|
Fire Fighter 2003
UB Robotics in the 2003 Trinity College Fire Fighting Robot Challenge
In 2003 we entered the expert division of the Trinity College Fire Fighting Robot Challenge.
The expert division is much harder than other divisions.
In other divisions a robot can be made optimized for the know floor plan.
It can be programmed to do only what it needs to in the least amount of time, but in expert, the floor plan is not known so the robot has to be programmed more generically.
We used a laser range finder and 2d mapping system to ensure that the robot knows about its surroundings and could navigate any house it was put in.
Some details of our robot
Our robots systems were separated into 3 major parts, as can be seen in the pictures as 3 separate PCBs.
Firstly was the motor control board.
This board is responsible for abstracting complicated PID control loops and tracking position for the higher level software.
It inputs the high resolution encoder information from both motors and regulates the speed of them, maintains the robots current x, y, and angle readings and executes move instructions handed down by the main controller.
The top board in the robot is the auxiliary board. This board is responsible for the 5 sharp rangers, 3 photo transistors, UV sensors, sound activation, fan, and the servo to move the fan. This board is the robots main connection to its peripherals.
It continuously polls all of its attached sensors for most up-to-date readings and provides them to the main processor when required. Because of this always ready approach the main processor doesn't have to wait any time to run a sensor to get a reading, it just asks the auxiliary controller for the most recent reading.
Finally is the main controller. This controller equiped with 512K of external memory and laser range finder driving circuitry controls the high level actions of the robot. All the low level functions of the robot are abstracted for this controller by the other boards. It is free to run complicated decision making and map building routines.
A map of the arena is generated as the robot moves around and is stored in ram. Only about 2KB of ram is used for this information, the rest of the ram is unused at this time.
More details on mapping algorithms to follow.
How we did
We competed on April 13th 2003 with 8 other competitors. We placed 5th among them.
Unfortunately the robot did not perform as well as in practice sessions mainly because of obstacles like furniture and carpets. We also turned down the sensitivity of the UV sensors while practicing because they would see the candle from too far away but the candles in the competiton burned with a very small flame so when the robot was right next to them it though it was still too far away.
The carpets were a particularly nasty ridged entranceway mat. The robot had only small problems turning on the carpet, more marjory the provided little traction when the robot brushed a wall or a piece of furniture causing the robot to loose track of its position within the house.
Overall i would have to say we ok with our score. Time ran very short for programming and some bugs remained and surfaced while running the competition. We also ran out of code space for the main processor and could not implement effective furniture avoidance routines so the robot pretty much hit the obstacles all the time.
Our Robot was constructed by Michael Licitra, Stefan Zickler and Dave Francis.
Competition Pictures
Laser Range finder Information
Progress as of April 9th 2003
Here are some pictures of the final state of the robot.
Yeay! (200KB each)
Progress as of Feb 28th 2003
Its getting there. We didn't get too much done recently because of another competition but we are back on the project now. Finally made the hardware so the robot can scan 360 degrees in one shot and got the Sharp distance sensors mounted so we can try to stay straight when moving, going over ramps. Stefan has been trying to optimize the code some and it is coming along. The main controller has about 3500 lines of c++ code in it, so far most of it works.
Will the pictures never end? (300KB each)
Progress as of Feb 1st 2003
At this time we have 2 of the 3 PC boards made, the motor driver and the auxiliary board (sharps, uv tron, fan, etc). The mapping algorithm is coming together very smoothly. The robot can currently explore the whole maze until it knows the position of every wall. Of course if there were candles and it put them both out it would be done, and wouldn't need to continue exploring.
The laser range finder is supposed to scan 360 degrees someday but it is still mounted right on top of a servo for now. The 360 degree scans will speed things up because the robot will have to stop and look around less. Currently the robot stops to make a scan which usually takes 3 or 4 seconds. We don't see much of a point to make it scan while moving because it will probably corrupt alot of data.
Currently the robot relies exclusively on dead reckoning for its position but that will be fixed soon. We will have some Sharp range finders to make sure we don't get to close to anything while moving and will be able to readjust our current position based on new readings.
A few more pictures (300KB each)
Progress as of Jan 9th 2003
I got alot of the mechanical parts built and the circuits are nearly finalized. Software is being developed and is working well.
Some Pictures (300KB each)
Progress as of Dec 7th 2002
At this time we know what we are going to do. We will be preforming 360 degree mapping to detect our surroundings.
Some Pictures
|
