I wanted to write this article to highlight the difference of mass distribution between humans and humanoid robots (small to medium size) in order to highlight the difference and the consequences in terms of control and the need for new actuators dedicated to humanoid robots.
According to [1], the average percentage of weight for each body part is as follows:
In an interesting article of Drillis and al. [2] I discovered the work of Harless who used a specific method in order to calculate the mass of body segments.
Harless [3] in Germany first used the immersion method. In 1858 he published a text book on Plastic Anatomy, and in 1860 a treatise, The Static Moments of the Human Body Limbs. In his investigations, Harless dissected five male cadavers and three female cadavers.
For his final report, however, he used only the data gathered on two of the subjects. The immersion method involves determining how much water is displaced by the submerged segment. Previous researchers, including Harless, have relied on the measurement of the overflow of a water tank to find the volume of water displaced. Harless started his studies with the determination of the absolute and relative lengths of the body and its segments. The absolute lengths were measured in centimeters. For determining the relative lengths, Harless used the hand as a standard unit. The standard hand measurement was equal to the distance from the wrist joint to the tip of the middle finger of the right hand. Later Harless also used the total height of the body as a relative unit of length. In the more recent studies on body parameters, this unit is accepted as the basis for the proportions of the various segment lengths.
Although the work of Harless if outdated, one can see that the two studies are giving quite similar results.
To visualize the mass distribution of the human body, Harless constructed the model shown in Figure 1. The linear dimensions of the links of the model are proportional to the segment lengths; the volumes of the spheres are proportional to segment masses. The centers of the spheres indicate the location of mass centers (centers of gravity) of the segments.
Figure 1: Body mass distribution (After E. Harless [3]).
If we do the same exercise for ODOI, the result is shown on Figure 2.
Figure 2: Mass distribution for ODOI
If you compare the two drawings you can see the differences! The main one is the weight of the lower extremities – specially the foot – quite huge discrepancies between a human and this robot.
This is due to the fact that actuators are located at the joins. More degrees of freedom at given join means more actuators and thus more weight. And if more torque is required, then the motor are heavier.
Consequences in term of control
To feel the difference as a human, fix 3 or 4 kilos at your ankle and then try to walk. You will start to move your torso back and forth while moving your swing leg in order to avoid falling forward or backward.
How to overcome this problem or redistribute the weight?
Reducing the weight on the foot is the main priority, but how to do it?
The first thing we can do is to redesign the leg for instance (our main concern regarding ODOI) in order to redistribute the weights (actuators) as it is shown on the picture below. The new design is clearly better but there is a price to pay: we loose the alignment of the foot axis (frontal plane) and ankle axis (saggital plane).
Figure 3: Redesign the leg
Another possibility is to look at the human muscle and understand how it works in order to design new actuators.
I will focus on the Hill muscle model which has been the dominant theoretical model for understanding muscle mechanics and is usually used in biomechanical computer models employed to simulate human movement.
The Hill muscle model [4] - see Figure 4 - has two elements in series and one element in parallel. The contractile component (CC) represents the active tension of skeletal muscle, while the parallel elastic component (PEC) and Series elastic component (SEC) represent the two key sources of passive tension in muscle.
Figure 4: The Hill muscle model.
With this model it is possible to model the difference phase of contraction/expansion of the muscle.
The Series Elastic component of the Hill model brings us to Series elastic actuator [7] which have been developed for the amazing Flamingo biped robot [6]. Such actuators are still being used in humanoid project but only human-sized ones such as THOR for instance. So far I did not see any development regarding Series elastic actuator for small or medium sized robots. I do not know if it is a problem of design or mechanical/engineering limitations but it is probably a direction of research.
Another direction of research is led by MIT Biomimetic laboratory with the Cheetah project [5]. It consist in developing a new actuator with high torque, light weight [9]. The actuator then drives the limbs though cables as it is shown on Figure 5.
Figure 5. MIT Cheetah robot
The work achieved by the MIT Biomimetic laboratory can be reused for a biped robot? I am convinced the answer is yes, but more research need to be done. However the work achieved with the fast runner [8] is a good start!
Finally a special type of actuator is the electromagnetic actuator. Control variable of this actuator is electric current and acting variable is force interaction and its effects. The principle of transformation in these actuators is based on force interaction in a magnetic field. Electromagnetics actuators are used in many applications (from small devices for a very precise control of position to quite powerful units such as drives of rods in nuclear reactors).
The solved actuator consists of two basic parts – electric and magnetic circuits. The electric circuit is formed by a cylindrical coil wound fixed on the frame. The magnetic circuit is formed by the shell and movable core. Movable core is placed on the axis of the actuator and can move freely in it. To reduce the friction force between the moving core and shell as well as to prevent their mutual impact the core is placed in a nonmagnetic sliding tube. The current in the coil produces magnetic field that gives rise to the Maxwell force which acts on the ferromagnetic core.
So far I did not see any robot powered by electromagnetic actuators. Few years ago I saw on article published by the MIT Media Lab but I cannot find it anymore. I saw also some research Osaka University but it was in 2009.
In this article I just wanted to outline the need for new actuators at least for the lower limbs of a biped robot (hip, knee and ankle). They provide high torque, compliance and light weight. They can be located in the pelvic for instance and thus redistribute the mass over the whole body.
I think that similar research should be done for small to medium size robots (less than 1 meter).
References
[1] Aydin Tözeren, Human Body Dynamics: Classical Mechanics and Human Movement, Springer-Verlag, 2000.
[2] Drillis R., Contini, R., & Bluestein, M. (1964). Body Segment Parameters; A Survey Of Measurement Techniques. Artificial limbs, 25, 44.
[3] Harless, E., The static moments of human limbs (in German), Treatises of the Math.-Phys. Class of the Royal Academy of Science of Bavaria, 8: 69-96 and 257-294, 1860.
[4] Hill AV. The revolution in muscle physiology. Physiology Revue 1932; 12-56.
[5] Hyun, D. Jin, S. Seok, J. Lee, and S. Kim, "High speed trot-running: Implementation of a hierarchical controller using proprioceptive impedance control on the MIT Cheetah", The International Journal of Robotics Research, vol. 33, pp. 1417-1445, 2014.
[6] Pratt, Jerry 2000. Exploiting Inherent Robustness and Natural Dynamics in the Control of Bipedal Walking Robots. Ph.D. Thesis, Computer Science Department, Massachusetts Institute of Technology, Cambridge, Massachusetts, 2000
[7] Robinson, D. W., Pratt, J., Paluska, D. & Pratt, G. (1999). Series elastic actuator development for a biomimetic walking robot, Proceedings of the 1999 IEEE/ASME International Conference on Advanced Intelligent Mechatronics, Atlanta, pp. 561 – 568
[8] Sebastien Cotton, Ionut Olaru, Matthew Bellman, Tim van der Ven, Johnny Godowski, and Jerry Pratt, FastRunner: A Fast, Efficient and Robust Bipedal Robot, Concept and Planar Simulation, Proceedings of the 2012 IEEE
[9] Seok, S., A. Wang, D. Otten, and S. Kim, "Actuator design for high force proprioceptive control in fast legged locomotion", Intelligent Robots and Systems (IROS), 2012 IEEE/RSJ International Conference on, Vilamoura, Portugal, IEEE, 10/2012.