Atlas, a humanoid robot 1.5 m tall, made the headlines in 2013 as one of the most advanced humanoid robot ever developed. It weighted 75kg, could carry 11kg of groceries (or anything else!), had a stereo vision (using Lidar) and 28 joints providing flexibility in its movement and the capability to manipulate objects. The first version weighted 150kg, the second 82kg. he decrease in weight has been made possible by using 3D printers to produce most of its parts.
What made quite an impression was its sense of equilibrium. Even if pushed around (see in the photo an engineer trying to make life difficult for Atlas as it was picking up a box) it could manage to keep its equilibrium and it was even capable of doing some hops and acrobatics.
To provide power to its brain and hydraulic muscles it was connected to a power line although a small battery could provide some very limited autonomy.
The new version, just announced, see the clip below, has learnt to run in the open space and most importantly the power required has been significantly reduced so that it can operate for some (still limited) time on its own.
Technology has made incredible progress although in the “efficiency” area is still way behind Nature. We can move around for the whole day with the power provided by a few sandwiches, a coffee surely helps, that is a thousand calories (2,000 calories, the average amount needed by a woman per day, corresponding roughly to 2kW hour) is enough fuel for the day. Not so for a robot. It takes much more power and that constrains its operation away from the mains.
The progress has been impressive. Take a look at the video and you will notice that it is not just running around, it is observing the lay of the land. It might seem strange to see Atlas stopping in front of a fallen tree, pondering what to do and then deciding to hop. For a 4 year old kid that would be a matter of a second, not requiring to stop the run at all. It shows that what we are taking for granted as “easy” is not easy at all, it requires a lot of computation for a robot. So in a way we can be happy that the distance between “them” and “us” is still significant. On the other hand they are quickly catching up.
Robots are becoming more and more autonomous, in doing activities and most importantly in evaluating the context and tailoring their activities to the context to achieve a certain goal. So far we are the ones setting the goal but this is also changing rapidly. More and more we are training robot to achieve an objective, no longer in ways to achieve the objective. We have seen that with the (soft) robot playing Go: it worked out the strategy to win by itself, actually taking some actions that the human experts would not have considered.
Autonomous systems are more and more designed to self learn. By doing that we are empowering them … and we might be losing control. Some sort of symbiotic relationship is needed so that we can remain part of the game. This is an issued being addressed by the IEEE FDC Symbiotic Autonomous Systems Initiative.