For many years now I have searched for practical uses for my robots, including my latest research project into behavioral control software that simulates a pet's behavior . Most of my previous projects have been solutions looking for problems until last Thursday evening when my wife and I discovered a visitor in our house. We were first alerted to the presence when I discovered a little (well not that little) cat suprize by the door. I thought this was strange because we do not have any pets. We looked through the house and could not see any signs of a cat, or what ever it was, and hoped that it was gone by now.
The next morning Amanda put a small dish of assorted food on the floor of the kitchen as a precautionary measure to see if it really was gone. At lunch time we came home to find that the creature liked the tuna, not hot-dogs. We opened all the doors to the outside and searched the house to find a black cat hiding by the hot air vent in the living room. Though we chased the cat into another room where the door was open to the outside, we never saw it actually leave. Needless to say, we were still a little suspicious. With lunch hour running out we closed all the doors and put another dish of food out to see if anything would bite. What has this to do with robots? I will get to that soon enough.
In the evening when we returned home, we knew to look for the cat by the heating vent. We opened all the doors again and the chase was on! At least for the first 20 seconds or so. When the cat ran to the kitchen, I could not believe my eyes when I saw it disappear into a small hole about 4" x 5" in the kickboard of the kitchen counter. Looking through the hole I found a small crawl space just big enough for the cat. There was no way we could reach in and grab it, the hole just was not big enough for an arm and cat to fit through. We agreed more bait would get it out. Not wanting to drill a big hole in the cupboard we needed a trap that would close off the hole when the cat came out. For a computer engineer it did not take long to think of a solution involving a robot.
Just a week earlier I finished building a small mobile robot named CURI-4 equipped with a small CCD camera and a gripper. CURI is tethered to a 486DX40 PC which is equipped with a frame grabber for capturing images. The camera is equipped with infrared LEDs allowing it to see in complete darkness for a distance of around 10 feet.
CURI's drive mechanism consists of two individually powered wheels each driven by a geared DC motor. CURI also has a servo which controls the pitch of the camera, 2 servos that are used by the gripper and a fourth servo is used for a leg that allows the robot to raise or lower the gripper. The camera can be positioned so that it can see the objects the gripper is picking up. CURI's size is a mere 7.5" x 7" x 4" ( 20cm x 18cm x 10 cm), and even shorter with the camera fully retracted. Naturally I tried to see if CURI would fit in the hole so it could drag the cat out by the tail, but the hole was just too small.
Normally CURI's primary behavior is to play with a ball. This involves finding it, picking it up and hiding it. For convenience it forgets where the ball is hidden and repeats the finding/hiding behavior. Superimposed are higher priority behaviors that help CURI avoid obstacles and deal with other real world situations. For example if CURI hears a loud noise, it suspends the ball finding/hiding behavior and attempts to hide itself in a dark corner. After a period of silence, the original behavior resumes. All this hardware and software was about to meet its first simple, but practical test.
I recalled that about 6 months ago I wrote a very sensitive motion detection program that worked with the camera onboard CURI. In 30 minutes, CURI's behavioral software was reprogrammed to move one of her gripper arms when it detected any movement.
I suspended the lid of a cardboard box above the hole by a thin tread and wound it several times around a makeshift pulley. I then led the thread to CURI, which was positioned on the floor about 3 feet away from the hole where CURI's camera was aimed at the third dish of food. I placed CURI so that its gripper would release the spool of thread which lowered the trap door. After several trial runs and fine tuning, the trap was set! We turned off the lights and went to bed.
As I fell asleep thinking about CURI doing battle with the cat, (CURI's behavioral software telling her to move anytime the cat did), a horrible thought struck me. "Hang on ... cats eat little moving things!". However, I was confident that the cat would take one look at the robot and laugh (in a cat sort of way).
Thud!! 2 a.m. Amanda wakes me up and we found a vase lying on the carpet. The cat was out. I rushed to see if CURI survived the fight. CURI was sitting there, spool knocked over and trap door shut!!! I could see that CURI was happy to see me because it was moving its gripper back and forth.
"It worked!!" I shouted as Amanda pointed out that the cat was still somewhere in the house. I put a couple of paperback books over the hole and we went to the cat's favorite resting place (when it was not in the hole). Again the chase was on. The cat did not hesitate this time, 10 seconds later it was dashing for the hole in the kitchen. We were both shocked when we saw the cat struggle with the books and manage to push them out of the way and slip back in the hole. Arrrrrggghhhh!!! CURI just sat there waving at us.
We decided to just wait until morning to see if the cat would come out. We put more food out, but by now the cat had more food in it than our fridge. By 7 a.m. no more food was touched and we resolved to drill a hole in the cupboard. Five minutes later Amanda reached in the new hole with thick gloves and the cat ran out through the little hole. We now placed heavy pots and pans and books and anything else not bolted to the floor over both holes. We found the cat again and with all the doors opened, it finally was last seen scrambling across the green lawn outside. Hopefully heading for another kitchen.
I guess the moral of the story is, a robot can be programmed to do something useful but you can't teach a cat to eat hot-dogs!
If you prefer fully autonomous robots over tethered robots, check out my previous project: CURI-3 developed in 1994 for the University of New South Wales Artificial Intelligence Lab in Sydney Australia. CURI-3 is a fully autonomous mobile robot with a built in IBM PC and stereo cameras. Currently it is being used for a number of research projects.