The most popular and comprehensive Open Source ECM platform
Let us know more
Robotics: Evolving Robots to be able to Improvise in the Real World
Robots nowadays can be trained to do seemingly anything. But making general-purpose robots that can interact under many different scenarios and environments are a long ways off. Robots today aren’t able to easily reuse in a different environment what they have learnt that works well in one environment.
Researchers have found that it’s easy to fool or confuse a robot and its base algorithms. Changes to color, position or surrounding objects can easily trip up robotic algorithms.A big problem is that scenarios that humans encounter in their everyday life aren’t as exacting as the rules needed to followed for doing something like industrial-robotic assembly-line work.
Ingmar Posner, deputy director at the Oxford robotics institute, told the WSJ that relying on robots to learn by requiring thousands of examples as input and hours or days of training may not be practical. “It’s not clear that’s particularly useful. You want the machine to pick up pretty quickly what it’s meant to do.”
Stefanie Tellex, robot researcher at Brown University, told the Wall Street Journal that one problem is that AI scoring functions don’t exist for the real world. Standard AI training programs are based on optimizing a scoring function. The computer manipulates parameters to be able to produce the highest score. “The reward signal is so important to making these algorithms work,” Dr. Tellex said.
