The problems involved in teaching a robot to play basketball
Economist
Sometimes the hand is really faster than the eye, which can cause difficulties if you are a robot. The robots see the world through video cameras. This means that they perceive it not as a smooth and continuous flow but as a series of individual images - about thirty per second.
It's not bad at all if you need to track the movement of slowly moving objects, but it can be extremely problematic when things are happening quickly. For example, a ball moving at a speed of 10 km/h will have enough time to pass nine centimeters before a robot sees a recent image of it. Therefore, if the robot tries to catch the ball, it will probably not succeed in doing so.
He knows where the ball was a moment ago, but not where it is now. No matter how fast he moves his grasping organ ("hand") to where the ball is supposed to be, it will be too late. Time, and the ball, will already move on.
Until now, engineers used to deal with this problem by using visual information and the laws of mechanics to predict in advance the trajectories of the movement of objects. In this way, the robot can calculate where it should place its grasping organ, despite the lack of direct proof that the ball is indeed there. But such a solution assumes that the trajectories of objects can be predicted, which is not always true.
This is especially not true enough when the robot is required to anticipate its own movement.
Like humans, robots do not always have complete control over their limbs. Although they may know exactly which direction they want to move, sometimes a combination of circumstances makes it difficult to achieve the goal: gears slip, wheels get stuck and various objects block the way. Visual feedback helps humans quickly perform precise movements (try bringing the index finger of the right hand together with that of the left hand behind the head). Buying such an ability for machines should also allow them to make use of their great speed and skill.
This has now become possible, thanks to research carried out by Masatoshi Ishikawa, an electronics engineer from the University of Tokyo. For more than ten years, Dr. Ishikawa has been working on developing a digital "retinal chip" that will locate images and process them, thus accelerating the visual tracking ability of machines.
This device consists of several detectors, which convert the light waves that are received into an electric current. The detectors are arranged in a grid array on the chip. Each of them constitutes the nucleus of a cell. The rest of the cell has an "analog to digital converter" (which receives the continuous input signal from the detector and encodes it into a chain of 1's and 0's), as well as a small number of logic gates (which collect information from their cell and those adjacent to it and perform simple image processing operations).
The result is a chip that performs initial processing of most of the visual information it receives - for example by highlighting the significant contours in the image - before it transfers the image to the robot's main processor. This speeds up the task of decoding the image to a great extent, because the initial processing of the different parts of the image is performed simultaneously, and not as it was done in the main processor - part by part. As a result, the retina chip is capable of processing 1,000 images per second. At such a processing rate, a car moving at a speed of 100 km/h will pass only 2.8 cm between two "blinks of the eyes", and a ball moving at a speed of 10 km/h - less than 3 mm. At this level of certainty about the position of the bone the robot does not need the ability to anticipate the movements of the world around it (or that of its legs and arms). He can simply see where each object is and reach in its direction with his organ of perception.
The demonstrations of this technology that have already been carried out have been impressive. In one of them, the device was able to follow the movement of a basketball bouncing quickly without any significant delay (which is more than most humans are able to do). In another demonstration, two chips were used as part of a larger robotic system. They provided information to a visual tracking system, designed to allow the robot to "look" over time at a moving target, and fed the captured images into a more sophisticated visual system that controlled the robot's perception organ. Thanks to the speed of the retina chip, the robot could reach out to objects that made a fast random movement, and even capture them. For the attention of talent scouts in various ball games.
Economist {appeared in Haaretz newspaper, 13/9/1999{
The knowledge site was at that time part of the IOL portal from the Haaretz group
