Opening locked doors using
Human Level Artificial Intelligence
Note: To make this website free to the public
please click on an ad to support my sponsors or you
can make a tax-deductable donation using Paypal
(click on the donation icon on the left).
DARPA, a defense
agency of the US government has announced the Robot
Challenge 2012 and the first contest will be held on
Dec. of 2013. Their mission is to give money to
Universities and technology companies to build them a
robot that can do "mundane human tasks". According to
the government, the reason for the the contest was to
build robots that can do rescue and search missions.
They wanted to build robots that can go into
dangerous zones to do tasks. The Japan earthquake was
one example.
There were several
tasks (about 8 tasks) the human robot had to
accomplish in the robot challenge. 1. walk around
dangerous terrains. 2 open a locked door. 3. identify
a broken pipe. 4. climb a ladder. 5. use tools, like
a sledge hammer. 6. fix a broken pipe. 7. control
vehicles with simple functions. 8. push buttons.
Basically, they are
trying to build a robot that can think, act, and
behave like a 5-year old child. Opening a locked door
is a very difficult problem to solve in current AI
(2013). If the door was locked and the robot didn't
have the key, then how can the robot enter the
house?? Human beings solve the problem by using
common sense. They can use logic to come up with
alternative methods to get into the house. For
example, breaking the window and entering the house
is one option. However, humans know that this method
is not desirable because the owner has to pay for the
window repairs. They can also find a spare key, which
is usually hidden in a secret location. Or they can
find someone that has the same key. These options are
analyzed and a decision is made on what actions to
take to open a locked door. This locked door problem
has been an unsolvable problem for AI researchers
since 1950.
In this video, the
robot is playing a videogame, where a character in
the game is locked in a dungeon. Within the dungeon
are 2 locked doors and 1 hidden passage. The player's
(the robot) mission is to get out of the dungeon. In
order to do that, he has to unlock 2 doors and find
the secret passage.
The robot uses common
sense and logic to unlock doors. In one scene in the
game, the player is locked inside a room. There is a
guard on the floor that is dead and a locked door.
Using common sense, the robot is able to understand
that the guard usually has keys to unlock doors. He
approaches the dead guard and searches him and finds
the key. Next, the robot understand that locked doors
require a key in order to open. So, the robot takes
the keys from the guard and unlocks the door. Without
common sense, the robot would not have been able to
unlock the door. In other settings the environment
might be even more complex. For example, there might
be no keys and the player is forced to find
alternative ways to get out (like finding a secret
passage). Or the player can take an axe from the dead
guard to chop open the locked door.
This video shows a
robot trying to open locked doors. There are no sound
in the video because I wanted to show the viewers
what the robot is thinking while opening locked
doors. The flashing text and freeze frames are the
internal thoughts of the robot and not instruction
text for the viewers. These internal thoughts
describe the details of how the robot produce
intelligence.
My robot doesn't use:
planning programs/heuristic searches (used by MIT and
Stanford University), Bayesian's probability theories
for decision making, Bayesian's equation for
induction and deduction, semantic networks for
natural language understanding, predicate calculus,
common sense systems, first-order logic, rule-based
systems, genetic programming, or MACHINE LEARNING.