Advanced Machine Learning Company Osaro Captures $3.3 Million
Osaro, a new, San Francisco-based startup that’s developing advanced machine learning known as deep reinforcement learning, has raised $3.3 million in seed funding to take its technology to market.
The money comes from Scott Banister, Jerry Yang’s AME Cloud Ventures, and Peter Thiel, for whom Osaro cofounder Derik Pridmore once worked (first at Thiel’s hedge fund, Clarium Capital, and later as a principal at his venture firm, Founders Fund).
Why it’s interesting: Osaro’s machine intelligence software combines perception (which we’ve seen plenty of in the past, including with image identification) with decision-making abilities to help computer and robotic systems act more efficiently and intelligently (which we’ve seen more rarely).
We talked with Pridmore yesterday to learn more about the nine-person startup. Our chat has been edited for length.
TC: You don’t have any customers yet, but one of the commercial areas you’re exploring is industrial robotics. Why should these companies pay attention to Osaro?
DP: First, industrial robots aren’t what the average person thinks they are. They don’t have brains. They really just have very precise motors that are controlled by software. But every time you need to do something new, it needs to be reprogrammed. The robots aren’t flexible.
That trend used to work, but what used to take a year to make now takes a month [in many cases], and if it takes a month to set up a robot and the robot runs for just one month, that’s inefficient. [Our technology] enables you to grab a robot and train it a few times and let it start training itself from there. You only have to tell it when it’s successful or it fails by giving it a score.
And we do think there’s a huge opening for companies to [employ Osaro] in manufacturing, but there are other, virtual, things like marketing platforms where it could also be used.
TC: Do you have beta customers we could name for readers?
DP: We’re working on a pilot right now. These are complex solutions, so we need to partner with people to show that we can handle technical requests that others can’t.
What’s the business model?
DP: Our software is like an operating system and we’ll license it.
TC: Is there a systems integration component? Who will be training this operating system at companies that you hope to transform into customers?
DP: Our end goal is to make it so a low-skilled technicians can set up and train one of these [robots], but that won’t happen any time soon. For now, we’ll deploy it in niche applications where there really is no other solution and we’ll build out from there.
TC: Will you learn and improve on the product remotely?
DP: That’s the super long term goal. If we have a brain in a Foxconn factory and we’re remotely monitoring the agent on particular robot, we can learn from that robot and make another robot in another factory smarter if we can get enough data.
TC: If you’ll forgive our asking, are you at all worried about the singularity?
DP: I have my own theories about how this will progress. As technologists, we have choices to make about whether we want technologies to make people’s lives better or something else. Obviously Google wants to help, but it’s not explicit about it. For example, you could tell a machine to produce paper clips, but if you don’t give it a number, it might endlessly produce paper clips. So some of those mission statements need to be more explicit. Robots take their goals from people.
This entry passed through the Full-Text RSS service – if this is your content and you’re reading it on someone else’s site, please read the FAQ at fivefilters.org/content-only/faq.php#publishers.
Read the article: