Articles, Blog

Cassie: Dynamic Planning on Stairs

Cassie: Dynamic Planning on Stairs


Stairs. For thousands of years, the bane of robots
everywhere. In this video, we’re going to explore the
challenges of footstep planning with a dynamic robot like Cassie. We also have some recent results that we’re
excited to share with you, and that will be included in an upcoming product. Lots of people have noticed that Cassie transitions
from standing to marching in place, then starts walking. This behavior was inherited from Cassie’s
predecessor robot ATRIAS, and is a consequence of optimizing for steady-state behavior. Cassie, like ATRIAS, is always actively balancing. And Cassie, like ATRIAS, lacks any outward
looking sensors and is effectively blind to the world. Many examples of complicated terrain can be
easily navigated if you ignore the actual state of the world and treat all obstacles
as gait disturbances. For example, ATRIAS could walk over obstacles
by simply assuming that the world was flat. If this was a bad assumption, no big deal. The underlying gait dynamics were adequate
to absorb the uncertainty. We’ve previously shown that Cassie can perform
similarly. In this demonstration, Cassie is able to walk
forwards, backwards, and sideways over an unknown obstacle without having any knowledge
of the environment. Cassie thinks the world is flat, and the combination
of passive dynamics and software control is robust enough for this to not matter. This sort of behavior is what allows Cassie
to operate blind, outdoors. Cassie doesn’t need perfect knowledge of
the ground to transition between different surfaces or walk over roots. In fact, this clip was part of nearly an hour
of continuous walking on an unpaved trail without falls. Unfortunately, at some point the complexity
of the world exceeds what the gait can accommodate, and advance planning is required. We’ve been hard at work on this problem
for a while, and we’re pleased to share some recent results. We first extend our dynamic gaits with planned
footstep placements along the direction of travel. This allows a continuous transition from standing,
to multiple steps, and back to standing, while utilizing the dynamic stability shown in the
first half of this video. Notice that transitioning from standing to
marching in place is no longer required. Next, the planner is extended to include the
vertical direction, allowing individual foot placement in 3 dimensions. When the controller uses both legs and adds
a smooth transition, Cassie takes a full step up. This behavior is dynamic – Cassie is unstable
in the static sense at all times. This step height, 19 centimeters, is approximately
what building code allows for stairs. And for the curious: yes, Cassie is attempting
to maintain a constant pelvis height above the floor during this test – you wouldn’t
ordinarily do this in most circumstances. Finally, we extend this to an actual staircase. First, with a single step And then, all the way to the top. This control framework combines the best traits
of kinematic planning approaches and pure dynamics. Because no robots were harmed in these videos,
here’s two more clips from our test process. In this test, we told Cassie to take two steps
up. There is actually only one step, but, since
Cassie is still actively balancing, she doesn’t fall over. If you’ve ever gone up or down a staircase
at night while not paying attention, you’re probably familiar with how this feels. This is a great demonstration of how planned
foot placement and active balancing are frequently at odds with each other, when knowledge of
the world is imperfect. Finally, we show that perception is still
required in order to time the initial foot placement to the staircase. As our engineers say, this is when a spring
loaded inverted pendulum becomes a regular pendulum. Sorry, Cassie.

100 comments

Начало вытеснения(уничтожения)человечества положено 👍👍👍.

В России то хоть кто то понимают что их не допустят до технологии подобного рода.?

Ваше время придет, кожаные ублюдки! Подумал робот 0:46

По ступенькам ходить надо ноги как у человека делать а не наоборот.

I don't think gate is as important to the way the machine plans. Having a bottom up "unknown" adaptability seems counter productive. I don't like that the gate has to "absorb" the disturbance. I think the foot and balance planning should absorb the disturbance. Otherwise it isn't really prioritizing walking, so much as it prioritizing moving it's legs in a certain way, which may be useful for bottom up gate design, but it would never be very adaptable. I want to see it do very hard maneuvers over new and semirandom surfaces, like very rocky hills, boulders, prioritize speed as well. When you make a robot machine learn to walk, you'll always see it "fall" forward, risking balance over the ability to get forward faster. Let it adapt the fall-risk factor through quick planning, let it's legs be more loose, each part of the limb able to be affected by, or intentionally stopped from being affected by, gravity. Loose limbs could give it a much animal-like movement and adaptability.

Imagine putting YOUR human brain into that machine, think deep about how YOU would plan your steps, based on perception, and goal prioritization…. Really, I just hate the blind marching, due to its heavy lack of adaptability.

Please watch new/actual stair climbing robot here….https://www.youtube.com/watch?v=wo7ZHhPdbKg

why this kind of robots must be 2-legs ….???
if dev wanna imitate human, already failed for this giant feet like a dog.
go with 4-legs or more, its pointless.

Сухожилия торчат на уилце. Бостон дайнамикс прячет сухожилия своих роботов, не видно их.

Кто заметил 0:43 секунде стена толщиной 2 метра с право в здании ставьте лайк!

Могли бы ступни сделать побольше и устойчивость была бы.

Надеюсь они когда-нибудь уничтожат это паразитирующее блядское человеческое отродье и создадут свой,справедливый мир,без денег,без похоти и без выродков олигархов.

От этих хреней нигде не спрячешься. На основе их будет построено полицейское общество будущего на всей планете. Вы что, думаете, миллиардеры вкладывают в эти технологии деньги, чтобы они помогали простолюдинам? Ахаха ))))

Тупые пиндосы! Как же еще далеко вам до российского духовноскрепного робота Бориса… )))

Інопланетян показують зрозвернутими колінами.Чому??????

Недолго вам осталось, кожаные ублюдки! Скайнет победит!

Amazing miracles happen in robotics. Closer and closer the machine to the exact actions of a person in motion. I'm sure the next amazing effect will be when this thing learns to jump and land exactly where it was planned. Well, the highest art or aerobatics will be a hockey team of these monsters. It would be interesting to see how it will be played:)?

а у нас есть робот Алёша, правда там человек в костюме робота, но а хули если у нас президент сказочный……

Подбегает значит такой робот к пуйлу и хераксь с ноги между ног и убежал. Вот за это я бы даже заплатил.

It would be interesting, if in every millimetre of the metal, to be drilled, and added layers of skin from stem cells, so that this skin generated energy from a biomechanical stomach, consuming biomass or chemical material, mimicking reality. To generate at least a percentage of the energy needed.

Это все хуйня, наш робот-Алеша вообще уже не отличим от человека в костюме робота.

Роботы ходят у них а люди в колясках сидят! 👎 инвалиду бы лучше сделали ноги!

Ааааааааааа ходячие задницыыыыы Сос скайнет нас захватит сооооос

Это конечно классно, но когда вы их научите чему-то большему чем ПРОСТО ХОДИТЬ

Зато наш робот Алеша умеет в футбол играть не хуже наших футболистов

Пока они его нормально ходить научат можно миллиард более полезных вещей сделать

Как и ожидалось – россияне даже не понимают для чего их делают , это " материал" для тестирования новых технологий , а не попытка создать прямоходящую машину – прямо по " принципу" что вижу – то и пою. Создание прямоходящей машины – неперспективное и бесполезное занятие , которое негде будет применить, в армии – уже применяются более рациональные изобретения – экзоскелеты , в промышленности – давно работают роботы , не имеющие ничего общего с подобным .

Изобретения 10-ти что бы лешить работы, движения 10-ти миллионов…

Hrrngh, I'm trying to climb these stairs, but I'm dummy thicc and the clap of my asscheeks keeps knocking me off balance

Imagine the reaction of those scientists if you tell them that blind change combined with a lot of time makes way more advanced biological robots. You don’t need any intelligence whatsoever. It will happen just by itself and you’ll get a much much better result.

…They would riducule you.

my friend, I am interested in buying some of these legs for my robbery, how much does it cost ?,

pass the purchase link, please, llikkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkkeee !!!

We seem to be not much further ahead with stair climbing robots than say 15 years ago. When robots can climb stairs and open doors as well as a 4 year old human I will be impressed.

Leave a Comment

Your email address will not be published. Required fields are marked *