Introducing Big Foot
Introducing Big Foot
by kemsulli
Wed Apr 18, 2007 7:11 pm
Hi all,
I've been lurking on the list for several months while I have gotten my RN up and working. The first goal I had for my RN, dubbed Big Foot, was to compete in the walking division of the Trinity College Home Robot Firefighting Contest: in this contest, the robots know the layout of a small "house", and have to autonmously find a candle and extinguish it. To get my RN ready, I made serval hardware mods. The first was to add a 400 Mhz Gumstix computer running microLinux and a Robostix microcontroller to the front, under the chest plate. Next, I added four Sharp IR range sensor, pointing left, right, front and back. Some aluminum brackets hold the left, right and front sensors, while some hot melt glue and two slots in the backback hold the rear sensor. I also added a 3-axis accelerameter, and 2-axis gyros in the shoulders. Next, I added an analog servo for a neck, and mounted a CMUCam3 for a head. Finally, I added the grippers from Matt Bauer. I am currently using the stock battery. See http://cs.gmu.edu/~ksulliv/robonova from some pics and more hardware details.
The Gumstix is the brains of the operation, performing high-level reasoning and controlling the entire robot. The Gumstix sends individual bytes over serial to the MR-C3024, which then executes the default moves from Hitec (well, not quite default, I tweaked them to compesate for the changed center of gravity, and slowed some down to improve balance). After issuing a command, the Gumstix blocks until the move is completed. The Gumstix also communicates with the CMUCam3 via serial. Communication with the Robostix is via I2C. Code on the MR-C3024 is in RoboBasic, while the Gumstix, Robostix, and CMUCam3 are all programmed in C.
In the Trinity contest, I found that carthwheels were the most effective method of locomotion, both in terms of speed and stability. Big Foot drew large crowds and lots of interest due to its ability to recover from falls and the cartwheels. Of the seven robots in the walking division, Big Foot finished first.
Future work will focus on software, particularly added localization ability (maybe using occupancy grids and a Monte Carlo approach). Also, I want to add path planning and partial order planning for determining move sequences to achieve a given pose.
Keith
I've been lurking on the list for several months while I have gotten my RN up and working. The first goal I had for my RN, dubbed Big Foot, was to compete in the walking division of the Trinity College Home Robot Firefighting Contest: in this contest, the robots know the layout of a small "house", and have to autonmously find a candle and extinguish it. To get my RN ready, I made serval hardware mods. The first was to add a 400 Mhz Gumstix computer running microLinux and a Robostix microcontroller to the front, under the chest plate. Next, I added four Sharp IR range sensor, pointing left, right, front and back. Some aluminum brackets hold the left, right and front sensors, while some hot melt glue and two slots in the backback hold the rear sensor. I also added a 3-axis accelerameter, and 2-axis gyros in the shoulders. Next, I added an analog servo for a neck, and mounted a CMUCam3 for a head. Finally, I added the grippers from Matt Bauer. I am currently using the stock battery. See http://cs.gmu.edu/~ksulliv/robonova from some pics and more hardware details.
The Gumstix is the brains of the operation, performing high-level reasoning and controlling the entire robot. The Gumstix sends individual bytes over serial to the MR-C3024, which then executes the default moves from Hitec (well, not quite default, I tweaked them to compesate for the changed center of gravity, and slowed some down to improve balance). After issuing a command, the Gumstix blocks until the move is completed. The Gumstix also communicates with the CMUCam3 via serial. Communication with the Robostix is via I2C. Code on the MR-C3024 is in RoboBasic, while the Gumstix, Robostix, and CMUCam3 are all programmed in C.
In the Trinity contest, I found that carthwheels were the most effective method of locomotion, both in terms of speed and stability. Big Foot drew large crowds and lots of interest due to its ability to recover from falls and the cartwheels. Of the seven robots in the walking division, Big Foot finished first.
Future work will focus on software, particularly added localization ability (maybe using occupancy grids and a Monte Carlo approach). Also, I want to add path planning and partial order planning for determining move sequences to achieve a given pose.
Keith
by kemsulli
Wed Apr 18, 2007 11:45 pm
i-Bot wrote:mmh tell us more !
remote control of the C-3024 has challenged us all at reasonable speed. Are you using ETX/ERX or making Robobasic code for direct execution ?
What do you want to know? As for remote control, the bot doesn't use any, its fully autonomous.
Keith
by kemsulli
Wed Apr 18, 2007 11:47 pm
limor wrote:locomotion through cartwheels sounds radical and must have been fun to watch. Did you make any particular use of the gyro and accelerometer for the cartwheel locomotion?
Not during the cartwheel; that was simply the default carthwheel from Hitec. After each move completes, the accelerometer is continually checked to determine if Big Foot fell over or not. I'd like to get the gyro working during individual moves to make the more stable. Maybe a summer project.
Keith
by kemsulli
Thu Apr 19, 2007 9:51 pm
Humanoido wrote:Really impressive all the innovative work you did. How can it do cartwheels with that delicate CMUcam board on the top of the head? Also, how did you write routines to communicate back and forth between microcontrollers?
humanoido
I changed the default cartwheel arm position to allow room for the camera. I'll admit, I held my breath the first time! There is enough room for the camera, but I'm having trouble getting right cartwheels to work, although left works fine.
As for communication, a friend and I wrote a Linux driver to handle I2C communication between the Gumstix and Robostix, so now the code can operate on file handles (pointing to say, /dev/robostix). All the gory details of master/slave and send/receive are in the kernel driver. Communication with the MR-C3024 is via standard serial. On the Gumstix, I use standard file I/O, and use ERX and ETX on the MR-C3024. I send a single byte which is then mapped to the appropriate action; a single byte was chosen to reduce latency. Communication between the Gumstix and CMUCam was slightly more difficult since the Robostix only exposes one serial port from the Gumstix. So, I wrote some bridge code for the Robobstix that performs file I/O on the Robstix serial port, and then transfers information on the I2C bus to the Gumstix.
Keith
by kemsulli
Fri Apr 20, 2007 5:23 am
I posted a video to youtube: http://www.youtube.com/watch?v=dKbBnltDHm4. Sorry about the poor quality. It's sort of long since the robot doesn't move overly fast.
Keith
Keith
by Robo1
Fri Apr 20, 2007 11:44 am
This seams like a stupid question but why does it cart wheel every where. Why doesn't it just walk it would surely be faster and easier to program?
I have to say it does look call rolling all over the place.
Have you tried any control methods for it like ZMP or COG.
Bren
I have to say it does look call rolling all over the place.
Have you tried any control methods for it like ZMP or COG.
Bren