Able to walk the steps by themselves.
△ safety rope is only used to prevent accidents
An accidental step in the air can also adjust the pace in time.
Up and down the slope is a leisurely step.
Can you believe that this bipedal robot named Cassie is actually a “blind” – without visual sensors, completely rely on the sense of touch to perceive the world of the kind.
This is a new study from Oregon State University. Researchers said.
As far as we know, this is the world’s first bipedal robot that can steadily climb stairs, or traverse similarly undulating terrain, by body perception alone.
Yes, not just up and down various high and low steps.
Cassie can also steadily leap over tree trunks that break off in the road like this one by feel.
Two feet in the plane of different heights, also does not affect the balance.
However, do not look at this guy now a look of ease, before the special training, it actually looks like this.
So, what kind of training did our friends at Oregon State University put the robot through?
The core method, still, is reinforcement learning based on LSTM networks.
The difference is that the Oregon State University researchers, added the effect of randomized terrain to the training.
That is, when the robot is trained in a simulated environment, it will encounter a stair-like changing terrain up or down, with variables such as the number of stair steps, the height, length, and width of each stair step, and the slope of the ground in front of and behind the stairs.
Specifically, the height of the stairs ranged from 10 cm to 21 cm, and the width ranged from 24 cm to 30 cm.
And, the multiple steps in the simulated environment are not uniform, i.e., each step is added into ±1cm error.
In addition, in order to effectively allow Cassie to apply the stair climbing strategies learned in the simulated environment to reality and prevent overfitting, the researchers also created various problems for it during the training phase.
For example, before the start of each training session, the robot’s joint damping, mass, offset, ground friction, and other dynamics parameters were randomized all at once, allowing the controller to try to face a variety of completely different environments.
Notably, during testing, the researchers observed that the robot learned a “counterintuitive” strategy: it performed better climbing stairs when it moved faster.
Jonah Siekmann, the paper’s first author, explains.
Because the robot is blind and cannot see its surroundings, it may land in a bad place.
For example, if it steps on the edge of the stairs, the force generated will push the robot down the stairs.
At low speeds, the robot does not have enough momentum of its own to overcome this backward push; but at high speeds, the robot will tend to jump over some steps, which in turn will allow it to find its balance again.
Seeing this, what do you think of Cassie’s performance?
Currently, this study has won the Robot Top Society RSS 2021.
Address of the paper: https://arxiv.org/abs/2105.08328
Posted by:CoinYuppie，Reprinted with attribution to:https://coinyuppie.com/all-about-foot-sense-the-robot-that-perceives-the-world-by-touch-can-climb-stairs-steadily-without-looking-at-the-road/
Coinyuppie is an open information publishing platform, all information provided is not related to the views and positions of coinyuppie, and does not constitute any investment and financial advice. Users are expected to carefully screen and prevent risks.