Question Considering Walking/Running Gate

[rudukai13]

So I've got a quick question. While looking at all the videos of the humanoid robots out there, I noticed a common trend. When humans walk, we lean our upper body forward slightly, moving our center of gravity to a point where we are no longer stable, and catching ourselves by taking a step forward. In order to go faster, such as a run, we lean farther worward. This is the same basic idea that Segways use to stay balnced-the center of gravity moves, the contact patch needs to move back underneath it to keep from falling over, and it rolls forward, in a perpetual fall. What I have noticed however, is that many humanoid robots tend to take a step forward, lean back, and then on a consecutive step over-compensate and lean too far forward, falling over. So, I guess my question is, why don't people program their robot to lean forward and then take a step? This would provide a much more stable gait, and would allow robots to move more quickly while running.

[sthmck]

The type of walking that you are referring to is called dynamic walking. It basically uses principles of physics to produce a very fluid motion. The problem is that with dynamic walking there is not definite stride length. Weight transfer like you said has a lot to do with it. I am in the process of writing a paper about dynamic walking. No my paper isn't very technical most of the physics is a little beyond me right now. i was basically introducing the idea of dynamic walking as a viable alternative to what is being done right now. http://www-personal.umich.edu/~shc/robots.html This is the first site that I came up with after running a search. Has some pretty good stuff on it. As to why dynamic walking isn't being used on the hobby level I couldn't tell you. Its possible that someone is working on something along these line. I know that Virginia Tech has a hobby sized humanoid that is supposed to be dynamic, but other than that I don't know.

[rudukai13]

Thanks for you help!

[limor]

Hi, that page refers to passive-dynamic walking which is not what humans do.
Dynamic walking involves processing thousands of inputs every second from different sensors, predicting what the sensory input is going to be in the short future and then actuating the muscles to shift our weight and optimize our locomotion objectives (running, walking, jumping..) based on this prediction..

Currently the hobby humanoids use "open loop control" which means they are not aware of what is going on around them. The servos are actuated blindly and if the environment is static, the gaits will repeat the locomotion sequence as they were intended. Gyros are a they only thing that breaks this paradigm but ultimately where many hobbyists and researcher are going is towards analyzing every servo position and torque 100 times per second and doing real dynamic walking.

The Bioloid servos with their Dynamixel protocol, and Hitec/Kondo with their HMI protocol allow to query the position and torque and on-board computation is becoming more reasonably priced (gumstix) so some of us with alternative controller boards are not far from having a full closed loop control of the robots.

[sthmck]

I understand that the website I listed is related to passive dynamic walking. While passive dynamic walking is very basic, the simple dynamic concepts are the same. Yes there is basically no input for passive dynamic walking. Although I would consider gravitational force to be an input. I mentioned dynamic walking because that is basically what rudukai13
was describing. Thanks, sorry for this misunderstanding.