About 11 people showed up, including Pete S., Fred S., Paul B., Wm L., Aaron?, Tim, Scott, Monty G., and a couple others I’m forgetting…
Wm Leler talked about the next Make PDX meeting:
Jan 23 — Focusing on… Robots! Wm’s requesting people to bring and show off their robotic projects. Contact Wm ahead of time at wm <at> opentechspace <dot> org or just show up with your robots.
Not much discussion (Mark G. out due to cold.)
This meeting focused on taking a step back from the comm protocol discussion of last week and discus what constitutes a swarm and how it differs from other types of collective AI’s.
Key design decisions:
- No master intelligence should be required, but should not be prohibited either.
- Neighbor-to-neighbor messaging support required.
- Standard mechanical platform should not be required, but it was recognized that there would be a desire for one.
Swarm vs Collective
University of Stuttgart (UoS) considerations for a swarm: No global coordinates, perception, or communications
PARTS: Design requirement: No master intellegence required, but not prohibited
UoS: Swarm should be able to accomplish tasks greater than a single swarm member can.
PARTS: Also consider tasks that can be done by a single robot, but can be done more efficiently by a swarm
UoS: Self organizing
Discussed larger picture of what all is involved in a swarm platform.
Communication standardization is a clear need.
Power maintenance standardization may also be needed.
Is charging an individualistic behavior or a swarm behavior? Could be either, depending on if chargers have a global beacon, or if bots need to pass messages to help others find the chargers; or is the charging station part of the swarm? Or do swarm members share charge they gather when in proximity to a charging station?
Inductive charging raised as a possibility.
Discussed chain formation of a specified nominal length as a basic unit of self organization.
Individual random-walk to find other units.
Assume leader, follower, or tail (tail caps chain to desired length).
Collective random-walk to find objects.
Adjust chain length to fit around object — redudant robots leave chain
Distance measuring accuracy can be low for chain formation
UoS used IR comm signal strength for distance measurement — good enough to keep adequate distance from neighbors.
Distance measurements will vary with changing environmental lighting conditions, so swarm formations may contract or expand.
Comm protocol potentially needs to handle different levels of measurement accuracy for different sensor capabilities
What tasks require accurate distance measurement? Mapping for one. (Were others discussed?)
Higher level organization
Marching-band like formation. Perceived as difficult w/o a master intelligence.
Geese formations. Question asked if this was truly swarm behavior dependent on communication between members, or if is emergent behavior from individuals acting upon sensory information of nearest neighbors. Sensory data could be considered a form of passive communication, however.
Need a standard platform? Should not be required, but there would most likely be a desire for one by others.
Depends on desired task and capabilities.
10cm robots with specialized abilities playing soccer as a swarm brought up.
Need different size classes and different complexity classes?
Different size classes for standardization of support structure interactions (i.e. charging).
Learnings from other swarms?
Should research other attempts at swarm robotics and learn what we can.
Commercial swarm, Centibots, and MIT swarm brought up. Also, robots.net article on SanDiego State University class that required students to purchase 2nd-hand dog toys (Aibo, Cybee (sp?)) and turn them into a robot swarm that could locate dangerous gas build-ups in a dump.
How long of range of communication needed?
Depends on swarm density, which is task dependent. Are we targeting a small controlled world, or the real world? The latter will require larger communication ranges. Mapping/exp will likely need larger comm ranges.
Neighbor-to-neighbor communication or more global communication?
UoS avoided global communication due to scalability concerns — too many units in a single area would create too much interference. Not as much concern expressed amongst PARTS? (Can’t remember)
Decided that neighbor-to-neighbor communications should be supported by all comm protocols, but shouldn’t eliminate possibility of wider area communication. Could limit wider area comm to certain distance??? (Don’t remember the details of this.)
Question asked: What is cheapest communication technology that could also be used for distance measurement?
RF discussed; Paul showed off a TI EZ340 developement board and discussed it’s capabilities. Does provide strength of signal measurement that could potentially be used for distance sensing. Range ~50′. ~$20/endpoint
IR clear winner on cost for short-range
Mesh network? Started discussing whether transient neighbor-to-neighbor communication where no more than two neighbors are in range at a time, but messages are stored and carried to other neighbors is really a mesh network, but got distracted.
Do individuals have unique ID’s?
How much time do we want to spend on comm HW vs using an off-the-shelf solution and focus on algorithms?
Larger goals for swarms
What do we want our swarm to do? Need a bigger goal to shoot for, but should have smaller goals along the way (like LineMaze tabletop challenges).
Mapping. Swarm units map independently and report back to central intellegence.
UoS: Swarm games (insert UoS criteria for good swarm games and example games)
We still need a good large goal for our swarm.
Smaller stepping-stone goals
Random walk w/o colliding. More of an individualistic behavior, but a basic first step towards swarm behavior.
Dispersion. This was discussed a bit. Idea here is for robots to arrange themselves to fill a space with a consistent density. Then how does the swarm behave when a door is opened to another room? What if the swarm consists of different sized members, only some of which can fit through the door? How are these members informed that they are needed to move to the other room. One possibility is to require members to keep changing their nearest neighbor so that smaller robots will eventually cycle around to the door opening. Another possibility is for robots closest to the door to recognize there is an opening of a certain size and pass a message throughout the group so that smaller robots will come forth.
$100: ~half of meeting attendees said they would be willing to pay $100 for an individual swarm member
$200: nobody willing to pay that much.
Really, depends on capability of robot
Discussed learning from PARTS’ past experience with Mark III.
Need to be clear up front what rights PARTS wants to retain and we want to allow people to do with our work.
Different open-source licenses and public-domain discussed.
GPL variations; some w/ limits on profits; some w/ required contributions of derivative works back to communinity;
Creative Commons licenses (CC Attribution seems most likely);
What about different licences for members vs non-members?;
Need to study options and discus more.