Social Robots Engaged in Collective Construction

Submitted by stefano on Sun, 2011-09-11 12:51.
in

This is an interesting video. Notice how they break it down:

  • Locomotion
  • Locomotion on structures
  • Autonomous navigation
  • Manipulation

Pretty neat, no? Each of the participants have relatively simple programs with regard to the structure they are building. There is no controller for the whole. There is just a system of autonomous agents. It's a great example of self-organizing behavior.

It's curious that they don't really talk about sensors. Rather, they emphasize the goals of the actors, not the means they employ. However, they do need to be "aware" of one another at some level. This is not social communication per se, but it is a degree of social behavior. They do recognize one another and act as if they are in pursuit of a common goal.

However, notice that this is not self-organization in the same way that living systems are. This is self-organizing construction "behavior" but the system and agents that "construct" are distinct from the "structure" they produce. If these robots were to manufacture THEMSELVES and they were incorporated into the structure, that would be self-organization.

Very neat indeed...

Submitted by gavinovillapiano on Mon, 2011-09-12 18:42.

What would be even neater would be to have self-replication, as you suggest.

Without it, the problem with this experiment is that it starts at too high a level. Intelligence, tho minimal, has been imputed into the robots from other intelligence (with your indulgence, um, us...).

I love stuff like this. I love order emerging from very, very simple organizing principles. Can't be much cooler.

So, can we find a way to start at a much more simple level - one rule - survival, ie, self replication - and see what happens?

Wouldn't that be slick?

GV

what do we mean by 'self organization'?

Submitted by stefano on Sat, 2011-11-26 03:40.

Exactly, Gavin. A community of autonomous sponge cells self-organize... but each of the cells began to cooperate to form a super-organism based on their own internal structure. In other words, each cell is self-organized AND the society of cells is self-organized at another level. In the case of these primitive robots, clearly the structure of each autonomous node in the network is NOT self-organized. On the contrary, each of these are manufactured by people. And yet, the society of nodes is clearly self-organized based on a highly distributed program executing in autonomous nodes.

Thanks for the comment!

Deconstructing the Constructors

Submitted by ensopc on Mon, 2011-09-12 17:47.

Hey Steve,

Great to see a new post! I love these kinds of demonstrations. I'm thinking this is not self-organization. In the video, right before we see the multi-robot simulation, we see a graphic that includes a "user input". Doesn't this qualify as an external plan, which means that it is not self-organizing? I.e., the collection of robots and materials by themselves do not constitute a self-organizing system, because the user input is required to achieve the pattern. I guess the robots + the materials + the input forms a self-organizing system, but that feels like cheating.

Also, I struggled with the "social" aspect of this. I don't think the robots were interacting with each other, but I can't be sure. They do seem to stop in certain situations and "wait" before moving again, but I couldn't tell what each robot is sensing and what level of understanding each robot forms about it's environment and it's "position" in the environment. Clearly, each robot knows the plan, but what else does it know? As you mentioned, I think we need to know more about the sensory, interpretive, and metadata functions of the robots. If they are not aware of each other as robots (rather than just more, undifferentiated environmental "stuff"), then it can't be called social interaction, can it? It may look like, or mimic, social interaction from a great enough distance, but it wouldn't be.

Cheers!

Paul

what exactly do we mean by 'plan'

Submitted by stefano on Sat, 2011-11-26 03:44.

Hi Paul!

See my comment to Gavino above.

The key is what we mean by plan. In fact, the plan held by each robot is not a plan consisting of operations to construct a structure collaboratively. In fact, there is no such plan. Rather, the sequence of operations and even the final state of the system is specified nowhere. It emerges from more primitive specifications we might call goals, strategies and constraints.

Hope this help.

Thanks for the comment!