[Korrelan]

User: Imagine a table
User: Imagine a saucer
User: Imagine a pepper pot
User: Imagine a salt pot
User: Imagine a cup
User: Can you touch the saucer
Can't touch the saucer because the pepper pot is in the way
User: Can you touch the cup
Yes I can touch the cup
User: Where is the cup
The cup (is) on top of the saucer
User: Where is the pepper pot
The pepper pot is in front of me, infront of the saucer (is) on top of the Table
User: Where is the saucer
The saucer is behind the pepper pot infront of the salt pot (is) on top of the Table
User: Where is the salt pot
The salt pot is behind the saucer (is) on top of the Table
User: Move the saucer forward
Can't move the saucer forwards because the pepper pot is in front of it
User: Move the pepper pot forward
OK! I moved the pepper pot forward
User: Move the saucer forward
OK! I moved the saucer forward
User: Move the pepper pot backwards
Can't move the pepper pot backwards because the saucer is behind it
User: Where is the cup
The cup (is) on top of the Table

[squarebear]

Pretty good. What bot was that with?

[Korrelan]

Hi Squarebear

It is a side project… the bot was my own design.

I usually work with massive spiking neural nets that store their learned information using synapse/ axons in semantic spatiotemporal maps etc. 

I spent an hour and wrote a simple sentence parsing routine that applies the same technique to a bot.  This should enable the bot to imagine scenes/ scenarios as well as experience/ view them.   I didn't have time for the bot to learn the laws of physics so I wrote simple object collision rules etc.  The bot has a set arm length and so can reach for objects, unless the path to the object is blocked by another object.  I had gravity but forgot friction… the cup would now stay on the saucer lol.

[Freddy]

That is clever. At first I thought it was an example of how a bot might work, I did not think it was an actual conversation. Very good 

[Art]

It very much reminded me of a somewhat dated but still remarkable program called, Blockworld, by Terry Winograd of MIT. Terry wrote it back in the 70's and it basically "knew" the physical shape and location of a variety of objects (cubes, pyramids, etc.) and would respond to various placement commands or queries.

I like your approach and would certainly like to see more.

Thanks!