Training,Learning,neural nets and etc
Training,Learning,neural nets and etc
by MoSin
Fri Jul 13, 2007 4:42 am
Hi,
I'm new here so first, I'm a student in electrical engineering and the end of the bacc arrives fastly! I'm thinking about doing a project about learning and robotics. I was looking at some robots over the net. I'm not very interested in building it on my own. I'm more a programmer than a designer, so it's the brain that interests me.
So, I saw the bioloid. I wanted to know, if someone knows a little bit about neural nets, does any body thinks train the robot to learn some motion is possible? The motion builder is cool, but I would be more happy to make him learn to walk that using poses. I know the controller isn't big, but I tought it would be possible to give him a packsack with other controllers with more rams and roms and make them communicate.
I'm really interested to have opinions about that so thanks in advance!
I'm new here so first, I'm a student in electrical engineering and the end of the bacc arrives fastly! I'm thinking about doing a project about learning and robotics. I was looking at some robots over the net. I'm not very interested in building it on my own. I'm more a programmer than a designer, so it's the brain that interests me.
So, I saw the bioloid. I wanted to know, if someone knows a little bit about neural nets, does any body thinks train the robot to learn some motion is possible? The motion builder is cool, but I would be more happy to make him learn to walk that using poses. I know the controller isn't big, but I tought it would be possible to give him a packsack with other controllers with more rams and roms and make them communicate.
I'm really interested to have opinions about that so thanks in advance!
by Biodog
Fri Jul 13, 2007 8:19 am
You can use your PC with a good size cable and leave aside your CM5 if you want, you'll just need components to connect your servos to your computer.
see this topic
http://robosavvy.com/forum/viewtopic.php?t=990
and go to Jon wiki, and things will be much more clear to you
http://www.bioloid.info/tiki/tiki-index ... r+Projects
(by the way, are you french? I am
see this topic
http://robosavvy.com/forum/viewtopic.php?t=990
and go to Jon wiki, and things will be much more clear to you
http://www.bioloid.info/tiki/tiki-index ... r+Projects
(by the way, are you french? I am
by limor
Fri Jul 13, 2007 6:00 pm
rustyarmor has linked Matlab Simulink Windows-realtime-target with CM5 to allow real-time control of the Bioloid at up to 100hz. http://robosavvy.com/forum/viewtopic.php?t=1514
If you have experience with Simulink, this is a great way to start exploring intelligent control without having to write too much code.
If you have experience with Simulink, this is a great way to start exploring intelligent control without having to write too much code.
by MoSin
Sat Jul 14, 2007 10:45 pm
I never touched mathlab or simulink but I hope to do so soon. I have no experience in that kind of simulation but I saw that there are some informations available here. I looked at the gumstix. Damn, it was the first time I saw that! It looks very nice! Is it difficult to implement somehting in that. I guess there is a thread that explains everything. There are so much things I need to learn!
by Robo1
Sun Jul 15, 2007 10:38 am
Hi there.
I'm doing the same sort of thing for my dissertation. the main thing you have to consider is how you get the feed back so that the program know how well it is doing. So for you I would say making a system to optimize a walking gait is a good start. So the the first thing to think about is how do you measure (obtain) a fitness value.
I'm going to use a ping))) and a digital compass. So the bot is in a square box with the ends at North and south. it will align it self so that it's facing north take a distance reading then take 5 steps then measure with the ping the distance again. These two readings give you the distance traveled the compass will give you how straight it walked and the accelerometers will tell you if it feel over (stability).
you can then put all this into a equation that will give you a fitness value. you then want the robot to automatically turn around line it self up and repeat the experiment with a different walking gait.
My project is on evolutionary algorithm to create the walking gait.
I'm also going to use a gumstix, but having a PC create the walking gait and the bot transmitting the final value distance direction and stability back to the PC.
this way I can log it and put it in to excel to plot pritty graphs.
Bren
I'm doing the same sort of thing for my dissertation. the main thing you have to consider is how you get the feed back so that the program know how well it is doing. So for you I would say making a system to optimize a walking gait is a good start. So the the first thing to think about is how do you measure (obtain) a fitness value.
I'm going to use a ping))) and a digital compass. So the bot is in a square box with the ends at North and south. it will align it self so that it's facing north take a distance reading then take 5 steps then measure with the ping the distance again. These two readings give you the distance traveled the compass will give you how straight it walked and the accelerometers will tell you if it feel over (stability).
you can then put all this into a equation that will give you a fitness value. you then want the robot to automatically turn around line it self up and repeat the experiment with a different walking gait.
My project is on evolutionary algorithm to create the walking gait.
I'm also going to use a gumstix, but having a PC create the walking gait and the bot transmitting the final value distance direction and stability back to the PC.
this way I can log it and put it in to excel to plot pritty graphs.
Bren
by Robo1
Sun Jul 15, 2007 5:58 pm
I'm going my dissertation for a Masters in robotics at bristol robotics lab http://www.brl.ac.uk/
To simplify your project have you thought of having a table of servo positions similar to the way people pose there robot and then use a learning program to optimize this table.
Bren
To simplify your project have you thought of having a table of servo positions similar to the way people pose there robot and then use a learning program to optimize this table.
Bren
by MoSin
Sun Jul 15, 2007 7:17 pm
Hum, I'm not sure about your question. I'm at the beginning of an idea of a project. I don't have the robot, I don't have all the skills I need. It is still at the stage of an idea! Right now, I read on what has been done. I don't know a lot about biology and human motion.It's a long term idea of project I have. So right now, I will stay at the theoric side. When money is there, I'll buy a robot and anything needed. I must get lots of knowledge before I throw myself completely!
by MoSin
Fri May 23, 2008 2:30 pm
well, my project is started for real now. It has changed a lot but at least I'm working with the robot.
Right now, I'm working on an animation engine. Something like Jon has made. My next step is to implement forward kinematic, robotic matrices to represent the robot itself in space. I'll use that for better interpolation and also to do an opengl viewport. I just saw that cosa did something like this. I left the intelligence thing aside for now. I want to build the best tool before going on intelligent balance control, adaptativity and etc. Well, I just wanted to update my project status!
Right now, I'm working on an animation engine. Something like Jon has made. My next step is to implement forward kinematic, robotic matrices to represent the robot itself in space. I'll use that for better interpolation and also to do an opengl viewport. I just saw that cosa did something like this. I left the intelligence thing aside for now. I want to build the best tool before going on intelligent balance control, adaptativity and etc. Well, I just wanted to update my project status!