View Single Post
  #13   Spotlight this post!  
Unread 29-07-2003, 00:37
dlavery's Avatar
dlavery dlavery is offline
Curmudgeon
FRC #0116 (Epsilon Delta)
 
Join Date: Jun 2001
Rookie Year: 1996
Location: Herndon, VA
Posts: 3,176
dlavery has a reputation beyond reputedlavery has a reputation beyond reputedlavery has a reputation beyond reputedlavery has a reputation beyond reputedlavery has a reputation beyond reputedlavery has a reputation beyond reputedlavery has a reputation beyond reputedlavery has a reputation beyond reputedlavery has a reputation beyond reputedlavery has a reputation beyond reputedlavery has a reputation beyond repute
Quote:
Originally posted by rlowerr_1

Thanks for the tips but im not to sure that the RCX will be able to read the sensors fast enough/ control the motors fast enough to perform any controlled falling.
Actually, the RCX can probably keep up with the overhead of a basic dynamic stability control system. For example, just one RCX can be used to control a simple Segway build from Lego parts.

You will want to be careful about monitoring the cycle time on the control loops. If the time that it takes to complete one control loop begins to approach either the sensor refresh rate or the response time of the motors/actuators, then you will need to back off a little. But you can do this by either decreasing the sensor processing required (tough to do), or by just adding an additional RCX and sharing the processing load between multiple units.

If you do end up using multiple RCXs, there are several ways to go. Lots of information is available on layered architectures (see things like the three-layer architecture by Eran Gat/JPL, or Reid Simmons work at CMU).

Alternately, you could break the entire system into similar modules and assign one RCX to each (e.g. each leg is a module - just mirror images of each other - and the torso is just a big leg module, etc., each with it's own RCX). Each RCX is then programmed with the control loops (and slightly higher-order behaviors) required to just run that one module. The fun part is figuring out how to have the RCX units pass information between themselves to create coordinated motion. This quickly gets into a really interesting problem (this is where Rod Brooks/MIT started with his subsumption architectures in the mid-1980s - the same Rod Brooks that Andy mentioned above).

As I noted above, if you really want to get serious about this there is a lot of information to explore and a lot of good work that has been done that can help you get started. But the best part is that walking machines are just plain cool! - if you can pull this off, you will have a lot of fun and the undying respect of us all!

-dave