AI & Humanoids

Poppy, a 3D-printed humanoid robot that defies conventions

Poppy, a 3D-printed humanoid robot that defies conventions
The assembled Poppy robot (Photo: Inria / H. Raguet)
The assembled Poppy robot (Photo: Inria / H. Raguet)
View 10 Images
Blueprints show that Poppy's structure more closely matches up to a person than to robots of a similar size (Image: Inria / H. Raguet)
1/10
Blueprints show that Poppy's structure more closely matches up to a person than to robots of a similar size (Image: Inria / H. Raguet)
Poppy's 3D-printed parts alongside consumer electronics components (Photo: Inria / H. Raguet)
2/10
Poppy's 3D-printed parts alongside consumer electronics components (Photo: Inria / H. Raguet)
The assembled Poppy robot (Photo: Inria / H. Raguet)
3/10
The assembled Poppy robot (Photo: Inria / H. Raguet)
Poppy's feet are small and have a thin strip that bends like our toes (Photo: Inria / H. Raguet)
4/10
Poppy's feet are small and have a thin strip that bends like our toes (Photo: Inria / H. Raguet)
Poppy's feet are so small they fit into children's shoes, which are then covered in pressure sensors (Photo: Inria / H. Raguet)
5/10
Poppy's feet are so small they fit into children's shoes, which are then covered in pressure sensors (Photo: Inria / H. Raguet)
Poppy's head contains a pair of HD cameras as well as a 4-inch LCD screen for displaying facial expressions (Photo: Inria / H. Raguet)
6/10
Poppy's head contains a pair of HD cameras as well as a 4-inch LCD screen for displaying facial expressions (Photo: Inria / H. Raguet)
Poppy's articulated spine gives it added flexibility when physically interacting with people (Photo: Inria / H. Raguet)
7/10
Poppy's articulated spine gives it added flexibility when physically interacting with people (Photo: Inria / H. Raguet)
Poppy's knees contain springs that help to keep the legs straight while walking (Photo: Inria / H. Raguet)
8/10
Poppy's knees contain springs that help to keep the legs straight while walking (Photo: Inria / H. Raguet)
Here you can see the dramatic difference in lateral movement between the Poppy with bent thighs (left) and straight thighs (right) (Photo: Inria / H. Raguet)
9/10
Here you can see the dramatic difference in lateral movement between the Poppy with bent thighs (left) and straight thighs (right) (Photo: Inria / H. Raguet)
The assembled Poppy robot (Photo: Inria / H. Raguet)
10/10
The assembled Poppy robot (Photo: Inria / H. Raguet)
View gallery - 10 images

A new 3D-printed robot called Poppy is helping a team of French researchers study bipedal walking and human-robot interaction. They were able to design, fabricate, and assemble a relatively large robot for around €8,000 (US$11,000) including servo motors and electronics. That's about a third the cost of commercial robots in the same size category like the RQ-TITAN, and is still cheaper than smaller humanoids like the Aldebaran Robotics NAO. And best of all, they plan to make their design open source.

One of the ways they managed to keep the cost down is by using lightweight materials, which means the robot requires less powerful (and cheaper) servo motors. Standing 84 cm (33 in) tall, Poppy weighs in at just 3.5 kg (7.7 lb). The Sony QRIO, by comparison, was 26 cm (10 in) shorter yet it weighed twice as much. Still, most of the cost lies in the robot's 25 servo motors: it utilizes 21 high-end Robotis Dynamixel MX-28s, two MX-64s, and two AX-12s. It's powered by a Raspberry Pi, and is equipped with 16 force-sensing resistors, two HD cameras, a stereo microphone, and an inertial measurement unit. Poppy's "face" is an LCD screen which can be used to show emotions (or to debug).

Making a more human robot

One of the main reasons the INRIA Flowers (FLOWing Epigenetic Robots and Systems) team opted to build its own robot is because none of the available commercial kits are truly biologically inspired. By rapidly-prototyping their own robot, Poppy could challenge some of the usual robot design conventions.

To begin with, it has an articulated spine with five motors – almost unheard of in robots of this size, but one of the ongoing topics at INRIA Flowers since its first humanoid, ACROBAN, from several years ago. The spine not only allows Poppy to move more naturally, but helps to balance the robot by adjusting its posture. The added flexibility also helps when physically interacting with the robot, such as guiding it by its hands, which is currently required to help the robot walk.

Looking at the knees, you'll see some springs spanning the upper and lower leg joints. The tension in the springs helps to keep the supporting leg straight during each step without motorization. And farther down, its feet are smaller than most robots of Poppy's size, and its toes are thin, allowing them to bend. Rather than planting each foot parallel to the ground, which is how most robots this size walk, the toes help the robot achieve "heel to toe" walking. And as for those children's shoes Poppy wears? They're equipped with five pressure sensors on each sole, which provide useful data.

One of the more obvious deviations in design can be seen in its upper legs, which bend inwards at an angle of six degrees. Despite the fact that this more closely models the human femur, most humanoid robot designs have opted for straight leg linkages. By bending the thighs in, the distance between the two feet is shortened, moving the supporting leg's foot closer to the robot's center of gravity. And that makes it more stable when standing on one leg and when walking.

Here you can see the dramatic difference in lateral movement between the Poppy with bent thighs (left) and straight thighs (right) (Photo: Inria / H. Raguet)
Here you can see the dramatic difference in lateral movement between the Poppy with bent thighs (left) and straight thighs (right) (Photo: Inria / H. Raguet)

The INRIA team produced two versions of Poppy: one with straight thighs and another with the bent ones. Experiments showed that the robot with bent thighs swayed far less during its walking gait, making it much more stable. However, the robot still can't balance on its own, so for now it needs a human trainer.

In the future, the team hopes to get Poppy walking on its own, and plans to share its designs with other labs to promote more biologically-inspired humanoid robot designs.

You can see Poppy in the following video.

Sources: INRIA Flowers, Poppy Project via 3ders

Poppy humanoid beta Overview

View gallery - 10 images
8 comments
8 comments
The Skud
Clever! I wonder if the team have considered using model drone/helicopter sensors and gyroscopic control along with the present spine system to keep the little chap upright? Some of the chips released recently are down to the size of a camera storage card. Sensing "sway" and adjusting posture would be almost automatic.
Robotbling
@Skud I believe that's what the inertial measurement unit will be used for, they just haven't implemented the control algorithm for it yet.
Thomas Bollinger
I am by no means an expert at all, but my spontaneous reaction is: A Robot with a cable? C'mon, you gotta be kidding. Anybody else?
Manke
@Thomas Bollinger This is standard for prototype robots, and is because, (i guess - no expert), they haven't implemented a battery yet :)
Anonymous756
"Poppy, a 3D-printed humanoid robot..."
This is simply not a true statement. Poppy is *not* a 3D-printed robot.
The plastic parts are certainly 3D printed. The motors, wires, cables, control circuits, metal bits and pieces are bought from traditional manufacturers. In other words, 95% of the complexity. The 3D printing is presently confined to the plastic bits. Still useful, but let's not violate the basic facts with headlines so poorly worded that they're false.
Anyone with any understanding of modern technology already knows this, but the utterly naive amongst us seem to actually believe that the robot exits the 3D printer fully assembled. Seriously.
Hit Or Miss
Interesting. As an inducement or alternative to parenting, it wins the cute vote, the 'helplessness' of aided walking and the like makes it clear that this is a child simulant, even before the scaling and big-eye open featured 'look'.
That said, the achievement of reduced weight comes at a penalty in structural effects design as you are essentially looking at a truss system which is going to constrain several design freedoms as both volumetric efficiencies in mounting larger or alternate actuators within the box, thrust as load lines for linearizing stress and keeping total displacement low. And of course the installation of higher kinematic pairings as sculpted joint pairs rather than simple revolute/cylindrical etc. systems.
In this we don't want to fight weight through miniaturization. We want to design system volumes by expected loads. A labor robot that can carry a 25lb load or 150lbs in combination with other robots for instance has to be designed around expected torque differentials that go beyond natural balance.
It's such a shame that we assume 'education, art and experimental' designs have to be functionally isolated from developmental approaches which yield practical applications.
Matthew Tedder
I am not sure what the extra back-bone flexibility provides over one forward-backward and a side-to-side swivel, other than appearing more natural.
However, the bending toes is good. My own latest design using springs to push-back on the toes.
One aspect missed, as per the gait, is the ability to wiggle the butt. Try sitting it on a chair, from the side. This also makes the gait look less natural when walking. To solve this problem, you could change the pelvis such that each side swivels forward and backward on the extremity. For example walking, the right hip swivels forward as the left hip swivel backward--this keeps the body facing forward straight as it walks forward.
Every humanoid design I've seen misses this ability.. although 3D animators tend to be familiar with the need.
JT DragonFlash
I love the design, reminds me of a toddler learning to walk but I can't help noticing the springs you are using as tendons to emulate tension on the joints when not in powered state. I would think the use of elastic would be more precise when it comes to getting the walking to be more controllable. Kind of like our muscles and tendons stretch and flex the use of elastic strands may help you get a more consistent and controllable gate for each step. Springs can be unpredictable in how they are going to bend and contort where elastic strands would emulate a contracting muscle better. You could implement that with some type of sleeve and capsule guide for the strands to adjust how the knee and ankle bend along with the toes. I also agree with the extra flexibility of the spine as this will also allow for some natural balance motion as the robot learns to walk though I would introduce more constraint to the spine as it is unnaturally moving which could also throw off the balance of the unit. If you can make it emulate the range of motion a human truly has and weight it proportionally I bet you could get it to walk on it's own introducing the same movement we have meaning the arms motion and side to side waggle for each step. Right now watching it walk the legs seem to have too much spring back for each step. It is not a fluid motion which also would contribute to balance issues. I have always wanted to get into robotics but never had the time or money to learn or experiment on my own. I have the ability to 3D print but no hardware to work with or software. I understand most of how the leg works as I ran track and field and have seen many videos of muscles working. I honestly think if you use what I have mentioned you will be able to fine tune the legs sort of like you would spokes on a bike vs trying to combine springs with servos and actuators. Trick is to build a bubble shape tube to emulate the muscle with channels for each elastic strand sliding a snug sleeve over each for each part you want to control and adjust accordingly. This should allow the servo to work more freely and the actuator the right amount of tension to not spring the leg back into position too quickly. If anyone has any idea of what I'm talking about I would love to know if i'm on the right track for my own version I hope to build one day. I hope to do so soon. If this at all helps please let me know as well.