Scientists at Columbia Engineering mark a paramount development in robotics. They were able to create a self-repairing robot that is also self-aware.
A robot that is self-aware means that it can learn what it is from scratch, without having previous knowledge of physics, or geometry, or even motor dynamics, the researchers revealed.
As explained by the researchers, the robot did not have an idea if it is an animal or a body part or what shape it takes. However, in just about a day of in-depth computing, or what researchers referred to as a brief "babbling" period, the robot initiated "self-simulation."
Using the self-simulators, the robot was able to reflect and adjust to various circumstances similar to humans. It can take on the things it needs to do and interestingly, it knows whenever it has incurred damage on its body and then, repair itself.
Prior to the invention of the self-repairing and self-aware robot, the robots existing today have not yet been able to create self-simulation, just like humans do.
"But if we want robots to become independent, to adapt quickly to scenarios unforeseen by their creators, then it's essential that they learn to simulate themselves," said Hod Lipson, a professor of mechanical engineering.
Lipson, who is also the director of the Creative Machines lab where the research was conducted, collaborated with Ph.D. student Robert Kwiatkowski, the lead author of the study.
Use Of Deep Learning
The four-degree-of-freedom articulated robotic arm, which was used for this study, applied deep learning, an advanced machine learning technique, to create a self-model. Although the first ones were inaccurate and the robot did not learn yet what it was, there was some progress after around 35 hours of training.
For example, the self-model robot showed a 44 percent success rate during a pick-and-place task in an open-loop system. According to Kwiatkowski, it can be compared to trying to pick up a glass of water with one's eyes shut — truly a difficult task, even for humans to accomplish.
The researchers tested the self-aware robot to further perform other tasks such as writing text using a marker. Using a 3d-printed deformed part to simulate the damage, the self-model robot was also able to detect the self-damage and retrain itself.
"This is perhaps what a newborn child does in its crib, as it learns what it is," said Lipson, who is also a member of the Data Science Institute.
While the robot's ability to imagine itself is primitive yet vis-à-vis humans, Lipson believes it could already be a route to machine self-awareness.
Both Lipson and Kwiatkowski understand that there could be ethical implications to stem out this major advance in robotics. They warned, however, that their invention is a "powerful technology," but it should as well be "handled with care."
This work, published on Wednesday, Jan. 30, is discussed in full in the journal Science Robotics.