We’re still waiting for that big breakthrough in artificial intelligence – the advancement that will take robots from performing parlor tricks to being truly intelligent. Currently, robots simply do as they are programmed to do. There’s no real thought or intent behind their actions. They follow the set of instructions given to them in the form of ones and zeros.
Mindlessly executing tasks is a good thing in factory robots, however. Industrial machines do what they’re told to do, and they do it well. They’ll assemble that part or move that load until they’re told to stop, something breaks, or the power runs out. But what if robots could wonder? What if their minds could wander, and they had a thirst for learning new things? That wouldn’t be good for manufacturing.
Making robots curious
Machine learning and artificial intelligence get a lot of attention from scientists, programmers, and researchers. The potential for well developed, intelligent machines is unfathomable. It’s difficult to really imagine what a truly artificially intelligent machine could accomplish. Of course, trying to recreate human-like intelligence in machines is easier said than done. So a couple of computer scientist got together to develop an algorithm that would put a fundamental aspect of human intelligence into machines.
Science Magazine recently published a story describing how two computer scientists developed an algorithm to teach robots how to be curious.
Todd Hester from Google DeepMind in London and Peter Stone from the University of Texas in Austin wrote a program that teaches robots to seek out new things and understand things it did not previously know.
The algorithm is called Targeted Exploration with Variance-And-Novelty-Intrinsic-Rewards (TEXPLORE-VENIR), and it uses reinforcement learning to try and get robots to pursue learning new, unknown things.
The robot tries new things leading towards an ultimate goal, and receives a reward for doing so. However, the robot also receives rewards for accomplishing internal goals even if the action doesn’t lend itself to the ultimate goal.
Basically, the robot is rewarded for discovering new things and reducing uncertainty (gaining familiarity with new things).
Curious robots need not apply in manufacturing
There’s definitely a need for robots to wonder and discover. For example, robots designed for space exploration, robots that need to adapt, or robots that are built for purposes that aren’t yet known can benefit from curiosity. Curious robots would be a bad thing for manufacturing, however.
The researchers notices that the algorithm can cause robots to become too distracted. Instead of successfully performing an assigned task, the robot was more focused on discovery. Imagine that your industrial robots decided to stop working because they wanted to explore some uncharted corner of your factory or ponder the meaning of life. We need factory robots to be exact, efficient, and reliable.
An algorithm that simulates curiosity is definitely a positive for robots that don’t yet know their purpose, or for robots that need to be able to do a lot of different tasks in changing environments. For a controlled factory environment, however, traditional industrial robots will do just fine.
Let us help you with your industrial robots. Call today for service, maintenance, or repair for your Indramat motion control system.