 |
| Artificial intelligence enables robots to control things |
A team of scientists from the Massachusetts Institute of Technology and Computer Science and the Artificial Intelligence Laboratory (CSAIL) has developed a system that will one day allow robots to have the same skills as young children - that is, to use their hands more intelligently.
The system can move more than 2,000 different objects with a single robot. This ability to handle everything from cups to cans of tuna helps hands quickly pick up objects and place them in specific poses and positions, and can even generalize invisible objects.
According to the team, this type of manual labor, often limited to a single task, can be a benefit that speeds up logistics and manufacturing, and helps with common needs, such as loading items into equipment locations or manipulating various tools.
The team used a simulated stereoscopic hand with 24 degrees of freedom and showed that the system could be implanted into a real robotic system in the future.
Parallel couplings are often used in industry, also for ease of control. But it cannot handle many of the devices we see in our daily lives. Even using a clutch is difficult for robots.
The team's system allows multi-fingered hands to manipulate these tools in ingenious ways, opening up new areas of robotic applications.
The system manages more than 2000 projects
Using non-paradigm reinforcement learning algorithms (which means the system must learn value functions from interactions with the environment) and deep learning, the team developed a 3D manual simulation that can process more than 2,000 objects.
Also, the system doesn't need to know what's going to be captured to find a way to move it around in its hands. But the system is not yet ready for use in the real world.
The team has to turn him into a real robot. Earlier this year, researchers from Zhejiang University and the University of Edinburgh successfully applied methods to improve artificial intelligence in their robot dogs. The system enables the robot to walk independently and recover from falls.
The new system can handle a lot of little things, including things like tennis and apples, with a success rate close to 100%. But tools such as screwdrivers and scissors presented challenges.
With this creature, the success rate is closer to 30%. However, the potential of this system is enormous. One day, it could make robots more efficient at handling tools and doing tasks like packing boxes more efficiently.
It should be noted that experiments in this area did not stop because OpenAI used a manipulator to experiment with Dactyl and solve the Rubik's cube software, a step towards a more general artificial intelligence, derived from popular thought becomes one task. . . . DeepMind has developed RGB, a vision-based system that can challenge robots to learn to grab and stack objects.