Not all of Elon Musk’s money goes towards Tesla, Solar City and SpaceX.
Musk is one of a set of Silicon Valley types who have invested in a company called OpenAI. OpenAI performs research into artificial intelligence, and part of that work includes reprogramming robots from Fetch Robotics.
Fetch Robotics builds hardware for warehouse automation. The OpenAI researchers are putting software into the robots from Fetch that will give them the chance to learn through trial and error.
The idea behind this is that innovations in software, as well as machine learning, will be better suited to giving robots new abilities, as opposed to advances in hardware. One robot can follow human workers around a warehouse and to pick up dropped items. The robots have a mobile base, 3-D depth scanners, a 2-D laser scanner, and an arm with seven degrees of motion.
Earlier this year, University of California, Berkley, professor Peter Abbeel joined OpenAI. Abbeel is an expert in robotic learning, and he is presenting how robots can, via the use of an approach of machine-learning called “deep reinforcement,” can acquire new abilities, such as being able to fold towels. These skills would be difficult to program into the robots by hand. Google’s DeepMind program has used the same approach to teach computers to play games at a level beyond human abilities.
The robots learn these new skills completely from scratch, thanks to a neural network that receives input from sensors while also controlling physical movement. The network can constantly adjust as needed to reach whatever goal has been programmed into it. So, for example, if a robot is learning how to fold towels, it can continually adjust its grip until it gets it right.
“If this goal can be achieved, then there will be economic and industrial benefits,” Marc Deisenroth, an industry leader on reinforcement learning at Imperial College London, told MIT Technology Review. “Imagine a Roomba not only cleaning your floor but also doing the dishes, ironing the shirts, cleaning the windows, preparing breakfast.”
Deisenroth predicts this could eventually drive the costs of robots down.
“Currently, the software seems to be the bottleneck,” he said. “However, independent of this, better hardware could also lead to substantial improvements.”
For example, researchers are working on elastic feet that work like a monkey’s, likely since they’d be better able to manipulate their environment.
A Japanese company, Fanuc, is testing this type of learning as a way to quickly train its industrial robots new tasks. This would save time. Google is working on something similar.
“Moving away from having to program robots by hand by endowing robots to learn autonomously is a key element for the future of robotics,” Jens Kober, a leader on robot learning at Delft University of Technology in the Netherlands, said. He also said that robots could be able to share information, thus saving time.
Industrial robots are nothing new, but with the exception of robots like the Roomba vacuum, household robots remain mostly in the realm of science fiction or futuristic comedies like The Jetsons. That’s because the constantly varying conditions of the typical household can throw robots for a loop. This new way of learning is meant to circumvent that.
The list of names behind OpenAI reads like a who’s who of Silicon Valley: Musk, Peter Thiel, Sam Altman (president of Y Combinator), and Y Combinator co-founder Jessica Livingston. OpenAI is a non-profit, and the investors have so far put $1 billion into the project, which is being led by Ilya Sutskever, a well-known artificial intelligence researcher who previously worked for Google, and Greg Brockman, who was one of the first employees at Stripe, a digital-payment company.
OpenAI has pledged to make the tech publically available, but it’s not hard to see how it could help Tesla in the assembly of its vehicles or components, such as the batteries being built at the gigafactory.
Or, perhaps, Tesla customers can spend a little extra to bring home a Tesla-branded robot to do their housework? A future in which some folks have a Model S in the garage and a Tesla housekeeper in the home doesn’t seem that far-fetched.