The self driving car movement was a big hit. Billions of dollars pouring into startups and all the automotive and chip companies waking up suddently. Just unbelievable. The question naturally then is what is the next big thing in AI. A very promising area is autonomous robots. Research in top Universities and in research labs of tech giants are in full steam.
Take for example this ambitious project from Nvidia. At GTC 2017, they announed the Issac initiative. It is a platform to enable fast development of robots. You see, for a robot to learn a simple task such as walking, there are a few different ways. One, watch someone walk and try to copy. This is known as imitation learning in academic circles. This is hard for two reasons. First, this needs lots and lots of data, for each different terrain and for each position you were in before taking a stride. Second, a robot cannot copy our muscular movements directly, they obviously have a different build and the torques needed at joints to achive a desired motion need to be deteremined. The other option then is learning by trial and error.
Trial and error is how human babies learn for example. However, this approach takes a lot of time. We try and fall over several months before we figure out how to walk. But robots can’t wait that long. They need to be up and running as fast as possible and get to work. This is where virtual simulations come in. If you could create a virtual model ( one that exists only in simulations ) of the world, then you could train the robot in that virtual world. If the virtual model was good enough and realistic, then in theory, the trained model can be loaded into a real robotic hardware and it would be able to walk on day one. The platform offering by Nvidia is an attempt in this direction. It is not ready for primetime yet. Lot of work needs to be done to get anywhere near commercial uses of the platform. However, it is a great start and sign of where things are heading.
Pretty soon, platforms like that will be where our robotic cops will get their training from.
Here is a link to the announcement.