[Dprglist] Deadend cul-de-sac

David P. Anderson davida at smu.edu
Thu Jul 22 22:25:53 PDT 2021


Sorry for the repeat post.

Howdy DPRG,

Karim wrote:

"There was a "what would David say" portion in the discussion tonight. 
We had some fun speculating on your take on Doug's desire to craft an 
explicit algorithm to effectively do your emergent cul-de-sac escape 
capability. I'll often fall into that natural how-would-a-human do it 
approach as opposed to the somehow harder how-would-a-bug do it 
approach. It make sense to argue that you should prove you can do it 
like a bug might - effectively, repeatedly, before trying explicitly 
higher level algorithms."

Well what a great discussion.   I'm not sure, though, that I can 
identify with the premise, i.e., that an explicit algorithm to replace 
the emergent cul-de-sac escape capability would necessarily represent an 
improvement, or be more efficient, or in some way have more Natural 
Goodness(TM).

As you might expect, I am not unfamiliar with the approach that Doug is 
suggesting.   I can only speak from my own experience which is certainly 
not exhaustive.  But I think you will find it extremely hard to do that 
well.   With emphasis on the word 'well.'

My experience is that approach tends to be brittle, and prone to 
unforeseen consequences.  False triggers and missed triggers.  And you 
end up running down all these exceptions and failure modes and the 
"explicit algorithm" grows and grows and gets more complex and harder to 
control.

To be clear, I would be tickled pink if Doug's "explicit algorithm" can 
be made as robust as the methods my robots are using.  And doff my hat.

I could go on at length as to why I think that is not the case, but 
experience is the best teacher.   And the proof is in the proverbial  
pudding: a working robot.  All algorithms work great in the imagination.

So get your code working and lets go out to the old TI building, or come 
here to SMU or better yet go down to Fair Park, and see your proposed 
algorithms in action.   That is the only real test. If they really are a 
better way of doing things, and able to better handle all the multiple 
vagaries of the real world, it should be obvious.  And I will knight you 
Sir Doug and refer to you thereafter as The Robotly One.

cheers!

dpa




More information about the DPRGlist mailing list