ASIMO Reaches Pinnacle Of Sophistication, Now Plays Frogger
Writing by Evan Ackerman on Wednesday, 5 of August , 2009 at 4:21 am
Earlier this week we covered several different incremental improvements in robot AI, including grasping and object recognition. In the video above, ASIMO is demonstrating another (arguably more) important aspect of robot intelligence: the ability to navigate around a dynamically changing environment. It’s not likely that ASIMO will find itself in a situation where it needs to avoid stepping on whirling pink blades of death, but at some point (soon, please) we’ll have ASIMOs walking around our homes, and they’ll need to be able to not accidentally crush our toes/children/pets/Roombas.
Currently, it seems as though the system depends on a camera with a view of both objects and robot, as well as some fairly heavy off-board processing. Obviously, the hardware is all going to need to get integrated into the robot itself for it to be practical, but there’s no reason that this can’t be done, and there’s another video after the jump which shows an HRP-2 robot (a more primitive but less freaky version of these) planning navigation from on-board vision data.
VIA [ Plastic Pals ]
Comments (2)
Category: Androids, Artificial Intelligence, Research
- Add this post to
- Del.icio.us -
- Digg
Comment by Zool
Made Wednesday, 5 of August , 2009 at 7:22 am
This is a very deceptive and somewhat disappointing methodology. At first it looks very impressive. But it didn’t make sense Asimo wasn’t looking down at the floor as it moved. It turns out its an overhead camera that is being used to then compute way points and then the walking is being adapted to this way point (I guess spline) list. Its very much along the lines of how video game AI characters currently work (and have worked ever since Doom was created, as that was one of the first vector based worlds (Wolf3d was grid based, so easier for AI characters to move around in)).
The more I think about this methodology the more its not only disappointing me, its actually starting to feel almost annoyingly fraudulent, in that its implying more intelligence than is really used (unfortunately).
I would be much more impressed if Asimo was looking at the floor in stereo (working out its 3d world) rather than using a fixed mono overhead camera.
As it is, recomputing way point lists dynamically is something games have done for years and they also have to combine that with adaptive skeletal animations to avoid feet sliding etc..
Comment by Zool
Made Wednesday, 5 of August , 2009 at 7:48 am
I should just add, the HRP-2 work looks more impressive (I missed that video at first as I was focused on that Asimo work when I was previously posting (I only got to see that other HRP-2 video after I posted)). Unfortunately however, HRP-2 is still using an external motion tracking camera and its still using video games AI techniques. But anyway their work looks more impressive. At least its (fixed position) motion tracking camera is seeing in 3D. (I wish the robot was seeing in 3d but a moving camera would make it harder to workout the world).