When I first came to Japan I wasn’t sure what I was going to do for my research. The only thing I knew for sure was that I was going to work with a robot hand for prosthetic applications. My supervisor showed me the robot hand of the lab and of course at that time I was amazed by it (I still am, but to a lesser degree). Soon I was going to realize that the hand didn’t quite work very well for precise manipulation or worst for my research purpose. So in the next few posts I will describe briefly the changes I had to do to this robot hand in order to use it for my research. Basically, my Master and PhD research was about sensory feedback for prosthetic applications, therefore I needed a working robot hand to do the experiments, so I didn’t have time to make a new one to fulfill my needs. That is why I had to adapt as fast, simple and cheap as I could the one we had in the lab.
First I want to briefly introduce the robot hand. This robot hand has become to be know as the Yokoi hand, after one of its designers, professor Yokoi. However, I think this name got famous on internet thanks to the AI Lab of the University of Zurich, but the company that makes it in Japan ( Tsukasa Kiko Engineering ) calls it: the EMG Robot Hand TYK-2005Ver III. Anyway, the hand itself is quite nice. It is a tendon driven robot hand with 13 degrees of freedom, including 2 for the wrist and 2 for the thumb. Its weight is only 350g not including the servo motors. The main feature of this hand is that the motors are not placed on the robot hand itself, but the weight can be distributed on the socket. This makes the weight more bearable for the amputee. Also, because of its structure and materials the robot hand doesn’t require very complex control algorithms to make different functional grasps, since the fingers adapt to the shape of the object that is being grasp. Therefore, we don’t need to program the grasping shape in advance, we just need to send a closing command to the hand and the object will be securely enclosed by it without the need of a complex network of pressure or position sensors (I will be talking about the cons of this point later). This is great for EMG controlled robot hands or myoelectric prosthetic hands, because it reduces the learning curve for the amputee for specific situations. This product it’s basically intended for research purposes (because of it’s price. Sadly as most of the prosthetic hand in the market nowadays), but there are some people in Japan that are using it in their daily life, which is great!
Sadly, this robot hand is not sold with any kind of sensor, which is something very important for complex manipulations or for sensory feedback, as I will talk about in a later post. The other thing that is not very good is the way the fingers are flexed and extended. Basically, the fingers use 2 independent tendons for the distant and proximal phalanx, one to flex and the other to extend, similarly to the human hand. However, in order to reduce the amount of motors needed, the flexion and extension is achieved by only 1 motor. This mechanism would work if the tendons were tightly adjusted (same tension in both tendons), which creates a stiffness in the fingers. This cannot be achieved by this robot hand because the tendons cannot be equally tensed, thus the stiffness of the fingers is very low and only depends on the frictions in the joints. This is, as you can imagine, very bad for position control. Basically you cannot maintain a constant position if you don’t have stiffness. Also, because the tendons are not equally tense, then when the motor moves to the opposite side there is a delay in the actuation (while the tendon gets enough tension to move it). This is not a problem if the intended motion are gross, but for precise manipulation is a big problem.
There are different methods that can be used to solve this problem. For this particular robot hand the fastest way (but not necessarily the easiest) is to use independent motors for each tendons, but of course this will double the amount of motor needed, which doubles the weight of the robot hand, which of course is not an option for prosthetic applications. Also, the control system for this method is not trivial. On the bright side this will give more freedom to control the stiffness and realize motion more accord to the human hand.
The other method is to realize the finger extension by a resting spring, nowadays this is normally done with an torsion spring (like the system used by the SmartHand), this way the motor is only used to flex the finger. This greatly reduces the position’s control complexity, but we cannot change the stiffness of the fingers.
The problem with our robot hand was how to add the springs to the structure. I was very tight with time, since I had to make an experiment with a real robot hand for the sensory feedback system I was designing at the time, and I needed the position control. So I considered the torsion springs, but meanwhile I found a great and cheap solution by using: Rubber Strings. As you can see in the right image, the rubber string was attached to the finger so it would extend it. I was very surprise that this idea worked for long periods. The only problem is that the rubber looses its elasticity and its not very strong. Of course I don’t even have to mention how ugly and messy it looks. Actually, now we are again using this rubber strings for a robot arm we are testing in the lab, while waiting for the real springs (I will get to this one of these days)… great material to have at hand! 🙂
After, I considered the torsion springs more, but sadly I couldn’t find a way to put them without cutting and making holes to the fingers. This of course wasn’t an option at the time because we only had this robot hand in the lab! So I came up with another idea on how to use the springs. I used the small canals that are used for the tendon tube in the base of the fingers to hold a tension spring, this way when the finger flexes the spring will never bend and can keep the force quite constant. As can be seen from the pictures down, 1 spring is needed for each phalanx. For the thumb was more difficult to put this kind of springs, so I used the setting you see in the figure, but this one is not that great when compared to the other fingers. This method is not optimal of course, but it gets the job done, without modifying the robot hand.
I’m guessing that by this point you are asking yourself, why on earth they didn’t just buy a new hand? or why they didn’t just make a new hand that can hold the torsion springs? I asked this questions all the time I was trying to make this hand work. Basically, the answer for the first question is money. Sadly, dexterous prosthetic hands are very, very expensive… even for universities. There are great devices out there (i-Limb, the Smart hand, etc.), but their prices, for various reasons, are … ufff… too much. This is something I would really like to help change in the future and a lot of people are trying to do as well. So we have to keep working on it. The answer for the second question, basically is that at that time we didn’t have a CNC or 3D printer in the lab, so we would have to pay a company to make the parts, which is very expensive. Luckily, as I hope you will read in future post, we got these machines now and are doing very cool things, one of them is a new finger for this hand.
In the next post I will discuss the sensors I used in the robot hand.
Some books, papers and links to read if you want to get into this topic:
The Open Prosthetic Project – This is a great link for an internet community dedicated on developing prosthetics! The only problem is that it seems there hasn’t been a lot of activity lately. Still is a great source of ideas and info.
Murray R. “A Mathematical Introduction to Robot Manipulation” (1994) – Great book to read about this topic!
Cipriani C. et al. “Progress Towards the Development of the SmartHand Transradial Prosthesis” (2009) – Discusses the mechanisms of the SmartHand.
Nishikawa D. et al.”On-line Supervising Mechanism for Learning Data in Surface Electromyogram Motion Classifiers” (2006) – a great method to classify up to 10 hand motions from only 2 EMG sensors.
Palli G. “Model and Control of Tendon Actuated Robots”(2006, Ph.D Thesis) – Great document on the subject.
Jacobsen S.C. et al. “Antagonistic Control of a Tendon Driven Manipulator” (1989) – talks a little bit about different Tendo Driven Systems and presents a control mechanism for a antagonist system.