www.cs.uwyo.edu/~dspears/videos/formation_20060816.avi
Strictly formation forces, no goal force. Robots "oscillate" quite a bit, but that's something we can reduce by including an offset in the coordinate
computations to account for an off-centered cone. I'll also try decreasing the step size, so that the force computations are done on more recent
trilateration data. With all of the improvements that Paul and Tom have made, we can completely localize the entire swarm 3 times per second! Thanks
guys -- awesome job.
www.cs.uwyo.edu/~dspears/videos/cluster_20060816.avi
A test run with a very high G; as expected, clustering occurs.
www.cs.uwyo.edu/~dspears/videos/chemo_jar_20060816.avi
The first CPT experiment: the small ethanol jar is hard to see, but 2 robots fight for it continuously. Watching this made me think that maybe we should
suggest that "RoboCup" use VOCs instead of the soccer ball ... but then one robot knocked the bottle over, and I had a real "chemical spill" scenario on
my hands (and my feet :) ...
www.cs.uwyo.edu/~dspears/videos/chemo_puddle_20060816.avi
Since there was no use crying over spilt alcohol, I decided to see if the robots could
find the puddle. I started them about 4' or 5' away, and used the small computer fan to induce a gradient. One Maxelbot made it to the edge of the
spill in 3 minutes. The desired inter-vehicle spacing was 2.5' (30"), so it's hard to say for sure that chemotaxis worked (it could have been just the
formation forces that pushed the swarm apart and toward the spill), but at least this particular run is an encouraging success. With more
experimentation I'll get a much deeper understanding of the performance factors, and should have a better explanation for what you're seeing right now.
www.cs.uwyo.edu/~dspears/videos/formation1_20060823.avi
Simple formation test. Works great!
www.cs.uwyo.edu/~dspears/videos/formation2_20060823.avi
Formation starts with a missing robot, which soon rejoins the group.
www.cs.uwyo.edu/~dspears/videos/formation3_20060823.avi
Everyone starts far, far away, but then pulls it together.
www.cs.uwyo.edu/~dspears/videos/formation4_20060823.avi
Initially a wide line, the robots come back into a triangle.
www.cs.uwyo.edu/~dspears/videos/goal1_20060823.avi
First try with a goal force: it's a minor fender bender, but we
crossed the lab in a little over 2 minutes!
www.cs.uwyo.edu/~dspears/videos/goal2_20060823.avi
Goal force again; no collisions this time, but with more turns.
www.cs.uwyo.edu/~dspears/videos/20060824_hookes_formation.avi
A mix of tests for the Hookes Law driven formation with no goal force.
All movies are shown at 3x the normal speed.
www.cs.uwyo.edu/~dspears/videos/20060824_hookes_goal.avi
Hookes Law team crosses the lab. At the end, when one of the robots is blocked by a chair, the other two wait for their neighbor -- the formation force
keeps the other robots from bumping into the wall.
www.cs.uwyo.edu/~dspears/videos/20060824_hookes_obstacle1.avi
A somewhat lengthy clip that shows just how robust the formation is. While one Maxelbot is bumping its way out of a cardboard box, the other one is
getting tangled up in bubble wrap. One can only imagine the noise on the acoustic channel, but the swarm recovers!
www.cs.uwyo.edu/~dspears/videos/20060824_hookes_obstacle2.avi
A light box is not a problem for the mighty Maxelbot :)
www.cs.uwyo.edu/~dspears/videos/20060824_hookes_obstacle3.avi
A heavy box is a challenge, but one we pass with flying colors. Note how the "free" vehicle on the right acts like a bridge between the other two --
it's a great example of emergent behavior.
www.cs.uwyo.edu/~dspears/videos/20060824_LJ_goal_vs_formation.avi
Our first test of the Lennard-Jones potential: in this scenario, two robots are disabled, and the third vehicle makes it's way back to the swarm against
the goal force (which is "up" in the video and North in the lab). We tested this case because LJ is strongly repulsive, so we wanted to make sure we
select correct parameters to give a reasonably strong attractive component. Seems to have worked well in this test.
www.cs.uwyo.edu/~dspears/videos/20060824_LJ_goal_strong.avi
In this run, the goal force is very strong, so the LJ formation is less stable, but we still managed to cross the entire lab as a group in just 2
minutes. Please note the large reduction in the number of rotations for this configuration. This might be a good way to proceed if we need fast
movement and are less worried about lattice imperfections.
www.cs.uwyo.edu/~dspears/videos/20060824_LJ_obstacle1.avi
LJ had a harder time going around the obstacle, probably because of the weaker attraction force.
www.cs.uwyo.edu/~dspears/videos/20060824_LJ_obstacle2.avi
In this case we could not get everyone past the big box, but take a look at how the two robots push together: you can see how the formation directly
causes this effect -- again, a good illustration of an emergent behavior. Also, note that because of the obstacle's size, the acoustic shadows must be
very large, but the swarm seems to cope well with that noise. Big congrats to our dedicated hardware team!
www.cs.uwyo.edu/~dspears/videos/20060829_chemotaxis_plate1.avi
I started out by pouring a little bit of ethanol into a saucer and just watching the Maxelbots. After wondering around for a little while, they managed
to find it!
www.cs.uwyo.edu/~dspears/videos/20060829_chemotaxis_plate2.avi
I wanted to see if the swarm can surround the source to get a feel for how feasible our emitter ID method might be, and in this case, things turned out
quite well (except for the fact that one of the robots flipped the plate onto itself -- I used to do that a lot as a kid, so I know how embarrassed that
bot must have felt ... :)
www.cs.uwyo.edu/~dspears/videos/20060829_chemotaxis_plate3.avi
To avoid similar accidents, I placed the saucer on a support stand. I also increased the starting distance to 10 ft. I'm quite impressed by this test
run. I am also amazed by the extent that the real vehicle behavior resembles what we saw previously in CPT simulations.
www.cs.uwyo.edu/~dspears/videos/20060829_chemotaxis_diffuser1.avi
I repeated the experiment above, but this time I used a large aquarium air diffuser attached to an air pump for the emitter. Again, the starting
distance is 10 ft. and plume tracing time is just 3 minutes!
www.cs.uwyo.edu/~dspears/videos/20060829_chemotaxis_diffuser2.avi
Another successful test. This movie clip is in real time (all others are @ 3x the
normal speed) to give you a better feel for the actual rate at which the robots are tracing. Also, you see that the swarm tends to zero in on a location
just in front of the emitter. This observation is supported by the CPT literature, and also something I've seen inside the flume: the chemical
concentration peak is often shifted from the true location of the source.
www.cs.uwyo.edu/~dspears/videos/20060829_chemotaxis_diffuser3.avi
This is my best result so far. Again, the robots covered about 10 ft. in under 5 minutes. Also, at the end, one Maxelbot was so ambitious, that it
ended up damaging itself in the pursuit of the emitter. I kept the camera rolling as parts started flying, just to show that the mission goes on, even
with the swarm taking on heavy damage :)