Can a robot design a humane mouse trap?
Last updated: 15 Nov 2010
Installed: 15 Nov 2010

A challenge for future robots

Much current work in robotics (e.g. The CogX project) aims to give robots sufficient understanding of the world to be able to explore a portion of the world, learn how it is structured, then later work out how to get back to previously seen locations, or to move objects from one place to another, or even to assemble objects to form larger structures.

The mouse-trap problem requires the ability to combine various kinds of knowledge, of animal behaviours, physical sizes and weights, gravity, balance, and how the shape of a container can constrain the movements of an animal within it.

When we noticed a mouse in the house and failed to catch it, Alison, my wife, looked for information about humane mouse traps and found this web page, by Chris Glass: http://journal.chrisglass.com/2005/09/how_to_catch_a_.html

We already knew that our mouse had a liking for chocolate, so she decided to make a similar trap and use a lump of chocolate as bait.

The first version on a shelf where the mouse had originally found and tasted chocolate (eating through the wrapping) did not catch anything.

Here are two views of it:

A better mouse trap
So we moved the trap to the top of a box file in another room, with a ramp for easy access and some bait on the ramp, and a few days later it worked. The mouse was taken to a park some distance away and set free.

(1)       (2)       (3)

Picture (1) Shows the ramp up to the trap, balanced on a box of music. Picture (2) gives a closer view showing a cashew nut at the top of the ramp to help tempt the mouse to the top.
Picture (3) shows the view from above. The metal clip was added to stop the tube falling over without the mouse. Chocolate inside the end of the tube is not visible.

(4)       (5)       (6)

Picture (4) from above shows the mouse hiding in tube, tail visible. In Picture (5) the mouse is right out in full view. Picture (6) shows the mouse on its way back
to the tube, before being taken to be released in the park.

What does this have to do with robots?
Some tasks:

• Try to work out what sort of ontology a robot would need in order to be able to think up this design, or understand it if presented by someone else, and then work out what to do to create such a trap.
• What empirical generalisations would the robot need to have learnt in order to be able to come up with such a design, or to understand it?
• What reasoning capabilities would be required?
  • E.g. in the picture the toilet roll is partly squashed: how could the robot understand why?
  • If no toilet roll is available, how could the robot think about a substitute?
    How would you think about a substitute?
  • How could the robot choose a suitable container for the mouse to fall into if a bucket is not available? Would a saucer or soup bowl do, and if not why not?
  • Suppose the robot gets a lump of chocolate and puts it at the end of the tube, but as soon as the tube projects over the edge of the shelf it falls: what options does the robot have and how should it choose between them?
• Is there a difference between the knowledge required to understand such a design and the knowledge required to be able to invent such a design?
• What else is required in order to be able actually to make such an instance?
  Including choosing a suitable location, suitable bait from among the items available in a typical house.

• http://www.commsp.ee.ic.ac.uk/~mcpetrou/V19N4O97.htm
• http://www.commsp.ee.ic.ac.uk/~mcpetrou/iron.html

Challenges involving vision

• Seeing a cup, saucer and spoon.
  How can a robot understand the possibilities for changing relationships between cup, saucer and spoon, and ways of grasping and lifting them.

• http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#talk88 different sort of visual challenge, suggesting an architecture using multiple multi-stable dynamical systems.

TO BE EXPANDED, POSSIBLY...

Maintained by Aaron Sloman
School of Computer Science
The University of Birmingham