There seems to be some sort of revolution going on in the walking abilities of small humanoid robots. The why and how isn't exactly clear, but watch the videos and you will see what I mean.
When you look at the way many of the famous research robots walk - take Asimo as an example - then to say that they don't look human is an understatement. They have to be so careful not to fall over and break all that expensive hardware that "move slow and keep the center of gravity low" seems to be the order of the day. Smaller robots can fall over more safely and do less damage, but even though Nao say is a lovable character his walking is less than convincing.
When it comes to Japanese robots, however, they may not be so cute but it seems that natural walking comes, well, naturally. Our first example is from Dr. Guero (Masahiko Yamaguchi) who designs robot hobby kits and is already well known for creating robots that can ride bicycles. The video shows one of his KHR-3HV robots with some modifications do a good job of walking naturally.
Notice that the robot manages to walk without lowering its center of gravity but it keeps its balance by swinging its arms. The gait places the heel down on the ground first - heel strike - which is characteristic of most human walking, but not running.
It may only be walking on a flat surface but overall - impressive.
The next video comes with even less explanation. However, its title is V-Sido, which is software that makes a robot mimic the actions of an avatar on the screen. This suggests that the technique isn't specific to a particular robot. On the other hand, the form of the robot, i.e. the long legs, is distinct. If you watch the video you will see that the gait is very different. It almost looks like a super model on a fashion runway. Clearly stabilisation is by way of swinging the hips:
It all looks like fun and if you want to join in you can get a Kondo KHR-3HV for around $1800, which is very good compared to a Nao. So Japanese robots aren't cute, but they are cheap and very capable.
Yes, another neural network breakthrough. This one provides us with a way of training using reinforcement learning, but without the need for the biologically implausible and expensive back propagation [ ... ]
The results of the 2017 Stack Overflow survey are available. They cover many different aspects of the developer landscape and one that is of immediate interest is what we can expect to be paid and whe [ ... ]