Robots can Learn

If the current research of Dr. Ashutosh Saxena of Cornell's Personal
Robotics Laboratory becomes successful then Robots of future will
learn to manipulate objects and will develop the ability of
generalization. This learning procedure will help Robots to adapt
according to new environments.

The main theory behind this is the use of machine learning programming
which leads to observation of events by Robots and find commonalities.
For an example a robot will see different cups, learn the common
characteristics of cups and will then identify cups in future no
matter what size and shape they will have. This kind of process will
also teach robots to handle cups correctly.

According to Saxsena, these new robots would be able to make decisions
like how to place cups on different places. Like a cup will be placed
upright on table while upside down in dishwasher. Robots will get
training of placing strategies and will be able to apply then on other
objects by the process of generalization.

In testing process scientists placed different things like a plate,
mug, glass, bowl, spoon and some other objects at different surfaces
like on flat surface, hook, stemware holder etc. Robots then tested
the suitable locations for placement of these objects after studying
the environment and gave priority to best locations to place an
object.

After such training Robot placed objects 98% correct at previously
seen environment and objects. This accuracy was 95% for new objects
and environments and that can be improved further by longer periods of
training.

The base of all this is the contextual relations in 3-Dimensions.
Although making a Robot that acts just like humans will take long time
but by making Robots like these is a giant leap in Robotics.