This week I looked at a paper written in 2008 that attempts to create a biomimetic robot eye based upon the human eye with some success. Since I’m presenting a summary of this paper this week I thought I’d share the presentation.
You can access the paper at http://www.sciencedirect.com/science/article/pii/S1672652908600347
Wang, X., Zhang, Y., Fu, X., & Xiang, G. (2008).
Design and Kinematic Analysis of a Novel Humanoid Robot Eye Using Pneumatic Artificial Muscles. Journal of Bionic Engineering, 5(3), 264–270. https://doi.org/10.1016/S1672-6529(08)60034-7
This paper proposed a novel humanoid robot eye, which is driven by six Pneumatic Artificial Muscles (PAMs) and rotates with 3 Degree of Freedom (DOF). The design of the mechanism and motion type of the robot eye are inspired by that of human eyes. The model of humanoid robot eye is established as a parallel mechanism, and the inverse-kinematic problem of this flexible tendons driving parallel system is solved by the analytical geometry method. As an extension, the simulation result for saccadic movement is presented under three conditions. The design and kinematic analysis of the prototype could be a significant step towards the goal of building an autonomous humanoid robot eye with the movement and especially the visual functions similar to that of human.
This paper looks at the creation of a robotic eye that is designed to mimic the biological construction of the human eye. The authors state that, at the time of writing, most research into the construction of robotic eyes either required the use of motors or was more focused on the aesthetics than the functionality. There is a proposed theory that mimicking the human eye may bring unknown computer vision benefits and allow robots to process vision in new ways. In particular, the focus of this paper is on replicating the saccadic movements of the human eye that allow us as a species to rapidly focus on a target and piece together individual image fragments.
The human eye is made up of two high-level components – the eyeball itself and the musculature. The eyeball, at a very basic level, is an almost-spherical organ with a retina at the back that has light focused on it. The retina detects this light and then passes it through the optic nerve to the brain where it is processed as vision. The musculature in the eye allows three degrees of freedom from six extra ocular muscles. The four rectus muscles provide up, down, left and right motions whilst the two oblique muscles act as a pulley around the side of the eye providing a rotation around the third axis.
In the creation of the robotic eye specific attention was paid to reproducing the muscles used in the human eye. The authors decided to use pneumatic artificial muscles (PAMs) due to their small and lightweight nature twinned with the fact that that they were able to find a PAM with the same contraction ratio and force displacement as used in human eyes. PAMs are basically prismatic joints that work by inflating or deflating a small tube causing it to contract or expand in a similar way to human muscles. These PAMs are combined with a standard (although low resolution) CMOS sensor to acquire images – performing a similar function to the retina in a human eye. Additionally an orientation sensor is required for the operation of the PAMs and is present in the robotic eye but not in a human eye.
These components were then constructed into a perfect sphere with all six PAMs attached – the oblique muscles looping through the back of the sphere whilst the other four were attached in place. The robot eye was held in place by a strong central pipe (which also carried cabling) and the pipe and muscles were attached to a back plate providing the support for the robotic eye. After construction a kinematic analysis was conducted assuming all rotations to be around the exact centre point of the sphere.
The authors then diverge from the focus of creating a robotic eye and perform a brief set of MATLAB simulations, based on these inverse kinematics, to replicate the saccadic movements of the human eye. The simulation has no analysis, very little description and feels completely out of place in the paper. It is followed immediately by a conclusion section which draws no conclusion other than this eye is a “significant advance”.
Whilst I do appreciate that the use of PAMs rather than motors to move a robotic eye is potentially a great thing to do, unfortunately the paper has numerous issues which can be summarised as:
- Unexplained choices in the construction of their system
- The paper has typographic and grammatical mistakes
- The authors do not set out a goal
- No conclusions are drawn
- There is no indication of a contribution
- The section of saccadic movements does not feel connected to the rest of the paper
- The kinematic analysis does not complete the analysis and leaves parts of it open
- The robotic eye is never actually used or even known to work
- There are no details on how this could impact computer vision techniques
Sadly a paper that could have promised a novel new approach for computer vision by this piece of hardware is let down by being unable to show any advancements using this technique.