CHANNELS

Aerospace

Automotive

Automated & Autonomous Vehicles

Battery & Electrification Technology

Defense

Drone TV

Electronics

Manufacturing & Prototyping

Materials

Medical

Motion Control

RF & Microwave Technology

Robotics & Automation

Sensors & IoT

Test & Measurement

Aerospace
Automotive
Automated & Autonomous Vehicles
Battery & Electrification Technology
Defense
Drone TV
Electronics
Manufacturing & Prototyping
Materials
Medical
Motion Control
RF & Microwave Technology
Robotics & Automation
Sensors & IoT
Test & Measurement
Aerospace & Defense
Search Results

Embodied Intelligence in the Hands of Robots

The future of manufacturing calls for the need of dexterous robotic manipulation — see how researchers at Columbia University focus on a new approach of "embodied intelligence" in robots.

“Dexterous robotic manipulation is needed now in fields such as manufacturing and logistics, and is one of the technologies that, in the longer term, are needed to enable personal robotic assistance in other areas, such as healthcare or service domains,” notes Professor Matei Ciocarlie.

Topics:
Manufacturing & Prototyping Robotics

More From SAE Media Group

    Transcript

    00:00:09 [Music] we focus our work on the  concept of embodied intelligence   so we have been building new tactile fingers for  robot hands and most people think of intelligence   as residing exclusively in the computer  if it's artificial intelligence or in the   brain if it's human intelligence but that's not  necessarily the case you know we are roboticists   and we're all about physical tasks in the real  world and when we do that intelligence no longer   should be confined just to the realm of the bits  the intelligent agent has a body it's a robot   and in particular we've been focusing on tactile  sense tactile sensing is key for embodied   physical agents interacting with the world if you  think about the human hand a marvel of engineering   more than 25 joints individual degrees of freedom  more than 40 muscles articulating those joints but  

    00:01:11 then of course also the sensing mechanism we have  hundreds of individual mechanoreceptors in each   fingertip that are constantly collecting different  kinds of information about touch the computational   intelligence the mechanical intelligence the  sensing intelligence in biological evolution   all of these components evolve in lockstep  together [Music] we've been asking can we optimize   the mechanical design of a robot at the same time  as the computational policy with and for each   other and this is where we come to the big big  advances that we've seen in the last five years in   the fields of machine learning and reinforcement  learning some of these modern reinforcement   learning techniques are really really well  suited for operating in situations where you have   imperfect knowledge of the world you can think of  it almost as a muscle memory right every time you   hit a tennis ball you don't have perfect knowledge  of the ball's exact velocity and spin you get your  

    00:02:24 sensor data which tells you something about  the world and then your muscle memory takes   over and says you know last time the word looked  like this these are the actions that i took and   things turned out very well in deep reinforcement  learning this takes the shape of a neural network   the input to the neural network consists of  sensor information and then the output consists   of commands to your motors and our sensors  are designed from the ground up knowing that   they will be used in conjunction with machine  learning and reinforcement learning methods   to mine it for the information that's  relevant to manipulation [Laughter] [Music] and all of those can be designed to react  appropriately to unforeseen circumstances   which is the definition of intelligence  right to do the right thing   when you are faced with a situation  that hasn't been predicted

    00:03:27 what we'd like to do is move away from the  paradigm where one group designs the mechanism   and then kind of throws it over the wall and  says hey computational people you take it from   there the mechanism looks about right it should be  able to do everything it's just a software problem   from now on and then you get the software people  saying oh it's the mechanism that's letting us   down our software is very intelligent what we're  saying is that if you think about them as a whole   that's when we're going to see a very  noticeable improvement in performance   because when you talk about embodied intelligent  agents there is really no choice but to look   at the mechanism the sensing suite and the  computational aspects all of them together [Music] you