Monkey See, Robot Do

Researchers implanted electrodes in a monkey's brain in North Carolina and used signals from the monkey's brain to make a robot walk in Japan.  The monkey was trained on a treadmill, but they eventually stopped the treadmill and the monkey could still send the appropriate signals to make the robot walk.  Here's the first part of the experiment from the NYT (and see highlighted comment at the end):

In preparing for the experiment, Idoya was trained to walk upright on a treadmill. She held onto a bar with her hands and got treats — raisins and Cheerios — as she walked at different speeds, forward and backward, for 15 minutes a day, 3 days a week, for 2 months.

Meanwhile, electrodes implanted in the so-called leg area of Idoya’s brain recorded the activity of 250 to 300 neurons that fired while she walked. Some neurons became active when her ankle, knee and hip joints moved. Others responded when her feet touched the ground. And some fired in anticipation of her movements.

To obtain a detailed model of Idoya’s leg movements, the researchers also painted her ankle, knee and hip joints with fluorescent stage makeup and, using a special high speed camera, captured her movements on video.

The video and brain cell activity were then combined and translated into a format that a computer could read. This format is able to predict with 90 percent accuracy all permutations of Idoya’s leg movements three to four seconds before the movement takes place. [emphasis added-AK]

I found this last statement rather surprising/confusing/unclear.  Presumably Idoya forms intentions to move less than three to four seconds before the movement takes place.  So I don't understand how the computer can predict "all permutations of Idoya's leg movements" so far in advance.

Speaking of exoskeltons, here's a non-neuro achievement from Tokyo University of Agriculture and Technology to help farmers: