In praise of less?
O?REILLY SOLID CONFERENCE: “Sometimes we like to have dumb droids which don’t have much initiative,” said Rodney Brooks, founder of Rethink Robotics, maker of the Baxter manufacturing robot. “And sometimes, maybe our machines have a little too much initiative.”
“How we’re going to interact with [intelligent machines], how we’re going to deal with them is different ? and it’s going to be scary for a lot of people as these machines come online,” said Brooks said he’s getting pushback from Rethink Robotics customers that don’t want intelligence and situational awareness in their industrial robots ? they just want dumb machines that will do exactly what they’re hard-coded to do, over and over and over again.
Intelligent machines, however, can be trained to do a variety of actions, even after installation and used for a different manufacturing chore. They can ?learn? various movements by being “shown” what to do through manual manipulation, or they can have software upgrades that can expand and extend their performance.
Are customers always right?
“But when robots start to have initiative,” he said, “sometimes the customers don’t like that so much.” As an example, he showed a video of a Baxter robot moving an object, but accidentally dropping it.
Knowing that it had dropped it, it didn’t continue the placement motion, but instead merely went back to pick up another object.
“Some of our customers hate that,” he said. “They hate having a machine that doesn’t do the same thing again and again.”
From Brooks’ point of view, an industrial robot should have the “common sense” to not complete a pointless action. But he believes it will take some time before we puny humans will tolerate such changes in behavior ? “initiative,” as he called it.
Rethink Robotics is working to develop machines that can take make intelligent decisions and take initiative, he said, “But I think it’s going to be awhile before more than the early adopters get used to that.”
After bemoaning the recalcitrance of the industrial-robot user base, Brooks went on to tick off four things that the robotic-development community should focus on that could jumpstart robot development and make people more comfortable sharing their world with their robotic brethren.
First on his list would be to upgrade visual object-recognition capabilities of robots to match that of a two-year-old child. “Computer vision has gotten fantastically better in the last 10 years ? Moore’s Law has been very, very good to it,” he said. “But we don’t recognize categories the same way a two-year-old child is good at recognizing categories.”
His next benchmark comparison was a four-year-old kid ? specifically, in terms of language-recognition. Children can easily pick out an individual voice in a noisy environment, he said, and handle accents and basic grammar. “They are much better at understanding language than any of our current systems,” he said.
Next up was a six-year-old, who has enough fine-motor dexterity to, for example, tie their own shoelaces. “A six-year-old child can do every single thing that a factory worker in China does in terms of manipulation,” he said. “Our robots right now are pathetic at dextrous manipulation.”
Brooks’ final example of the need for robotic improvement would be embodied, this time, in an eight-year-old child?social understanding. A robot should recognize us, know our range of capabilities, and treat as individuals as can an eight-year-old, he said.
“Every time you go to the ATM machine ? every day for the last five years ? it says, ‘Do you want to speak Spanish today?’ No. Stop asking that stupid question.” That ATM machine ? that robot ? should know who you are, what you know, and how to interact with you ? individually ? on a much more natural basis.
If these four goals are met, he said, “The machines of the future that have initiative are not so scary.”