Communications of the ACM
Dexterous Robotic Hands Manipulate Objects with Ease
The researchers used a model-free reinforcement learning algorithm with deep learning, and something called a teacher-student training method.
Credit: MIT CSAIL
Scientists at the Massachusetts Institute of Technology's Computer Science and Artificial Intelligence Laboratory (CSAIL) have designed a system to enable robotic hands to handle more than 2,000 different objects.
The researchers used a simulated hand with 24 degrees of freedom, and demonstrated that the framework could be adapted to a real robotic system.
The framework employs a model-free reinforcement learning algorithm that formulates value functions from environmental interactions, deep learning, and "teacher-student" training.
The "teacher" network is fed data about the object and robot in simulation, which it distills into observations similar to those found in the real world.
From MIT Computer Science and Artificial Intelligence Laboratory
View Full Article
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA
No entries found