OpenAI's Dactyl System Gives Robots Humanlike Dexterity (venturebeat.com) 22
An anonymous reader quotes a report from VentureBeat: In a forthcoming paper ("Dexterous In-Hand Manipulation"), OpenAI researchers describe a system that uses a reinforcement model, where the AI [known as Dactyl] learns through trial and error, to direct robot hands in grasping and manipulating objects with state-of-the-art precision. All the more impressive, it was trained entirely digitally, in a computer simulation, and wasn't provided any human demonstrations by which to learn. The researchers used the MuJoCo physics engine to simulate a physical environment in which a real robot might operate, and Unity to render images for training a computer vision model to recognize poses. But this approach had its limitations, the team writes -- the simulation was merely a "rough approximation" of the physical setup, which made it "unlikely" to produce systems that would translate well to the real world. Their solution was to randomize aspects of the environment, like its physics (friction, gravity, joint limits, object dimensions, and more) and visual appearance (lighting conditions, hand and object poses, materials, and textures). This both reduced the likelihood of overfitting -- a phenomenon that occurs when a neural network learns noise in training data, negatively affecting its performance -- and increased the chances of producing an algorithm that would successfully choose actions based on real-world fingertip positions and object poses.
Next, the researchers trained the model -- a recurrent neural network -- with 384 machines, each with 16 CPU cores, allowing them to generate roughly two years of simulated experience per hour. After optimizing it on an eight-GPU PC, they moved onto the next step: training a convolutional neural network that would predict the position and orientation of objects in the robot's "hand" from three simulated camera images. Once the models were trained, it was onto validation tests. The researchers used a Shadow Dexterous Hand, a robotic hand with five fingers with a total of 24 degrees of freedom, mounted on an aluminum frame to manipulate objects. Two sets of cameras, meanwhile -- motion capture cameras as well as RGB cameras -- served as the system's eyes, allowing it to track the objects' rotation and orientation. In the first of two tests, the algorithms were tasked with reorienting a block labeled with letters of the alphabet. The team chose a random goal, and each time the AI achieved it, they selected a new one until the robot (1) dropped the block, (2) spent more than a minute manipulating the block, or (3) reached 50 successful rotations. In the second test, the block was swapped with an octagonal prism. The result? The models not only exhibited "unprecedented" performance, but naturally discovered types of grasps observed in humans, such as tripod (a grip that uses the thumb, index finger, and middle finger), prismatic (a grip in which the thumb and finger oppose each other), and tip pinch grip. They also learned how to pivot and slide the robot hand's fingers, and how to use gravitational, translational, and torsional forces to slot the object into the desired position.
Next, the researchers trained the model -- a recurrent neural network -- with 384 machines, each with 16 CPU cores, allowing them to generate roughly two years of simulated experience per hour. After optimizing it on an eight-GPU PC, they moved onto the next step: training a convolutional neural network that would predict the position and orientation of objects in the robot's "hand" from three simulated camera images. Once the models were trained, it was onto validation tests. The researchers used a Shadow Dexterous Hand, a robotic hand with five fingers with a total of 24 degrees of freedom, mounted on an aluminum frame to manipulate objects. Two sets of cameras, meanwhile -- motion capture cameras as well as RGB cameras -- served as the system's eyes, allowing it to track the objects' rotation and orientation. In the first of two tests, the algorithms were tasked with reorienting a block labeled with letters of the alphabet. The team chose a random goal, and each time the AI achieved it, they selected a new one until the robot (1) dropped the block, (2) spent more than a minute manipulating the block, or (3) reached 50 successful rotations. In the second test, the block was swapped with an octagonal prism. The result? The models not only exhibited "unprecedented" performance, but naturally discovered types of grasps observed in humans, such as tripod (a grip that uses the thumb, index finger, and middle finger), prismatic (a grip in which the thumb and finger oppose each other), and tip pinch grip. They also learned how to pivot and slide the robot hand's fingers, and how to use gravitational, translational, and torsional forces to slot the object into the desired position.
Re: (Score:1)
nearly perfect (Score:4, Funny)
Only two out of seven researchers had their penises ripped off for the "big finish" after an otherwise perfect handjob.
Re: (Score:3)
Note to fellow researchers: teach AI the words "jerk it off" never, EVER mean that literally.
Re: (Score:2)
There are so many ethical issues to think about with sexbots, but as ever I doubt anyone really will and it won't be until penises are ripped off that some effort is made.
It's not hard to imagine the problems people are going to experience with their sexbots. Forgot to charge it, and there is no lock-out for initiating a bondage session with less than 80% battery life. Was getting ridden when their 150kg BBW model decides to install a critical update and goes limp. Driven into poverty by sexbot addiction an
Re: (Score:2)
Click the first link in the summary, there's a video immediately after the article title of a real robot hand performing tasks.
No one is actually stopping you from seeing the video.
Re:Impressive (Score:4, Informative)
Their current method took a lot of work (CPU time, space) and at the moment is basically useless (and in many cases performs worse than the shown video according to tfa), but it seems there is a lot of potential for improved efficiency here. It's not going to develop to strong AI, but it might be possible to develop the technique to where it can dramatically improve robo soccer [youtube.com], for example.
Re: (Score:3)
They did it the really hard way. All they had to do, believe it or not, was give one set of robot hands whiskers. Straight up whiskers with sensors to monitor feedback from the whiskers. They are pretty neat, not only will you generate feedback prior to contact and thus refine contact methods but they also provide inertial feedback and even resistance feed back for motion through the air and of course position relative to gravity.
Production unit would probably benefit from whiskers, just much fewer of them
Re: (Score:2)
Re: (Score:2)
I'm sure that even with minimal training, this AI would also outperform you at comprehending a basic article summary.
how fast (Score:2)
Re: (Score:2)
It's all fun and games until you get hauled in front of Congress to testify about usage of Performance Enhancing Algorithms.
I Know Where This is Going (Score:2)
"I know kung fu..." - Terminator
Humanlike dexterity? (Score:2)
Re: (Score:2)
Open Source Applications? (Score:2)
could be awesome for prosthetics (Score:2)
Could be awesome for prosthetics.
A general command could be issued for the appendage, and the "AI" can fill in the gaps about how to execute the action.