Google's (now Alphabet's) Boston Dynamics' Atlas Robot for DARPA wows, makes sci-fi feel like reality.
Chances are you may have already seen the new Agile Anthropomorphic Robot (Atlas) demonstration video since its already gotten over two million views in two days. For prospective, Atlas is the robot that the Defense Advance Research Project Agency (DARPA) gives away to contestants too poor to build their own in its Grand Challenges. Atlas is also used by the U.S. Army, the Navy and the Marines--its a tough dude.
One thing you may not know, is that Boston Dynamics was acquired by Google in 2013, hence they no longer pitch the press or even describe their newest innovations, but merely make YouTube videos about their results and refuse repeated requests for interviews.
What we do know is that the newest version of Atlas is a little less ominous at just five foot nine inch tall and 180 pounds--hefty for that height. Atlas can now operate indoors or outdoors, as demonstrated in the video where it slips and slides, but does not fall, on snowy unlevel ground. As usual it carries its own battery, but powers its manipulators with hydraulic actuators, making it plenty strong. It also has a new trick--it can get up if it falls (or is knocked over on purpose by a mean human with a hockey stick in the video).
Micro-electro-mechanical system (MEMS) sensors in its body and legs send orientation information to its processor which keeps it balanced. And it uses a range-mapping infrared laser-powered Light Detection And Ranging (LIDAR) system to sense obstacles, keep an eye out for potholes and generally manage its navigation and manipulation abilities. According to IEEE Spectrum, 3-D printing--an approach being used by other robot makers--was used to embed the hydraulic pathways and servo valves are inside the legs instead of adding them on with expensive components, thus lowering its cost (or increasing Google's margins--we could not find out which since the sales department is also not answering queries).
Boston Dynamics classic Atlas is a super-strong military-grade robot unlike the consumer trends toward co-robots that less intimidating and which can be safely cooperate and work alongside people without hurting them.
URL: http://www.bostondynamics.com/img/Atlas-x3c.lr.jpg (Source: Boston Dynamics)
Apparently, Google--and its new parent Alphabet--are allowing Boston Dynamics to further develop the dexterity, weight-to-strength ratio and cost-lowering strategies of 3-D printing on their own. As far as what Google has in mind for applications of Atlas and its other four-legged brethren, such as BigDog which refuses to be pushed over even by mean humans, remains to be seen since Google is mum.
Boston Dynamics classical leg (left) used external hydraulic hoses and servo-valves, that are build into the 3-D printed leg of the new Atlas (right) which lowers its cost--to Google--and makes it operate more reliably, according to Boston Dynamics.
(Source: Boston Dynamics, used with permission)
One aspect that these robots may or may not be developing is co-robot capabilities--the ability to move and work around humans without hurting them. One scene in the current video shows Atlas following a human, apparently to carry is pack or some such. However, the human provoker in the video who pushes Altas over and taunts it by moving the box it is trying to pick up out of range, uses a hockey stick to keep his distance from the robot, appearing afraid to get too close.
— R. Colin Johnson, Advanced Technology Editor, EE Times
It has lidar to sense obstacles and MEMS sensors to sense its orientation, but I('m not sure hoiw extensive its obstacle avoidance algorithms go. It has "reflex" algorithms in firmware for quick reactions that keep it upright and stable, but I believe it needs extensive specialized programming to properly operate in specific environments performing specific task. And you can tell its reflexes to not keep it from harming humans like a co-robot by the wyt that the guy with the hockey stick makes sure he is no where near Atlas when it is operating.
Agreed, the movement and stability is a great achievement in itself.
I also noticed that there were giant barcodes on the boxes and on the wall next to the door, as well as on the door.
That is another interesting point in terms of the control of the movement. Partially visual, distance control, on the other hand, plotted path through the forest. Can it recognize a tree, or another obstacle? So, of you put an obstacle suddenly in the way, does it move around it?
Is the control unit for that logic in the robot, or remote?
Yes, they have to specifically program tasks, but at a high level. For instance, his jaunt through the woods was plotted by a human, but the slipping and sliding to stary upright is part of his firmware. He can also pick up boxes--and chase them around for retrieval autonously from his firmware, but you have to program him to stack them on shelves and such--but at a high level (he performs the grasping and lifting and positioning from algorithms in his firmware. They intensionally made his actions appear all autonomous, like opening the door to take a walk in the fresh air--which a robot woulod never do on its own--in order to make his appear more human-like than he is. The important part of the video is his ability to stay upright and pick himsef up from fallling, which he previous generation could not do, but the rest depended on specific programming or path-plotting by a human. Still its a giant step forward from the old bigger clunkiret Atlas that was not that goog at unlevel ground and getting up after a fall.
- where the box is, and that it should pick it up?
- where it is supposed to go (though that door)?
- follow the human
- walk where to in the forest (goal)?
Are these programmed actions, or part of a bigger "plan"?
How do you tell the robot, what its tasks are, and what does it do, for how long?
When a human gets up in the morning, there is a number of habits and a larger plan to follow (you would certainly hope for, with most people) which is gradually changing and evolving over time, so how do you define the tasks for the robot, and within which larger context and logical plan to follow?