Imitation learning: learning from a person
INTERMEDIATE · 6 MIN READ
Show the robot how to do the task a few times and it learns to copy you.
In imitation learning, a person demonstrates the task, often by guiding the robot arm or using a remote control. The robot records what it sensed and what you did, then learns to repeat it.
It is much faster than reinforcement learning for many physical tasks, because the robot starts from a good example instead of random tries.
A person guides a robot arm along a path. No two demonstrations are the same. The robot averages them into one smooth move (orange). Add more demonstrations and watch it settle.
Learned path is off the ideal by about 3.0 units.
With 1 demonstration the robot copies one person's slip exactly. With 20 the slips cancel out. That is why varied demonstrations beat one perfect one.
You guide a robot arm to fold a towel twenty times. It then folds towels placed in slightly different spots.
Non-programmers can teach robots new skills by demonstration.
Every new skill needs a programmer, or millions of trial-and-error attempts.
Teach a friend to tie a knot using only demonstrations, no words. Notice how many repeats they need.
Common questions
How many demonstrations are needed?
Often tens to hundreds for simple tasks. Varied demonstrations work better than many identical ones.
What can go wrong?
If the robot drifts into a situation you never showed, it may not know how to recover.
Keep learning
Get your personal invite link to share R2BOT with friends.
Teachers and bloggers: paste the embed code into your page to show this demo for free.
Plain English. Free. Unsubscribe in one click.