Ra t’s Life Competition Entry Thavidu Ranatunga ANU-UCSD EAP Exchange Student Department of Computer Science University of California, San Diego [email protected] Abstract This report presents an entry into the Rat’s Life robotics competition. The competition pits pairs of robots in a maze with the objective of out lasting the other robot. The task is measured by an energy level which may be recharged from sources within in and unknown maze configuration. The robots are equipped with limited vision and distance sensors. We discuss the methods and strategies involved in an entry; covering the obstacles and various aspects that are required including motor control, mapping, and vision and distance-based sensing. 1 Introduction 1.1 Competition The objective of the project is to create a competitive entry into the Rats Life competition running at http://www.ratslife.org The competition is defined in both simulated and real platforms however we will be referring to the simulated platform hence forth. The competition runs matches of two robots in a random maze configuration containing power sources (‘Feeders’), walls, landmarks and open spaces. The robots have a simulated ‘energy level’ which depletes over time and the robots must recharge at power sources in order to stay active. ‘Feeders’ require time to recharge once they have distributed their allocated power. The objective of the competition is to outlast the opposing robot. In the current state of the competition, there are only 6 maze configurations being selected from, however there are intended to be more. 1.2 Robot Description The robot is designed to reflect the e-puck educational robot (www.e-puck.org) and as such has 3 types of sensors: - 1 Camera - 8 Distance Sensors - 1 Accelerometer- 2 Wheel encoders The camera is mounted on the front of the robot, centered near its top. It can only output a resolution of 52x39 and returns an RGB colour format. The distance sensors are IR-based and only have a specified range of 4cm. This represents roughly one tile on a standard map. They are positioned around the circumference of the robot. The wheel encoders measure the turn of the wheels not necessarily the actual movement of the robot. The accelerometer returns the acceleration in the x, y and z axis directions in units of m/s2. The robot also has 2 outputs: 10 LED lights positioned around the circumference of the robot and 2 wheels. The 2 wheels are positioned at the base of the robot such that turning on the spot is possible. 1.3 System Setup We used Webots version 6.0.1 for the simulating the robot and its environment. The original competition ran in Webots 5.8.0. The code for the robot controllers are written in Java 1.6 update 11. 2 Methods & Strategies 2.1 Movement The robot moves based on how much power is given to the wheels. Basic motor control is required to turn and move the robot into various positions. Turning can be achieved with applying an opposite polarity to the power of each wheel. Movement back and forth is done by having the same polarity on the power of each wheel. This is illustrated in Figure 1. The movements of the wheels are measured by the wheel encoders; however it only measures the turn of the wheels as opposed to actual movement of the robot. As such this leads to inconsistencies and is not always able to be a good measure of actual motion. To reduce this in post-movement it is possible to use the accelerometer to slightly enhance this functionality but this returns the acceleration values and may not be fine enough to be very useful. Alternatively it is possible to use the distance sensors pre-movement to test that there aren’t any obstacles in the direction of movement such that for example the robot doesn’t try to push against a wall (which would increase the wheel encoder amount but not actually move the robot). Figure 1. Basic Motor ControlAll sensors are only updated every cycle of the controller’s time step and thus it is not possible to stop moving at an exact measurement of the wheel encoders. However this can be compensated for and the robot can be re-aligned by other techniques such as using the camera for a vision-based measurement. There are black painted crosses at constant intervals on the ground to form tiles. These can be used to aid in measuring actual motion of the robot. On a larger perspective of movement, we consider two ways of moving through the maze: fluid motion and block motion. The difference is illustrated in Figure 2. In block motion the robot only moves in square tiles or blocks, as designated by the black painted crosses on the maze floor. All robots start in the middle of one such square in all maze configurations. This method is considerably easier to create and maintain a map with as it is possible to just store and reference it as an array or table. The drawback of the method is due to the inaccuracies of movement described above. It is difficult to get the robot to move in discrete squares and as such it needs to be compensated for and realigned over time. In fluid motion the robot moves fluidly without any restrictions. This is faster as there is little to no stopping whilst travelling, and movement isn’t restricted to moving horizontally and vertically. However this inherently makes it much harder for mapping and localization, and techniques such as dead reckoning are required to be used. The problems regarding alignment within the tiles no longer exists, however the inaccuracies with the wheel encoders still exist. This can have less impact here than in block motion, as it is more easier compensated for, by having an acceptable region of error. Both methods of motion are viable with different strategies, but we decided to use block motion as it we estimated it would have a better return on interest for the amount of time available for this project. Figure 2. Top: Block Motion, Bottom: Fluid Motion2.2 Distance Sensors The distance sensors on the robot are IR-based and short range, with the specification listing it as being about 4 cm. In terms of actual maze tiles however, the robot is at best able to only detect the 4 walls on sides of