In the first part of our Illustrated Guide to Designing Humanoid Robots, we covered some of the dos and don’ts, as well as various styles and approaches. Now we’re going to apply some of those lessons to the creation of our very own humanoid, while also adding a couple of pointers on proportions and consistency. What follows is a break down of our concept for PAL Robotics’ contest to design their forthcoming humanoid robot, REEM-C. In case you missed it, be sure to read part one in advance.
Your robot should look consistent from top to bottom – avoid putting a realistic head on a mechanical body. The head, arms, hands, and other parts shouldn’t be too large or too small. That said, proper proportions aren’t always possible when building a robot using prefabricated parts. NTUST’s Adam has a bulging chest due to its internal mechanisms, and KAIST’s Albert Hubo has an oddly shaped torso that covers its neckline.
If the robot has an exoskeleton, it should cover the whole body. Otherwise, the robot may look incomplete or unfinished. A consistent, well designed exoskeleton can even make up for proportional inconsistencies. Take MIT’s Nexi as a cautionary example. The face looks finished, but the rest of the body is not. The head is too large for the rest of the body. The lower body is truncated prematurely at the waist. Compare the following robots and ask yourself:
- Which robot looks incomplete?
- Which robot has inconsistent proportions?
- Which robot looks dangerous to touch?
Nexi (MIT), Robovie R3 (ATR), and Posy (Flower Robotics)
Even though both Posy and the Robovie R3 have large heads and tiny arms, they get away with it because of their more natural body length and overall finished appearance.
Compare the following robot heads. The REEM-H1′s microphones (brow region) and eyes are fighting for your attention; the face has no clear focal point. In order to incorporate sensors, Nexi’s forehead was elongated which looks weird and unnatural. In contrast, Hitachi’s EMIEW-2 has cameras and a microphone array which are largely concealed by the red helmet, which also looks reminiscent of a frog, lending it a childlike feel.
REEM-C has a helmet which can be used to hide the microphones. As with EMIEW-2, a number of microphones could be distributed around the crown in order to get better sound source estimation. We haven’t explicitly shown where the microphones and speakers would go, but there is plenty of room for them in the inner rim of the helmet.
REEM-C’s face has been kept minimalistic, partly because I prefer it this way, and partly because none of PAL Robotics’ previous robot designs have incorporated eyeballs, eye lids, eye brows, or lips. The REEM-H1 was the first of their robots to feature a nose, and it doesn’t work very well. An expressive face doesn’t appear to be a priority, so we have used simple cut-outs for the eyes (stereo cameras). We have opted to rim the eyes with color-cycling RGB LEDs to focus your attention on them. Simple and elegant.
In the above image you can also see the main indicator light, in the form of the REEM-C logo we designed. This not only helps to brand the robot, but can change color to indicate the robot’s internal status (kills two birds with one stone).
Here’s another shot of the upper body, with the focus on the shoulder. The shoulder joint is able to roll 90 degrees into this position without the upper arm penetrating the shoulder pad. In part one of this guide we saw that two-toned color schemes work well – and here you can see we are using black and silver. The fingers are a lighter color because they are covered in semi-translucent rubber.
Here’s a shot of REEM-C’s elbow joint (it can bend even further), largely inspired by the SONY QRIO. Note that the link for the upper arm slides around and inside the link for the fore arm, leaving no gap on the outside of the elbow. The inside of the elbow could still potentially trap fingers, which is why we would recommend putting pressure sensors in this location. If something gets caught in the gap, the sensors would detect it and cause the joint to reverse direction. Also note that the wrist joint has 2 degrees of freedom, allowing it to twist and roll.
REEM-C’s upper back. The concave oval is where the rear-facing camera goes. The rear-facing camera was present in the REEM-H1 and I decided to include it here, since it allows the robot to track what is going on behind it. This would also be a good place to put ventilation, but I have left the precise location of such a detail to the experts.
The back of the legs. Similar to the folding elbow, the back of the knees fold inside the calve link to hide the internal mechanisms even when the legs are perfectly straight. The legs are able to roll and twist 90 degrees at the hip, which would allow it to turn 90 degrees to the left or right in two steps.
The foot has a toe joint, which lends a little added support during the walking gait. This feature was primarily inspired by Waseda University’s Wabian-2, which has a very natural gait. The sole of each foot is covered with an impact-absorbing rubber, and the pressure sensors are located in the corners to help calculate the center of gravity. Reading some of Honda’s history building robots, they said that the walking was made much more stable when cushioned by rubber soles.
Waseda U.’s Wabian-2 has a toe joint which gives it a smooth gait
Part of the contest’s criteria is that the design should maintain stylistic consistency with PAL Robotics’ previous robots. Clearly PAL Robotics is interested in branding their robots so that when you see one, there’s no question who developed it. With that in mind, let’s examine how other institutions have maintained stylistic consistency between their robots.
KIST’s Mahru-M and Mahru-Z had identical head designs to the original Mahru
It seems the best way to maintain consistency across multiple robots is to use similar head designs and color schemes. Unfortunately, PAL Robotics has not been practicing this unified approach. REEM-A and REEM-B have somewhat similar heads, but they have very different color schemes overall. The REEM-H1 has a completely different head design and color scheme. And the conceptual renderings of the forthcoming REEM-H2 has nothing in common with any of the previous REEM designs.
What’s done is done and can’t be fixed, but PAL Robotics can begin the process moving forward. There are two ways to do this. Consistency could be achieved if REEM-C was based on REEM-H2. The main issue I have with this is, quite simply, the REEM-H2 does not look very friendly – its face reminds me of the killers’ masks from horror films like Friday The 13th or Halloween. Furthermore, if we were to retain the head design and upper body of the REEM-H2 and simply add legs to create the REEM-C, this wouldn’t be much of a design contest, would it? So we are suggesting that they abandon the current REEM-H2 design altogether.
We’re suggesting that the new design of REEM-C should largely inform the look of the REEM-H2. When seen side by side, this leaves no doubt that they came from the same company. In the future, REEM-D and REEM-H3 should also share common traits with one another to continue this trend. This helps to create a cohesive “brand” for the company’s robots.
Our version of REEM-H2 uses many of the parts from REEM-C, but has a unique chest to hold the touch screen, and features an omnidirectional wheeled base which connects at the hips. The logo just below the touch screen is again used as an indicator of the robot’s status using color-changing LEDs. Like the REEM-H1 and REEM-C, it has a rear-facing camera on its upper back. We made a minor change to the back of the helmet to show that this robot could feature a different helmet design, but due to time constraints we couldn’t make it truly unique.
We have adopted an opening similar to that of Nissan’s EPORO for where the laser range finder goes (near the base). A different option would be to have a smaller slot with the laser range finder in a tilting configuration similar to Willow Garage’s PR2, which would provide even more point cloud data. It is worth noting that neither of these are particularly attractive solutions, but they are the best of what is out there.
Other sensors include a rear-facing camera in the center of the robot’s upper back, which was carried over from the REEM-H1. There would also be a series of ultrasonic sensors around the midriff and base (see next figure), but again due to time constraints we didn’t add them. On the back’s cargo bay, we have implemented a combination of the REEM-H1 and the current REEM-H2 concept’s protective covers. The REEM-H1 has a series of rubber grips to prevent objects from scratching the robot, and these also provide some extra grip to hold objects in place. The REEM-H2 concept had a trio of rubber nubs which would help hold an object in place.
Our concept combines the cargo bay configurations of the previous designs
If we had had more time, we would have also added some lights to the front and back of the robot, similar to a car. These RGB LEDs would blink to indicate if the robot is going to turn left or right, and change color when slowing to a stop, which would notify people of the robot’s intentions.
Had this been a contract rather than a contest, we would have worked with the client throughout the design process, making changes according to feedback. Given free reign, we hope our design meets the criteria of an “elegant, strong, and friendly” robot. With the redesign of the REEM-H2, there is now consistency across the platforms. We have tried to keep things realistic, the joints have good range of movement, and have based much of the design on various bits and pieces of robots that already exist.
It would be an honor to design one of the world’s full-scale humanoids, but even if Plastic Pals doesn’t win the contest, it has been fun just participating. If you are a company in need of a robot design, please feel free to contact us.