement odometry by counting wheel rotations. This will give us an estimate of where the robot has been it will also serve as feedback to tell us if a control command has done what it was intended to do. There are a couple of different motor types that can accomplish this such as stepper motors or DC motors. Stepper motors rotate a specific angle for each input command, so wheel rotations can be counted by counting the input commands. Another way to count wheel rotations is to use an encoder on a DC motor. The encoder gives a certain amount of counts per shaft rotation so this setup has essentially the same result on odometry as the stepper motors. We ordered a pair of 12V DC motor with encoders. We decided on DC motors because they are easy to control with a pulse width modulation (PWM) signal. Also, stepper motors require a lot of current to run accurately which is an important consideration since the robot is battery powered. [1] We chose a 12V motor because that voltage is safe, accessible to us as students and 12V battery packs are readily available. The DC motors have gearboxes which step down the motor shaft speed which also increases the torque on the output shaft. The gear ratio is 100:1 and the maximum output shaft speed is 100 rpm. The encoders give 64 counts per rotor revolution which translates to 6400 counts per output shaft revolution. We ordered two pairs of wheels, one is a pair of 9*1 centimeter wheels and the other is 7.62*1.9 centimeters. The first pair of wheels has hard tires, which is good for our step counting system because there will be less error due to distortion. However, the hard wheel tire might cause loose connections for our robot because it cannot cancel much vibration. In comparison, the second pair of wheels has soft tires, which is not good for our step counting system, but good for cancel vibrations. On the other hand, the radius of the first pair of wheels is larger (9 cm) comparing to the second pair s radius (7.62 cm). The bigger wheels provide a higher top speed for the robot, but will require more torque from our drive motors for acceleration. Sensors

5 There are a few separate sensing capabilities that our robot must have. First, it must find its way around. This includes navigating hallways and rooms and avoiding obstacles. It must also be capable of finding a candle flame. These sensing capabilities must work in unison so that the navigation system guides the robot until the flame is detected and the flame sensing system can then guide the robot to the flame. There are operating environment considerations in choosing sensors. The arenas which the contest will take place in are made of plywood. The floors will be relatively smooth. The partition walls will be vertical and all corner joints will be roughly 90 degrees. The hallways have roughly uniform widths and the walls will be painted a light color. However, we must allow for some variations in all of these environmental properties. This means that we can t pre program our robot with exact arena layouts and just use our wheel encoders to navigate via dead reckoning. We need some range measurement capabilities to use as feedback to compliment our dead reckoning navigation. There will also be different types of external interference that can affect sensor operation. The contest takes place in a large gymnasium with overhead lighting. There will also be a lot of flash photography happening during the contest. Both of these light sources will interfere with any range sensors that use light for measurement, especially noncoherent light sensors like infrared (IR) range sensors. These interferences could also confuse our flame detection sensors if we rely solely on the presence of light to indicate the presence of a flame. The main types of range sensors which are readily available are ultrasonic, infrared, and laser rangefinders. Ultrasonic range sensors are very inexpensive and are very simple to collect data from. The drawback is that they have low resolution meaning the sensing is done over some angular range (~30 degrees) and the sensor simply gives a distance measurement for anything in that angular range. This is probably acceptable for obstacle avoidance. But if we want to map the locations of walls and objects in the operating environment, it will require something with better resolution. IR rangefinders have similar resolution characteristics to ultrasonic sensors. The maximum distance measurement is also limited to ~2 m on readily available models. IR range sensors are also the type that is most likely to receive interference from the operating environment. Laser rangefinders are the best option for high resolution sensing. This is because laser light is highly directional so practically pinpoint range measurements can be taken over considerable distances. Laser light is also monochromatic and coherent which will help a laser rangefinder reject light from interference sources. The drawback is that these sensors are by far the most expensive of the three types. We decided on a infrared laser scanner made by RoboPeak that can measure a range of 6m, which is farther than any IR or ultrasonic sensor we could find. It continuously scans 360 degrees taking a range measurement every ~1 degree. It has an angular encoder to determine the exact angle relative to the robot s heading of each measurement. It returns range measurements in polar coordinates. Our intent is to use this range capability and the inherent resolution of the laser to see openings and corners before our device is at them, allowing for early decision making and smooth operation. We also discussed the possibility of implementing something similar with a stationary range finder and a rotating mechanism of our design. This would cost less, but require time to build and program. The advantage of this scanner is that its operation is already programmed and controlled and all we have to do is process its output. Laser rangefinders in general may be overkill for the particular environment we'll be competing in (evenly spaced, light colored walls). But our intent is to design something that could be extended to a less ideal environment where this type of measurement would be absolutely necessary.

6 For flame detection, we intend to use a combination of light and heat sensing capabilities. A setup that only senses light will be subject to a lot of interference. A setup that looks only for heat may be limited by the relatively small amount of heat put off by a candle. However, we think it is safe to assume that the candle will be the only significantly hot light source in the operating environment. So if we sense both light and heat it should indicate our target. For light sensing we will test different arrangements of photodiodes and photoresistors which sense the grayscale of ambient light. We will attempt to shade out light from directions other than the direction we expect to see the candle in. For heat sensing we ordered a 4 x 16 thermopile array. This is essentially a 64 pixel camera that assigns a heat signature to each pixel instead of a color so we should be able to locate the candle once it s in the field of vision. Experimental testing with this sensor has shown that it can detect a flame from a distance of approximately 1.5 meters, which is far better than we were expecting. Routing/Navigation For navigating the arena environment, we plan on implementing a Simultaneous Localization and Mapping (SLAM) algorithm. There are many online resources that describe possible implementations [2]. The main idea is to use features that can be sensed in the operating environment as reference points to check the estimate of the robot s motion and to create a map of the area. Since we are using a laser scanner it should be relatively simple to detect corners and terminal points in walls. We can use these points as our landmarks. SLAM involves identifying landmarks in the operating environment. The robot s motion is predicted from the control input and the resulting odometry. This prediction is then compared to the results from the scanner measurements. It is an adaptive algorithm that continually reassigns trustworthiness weights to the sensor and odometry readings depending on the previous results. This lets the robot know which information to regard as more accurate in calculating its position at any given moment. It is important to have such an ability since the odometry from wheel counting, while accurate over short distances, can suffer from small compounding errors such as wheel slippage and slight errors in wheel radius. The implementation of this algorithm will require data processing and calculation. We are currently able to gather data with the Arduino from the laser scanner. These measurements are in polar coordinates (r,θ). Data in this form may prove to be useful in looking for openings in walls since a drastic change in r between consecutive measurements will usually indicate an opening or terminal point in a wall. However, the calculations for localization and the map of the environment will be done in cartesian coordinates (x,y). This will require transformation of our sensor data which should be relatively simple mathematically. We plan to do all of the necessary calculation with the Arduino. However, it is unknown at this point how well the Arduino will handle all of the sensor data processing, the control calculation, and the SLAM calculations and memory. In the case that the Arduino runs slowly doing all of the required processing, one of the group members has a Raspberry PI microprocessor they are willing to donate to the project. If necessary, we will establish communication between the Arduino and Pi and use the Pi to do a lot of the mathematical calculations for SLAM. This will decrease the processing load on the Arduino and it will act mainly as a microcontroller while the Pi will act as more of a processor.

7 Ultimately, from SLAM we will have a map of the robot s surroundings and the robot s current location in that map. We intend to store some previous locations in memory. This will allow us to know where in the map the robot has been. This will aid in efficient searching. It will also enable the robot to return to its starting position which gives bonus points in the competition. Memory may become an issue as well. As mentioned above, we want to remember some of the robot s previous positions. Also, the map of the area will be stored in memory. If the required memory exceeds the available flash memory on the Arduino, we will need to figure out a way to add memory. Once a decent map is generated and stored in the memory, we can apply some logic in the programming for searching that map. Because of the range of the laser scanner, it is reasonable to assume that we will have a map of the area well ahead of the robot s location. This will allow for logical processing and decision making to occur while the robot is moving. So the robot should never have to stop at a corner or intersection to decide which way to go. This should greatly reduce the time it takes for the robot to accomplish its goal which is the main factor in determining its score in the contest. Flame extinguishing There are several ways we discussed on how to extinguish the flame. The first and most preferable solution is to use the compressed gas. There will be a bonus added to our score if we use compressed gas and it is effective because the compressed air will blow with a very high speed. We planned to put the compressed gas can at the back of our robot and connect to a nozzle which will be mounted at the front of our robot about 17 to 18 cm high. The trickiest part of this operation is aiming the compressed gas. Since the nozzle will be relatively small diameter (~0.25 inch) it will need to be pointed directly at the candle flame to be effective. We are hoping that the thermopile array will aid in this process. If set up properly, there should be a particular pixel in the array that is aimed at the nozzle tip. We can then manipulate the robot s position until the flame is in this pixel. Once the robot is aimed at the candle, it will release the compressed gas and extinguish the flame. The other challenging aspect is how to release the gas. We have a frame that holds a CO2 cartridge and has a spring loaded release valve. However, to release the gas, a force must be applied to the valve that is strong enough to overcome the spring and the pressure of the gas (~1000 psi). This is a lot of force so will require some type of lever to open the valve. There is a chance that we will be able to find an affordable pressure regulator that can step down the pressure to something that is easier to handle. The other option is using a fan. It is much simpler than the compressed gas. We will connect the fan to the Arduino and we program the Arduino so that the fan will be turned on when the robot detects the flame is in front of it. We think this is a safe fallback option so we plan to try a compressed gas setup first. If we run into difficulties with that, we will go with a fan. Start (start button, LEDs, microphone, kill power plug, handle) The contest requires each robot to have 5 key elements, a start button, LED, microphone, kill power plug, and a handle. The start button is required to be on a green background and is used to initialize the robot. The LED must be red over a white background, and is used to indicate flame detection. The microphone needs to be on a blue background and when it detects a sound of a specific frequency and amplitude it activates the robot. The kill power plug is used to immediately remove power from the robot s sensor and control system and its drive system (when plug is removed, all robot systems are

8 turned off) in case of emergency and needs to be on a bright yellow background. All of these devices, as well as a handle used to safely lift and move the robot will be placed at the very top of the robot. Programming We chose Arduino as our microcontroller, and that means we will use C++ as the language of our programming. There will be a learning curve with the programming since no one on the team has much of a background in that area. Timetable The contest is the weekend of April 1st, This gives us roughly 5 months to complete the project. Our current goal is to have the rough robot constructed by the end of Fall 2015 semester and have the test arena constructed before the start of Spring 2016 semester. We think that programming the robot s behavior will be the most time consuming aspect of the project, so we would like to allow the majority of the spring semester for testing and programming. Budget Our sponsor is the University Of Connecticut and the initial budget is roughly $1000. As of November we have spent $875 on buying parts and materials for our robot. We plan to use the remaining part of our budget for building our test arena, creating our extinguishing mechanism, and saving some in case of emergency to replace or fix any issues. Quantity Item Amount 1 Arduino Mega 2560 Microcontroller Rev Pololu 12V, 100:1 Gear Motor w/encoder Pololu 37D mm Metal Gearmotor Bracket (Pair) Pololu Universal Aluminum 6mm Mounting Hub (4 40) Pololu Wheel 90 x 10mm Black (Pair) Pololu Ball Caster with 3/4" Metal Ball Lynxmotion Neoprene Foam Tire NFT 07 3" x 0.75" (Pair) Sharp GP2Y0A41SK0F IR Range Sensor 4 to 30cm RoBoard RM G212 16X4 Thermal Array Sensor RPLIDAR 360 Laser Scanner 398.9

9 1 1 1 Polycarbonate (PC) Sheet, Transparent Clear, Standard Tolerance, ASTM D 3935,1/4" Thickness, 24" Width, 48" Length Pcs 14.8V FLOUREON 5500mAh 35C 4S Lipo High Power Battery RC Battey Pack with XT60 Connector for RC Airplane, RC Helicopter, RC Car/Truck, RC Boat TLP 2000 Tenergy Universal Smart Charger for Li Ion/Polymer battery Pack (3.7V 14.8V 1 4 cells) Total Information about personal and collaborators Sponsor: University of Connecticut School of Engineering Project Advisor: Professor John Ayers Team Members: Katherine Drogalis Majoring in Electrical Engineering Zachariah Sutton Majoring in Electrical Engineering Chutian Zhang Majoring in Engineering Physics (concentrating on Electrical Engineering) Citations [1] Stephen Chapman, Electric Machinery Fundamentals, McGraw Hill, 5th edition [2] SLAM course by Cyrill Stachniss