Remote robotic-data farms
Most robotics researchers start with some level of love for the hardware side of the work. The next big thing happening in robotics infrastructure is going to make it so us researchers no longer need to worry about all the hardware details — it is a series of technological developments allowing hardware papers to be written without going into the office. After a Ph.D. of on and off hardware development, I am sure most of my desire to work hands-on with hardware is a graduate-student-special version of Stockholm syndrome.
I have heard of multiple researchers at big technology companies that have published robotic learning work without physically going into the office. The technology that allows remote operation also gets most of the way towards robots that can run extra experiments on their own to collect more data for future tasks that may not yet be known. This may seem somewhat backwards: what is the extra data going to be used for? To date, most robotic learning research has a very tight link between a paper and a set of experiments, but breaking that one-to-one mapping is key to the field leveraging other advancements commercializing deep learning (mostly bigger data).
As most people who work in tech know remote work is the future, this can become a major perk that some companies leverage to keep talent. Anyways, I am okay with saying "hello" to the robots executing our experiments when we have our on-site workshops with our team - the rest of the time we'll be doing long-distance.1
