Building a fire-fighting robot
I’ve been looking for a reason to build another robot. A few weeks ago I found a list of robot competitions around the world. Ignoring the battle bot options, geography and capability led to me to the Penn State Abington Fire-Fighting Robot Contest:
The objective of the fire-fighting robot contest is to design a computer-controlled robot to navigate a maze (8 ft. by 8 ft.) that consists of 4 rooms. Rooms are surrounded by walls except for a 18″ entrance. A single candle is randomly placed in one of the 4 rooms. The goal is for the mobile robot to explore the maze, locate the candle, and extinguish the candle in the minimum time. Robots must be within 12″ of candle before extinguishing candle. The layout and dimensions of the maze and rooms are fully known to all contestants prior to the contest. For the advanced divisions, the hallways and room may be covered with carpeting, and there is a small staircase located within the maze. Bonuses are earned for returning to the start position after extinguishing the candle, and allowing obstacles to be placed in the rooms. Participants are permitted to use any combination of building materials and computer technology. All robots must operate autonomously except for the K-5th grade remote control division.
This is not a simple task, but I was particularly excited to see the K-5th grade remote control division. This allows us to participate as a family; my kids can drive the robot to the candle and extinguish it, while my entry must be autonomous. Robots cannot be larger than 12.25″ in any dimension.
At this point, I’ve figured out the basic design. I believe the most difficult task will be navigation, especially as the robot will have limited sensor capability. The robot may start from a known position in the maze, and I will take this option. I would like to use existing software; many brilliant people have worked on this problem. A few months ago I took a great course on the Robot Operating System (ROS), taught at HacDC by Andrew Harris.* ROS has the firepower to do the job, but there is a steep learning curve, it must run on a full PC, and I will have to use Ubuntu and Python, two environments I’m not familiar with. In doing the course assignments, figuring out Ubuntu and Python slowed me up more than anything else. However, I already have a working setup from the class, and with Google all things are possible. So my plan is as follows:
I will run ROS on my laptop. The robot will have an Arduino microcontroller as its brain. The laptop and the robot will communicate wirelessly using a pair of XBees. ROS has a package, rosserial_arduino, which allows the Arduino to run little ROS nodes that can send sensor and odometry data over the serial port to the ROS nodes on the laptop. The laptop nodes will do the heavy computation and reply back with navigation directives. The XBees make the serial port communication wireless.
Although the ROS site has some good tutorials, setting up the XBees on Ubuntu to run the demo Arduino programs was not obvious. Their rosserial_xbee package is designed for a mesh rather than point to point, and the two modes are not compatible. And the XBee configuration tool is Windows only. I got it working — if I have time, I will write up a post on how to do all of this — and was quite impressed with the performance of the XBees. I walked outside my house for a few hundred feet and was still able to communicate with the XBee in the upstairs office. The big risk with this design is that if the wireless communication fails, the robot is dead in the water. My XBee series 1 802.15.4s have a range of 90m and operate at 2.4 GHz, so they should work well. I did of course change the default PAN ID to avoid interference from any other competitors with the same device.
One nice thing about the wireless design, the kids can use the same robot for their entry. I will just hook up a joystick to the PC and it will send commands via ROS. Even if my autonomous design fails, and I give it a 50% chance given the total novelty of all this, theirs should have a better chance.
I purchased 10 ultrasonic sensors (HRC SR04) from China for the price of 1 PING))) — $3 each on eBay, free shipping, and they arrived quickly. Currently waiting for the chassis, gearmotors, wheels, and encoders to arrive. I opted for a mostly circular ABS chassis that is 7″ in diameter. The wheel encoders were more expensive than anything else, so it will be interesting to see how well they work. I already have Arduinos and servo motors from other projects.
In the meantime, there is a lot of software work to be done. The ROS navigation stack is not at all simple to understand, nor are the simulation tools. But I am making progress. I already have a simple robot model URDF file, and understand how to translate frames of reference using tf. I should be able load a 2D map of the maze using a properly dimensioned image, so the next step is to generate that.
The contest is April 21st — not far away!
*The HacDC ROS class wiki is publicly available and is complete enough to learn from on your own.