【Prosthetic Hands】17 Third Generation - Software Design PART III

 In my research, the decision strategy is the decision process between (classifier) generating a result and (prosthetic hand) performing a motion. After the classifier categorizes the motion signals into a specific class, the decision strategy will control the actual motion of the prosthetic hand based on the given class. Here I investigate two types of decision strategies - the Intuitive Strategy and the Practical Strategy. 

Intuitive Strategy resembles how humans perform grasps. To allow the user controlling the prosthetic hand “intuitively”, the motion of the prosthesis is designated to be as similar as possible to the user’s intention. For example, when the classifier categorizes a motion signal into "Power Grasping", the program will move the prosthesis and perform a motion similar to "Power Grasping". This strategy is quite straightforward, yet some problems show up when it is applied in daily life. (I will discuss them later.)

Before we look into the Practical Strategy, I would like to first talk about "how to complete a grasping task". Let's consider a common scenario. A person wants to lift a water bottle from a table, drink some water, and put the bottle back on the table. During the process, he/she needs to perform four motions. 

1. Stretch the fingers to make space for the water bottle

2. Grasp and lift the bottle

    (Drink some water à no need of changing hand motions)

3. Stretch the fingers again to put down the bottle

4. Return the hand back to resting position

In this scenario, it takes three motion classes to finish the task, and the user must produce four motion signals. If any of the classification result is incorrect, the user cannot finish the grasping task (Or cannot return to the resting position). Moreover, in some situation, the wrong classification result my accidentally spill the water or cause structural damage to the prosthetic hand. These mistakes are unpredictable, so they may impose dangers on the user or the hand itself.

The most straightforward to avoid wrong recognition results is increasing the accuracy of EMG classification, but the extent of improvement is limited by the acquisition of surface EMG signals. Instead of stubbornly chasing the high accuracy of classifiers, I turn to proposing a new decision strategy - Practical Strategy.

Practical Strategy is designed to be used on prosthetic hands to enable safer and more efficient grasping experience, and it also reaches higher success rate compared to Intuitive Strategy. The idea is simple. Instead of designating a motion class to a specific hand motion, each motion class is designated to one specific type of grasping task. Based on the research by I. M. Bullock et al. [1], the top 10 most-frequently-used grasping motions cover more than 80% of the ones used in daily life. That is, by designating the available motion classes to the most-frequently-used grasping tasks, the hand can achieve most of the motions needed in daily life. 

In Practical Strategy, the user perform the same motion signal (based on the grasping task) for two times. The first motion signal moves the fingers the preparing position (in the previous example of grasping bottles à stretching the finger), and the second identical motion signal would trigger the main grasping motion. After the grasping is complete, the third motion can be arbitrary, and the hand will automatically looses the object and return to the resting position.

Practical Strategy provides a safer user experience because of its double-insurance mechanism. If the second motion signal is not the same as the first one, the hand will automatically be returned to the resting position, so even if the recognition result is incorrect, the user can foresee the upcoming motion and take action before danger happens. In addition, Practical Strategy is more efficient in most situation. Because the forth motion (returning to resting position) has been integrated with the third motion, users only have to produce three motion signals to finish a grasping task. Last but not least, Practical Strategy increases the success rate. As I mentioned in the previous part, Intuitive Strategy takes four motions to finish a grasping task, and each motion classification result must be correct. On the contrary, in Practical Strategy, only the first two motion signals need to be correct, so obviously the success rate is increased.

You may wonder if Practical Strategy restricts the diversity of hand motions, but in fact, Practical Strategy enables more hand motions. The number of accessible motions in Intuitive Strategy is limited by the number of available motion classes. On the other hand, the motions available in Practical Strategy depends on the grasping task. 

Based on the chart above, Intuitive Strategy enables eight motions, while Practical Strategy enables totally 12 motions (repeated motions eliminated).

To conclude, Practical Strategy outperforms Intuitive Strategy in many aspects (I will present some experiment results in the future articles) and is thus more suitable for daily use. 

After completing the software design, I will then introduce how I integrate hardware and software in the next article. Hope you enjoy this one!

[1] I. M. Bullock, J. Z. Zheng, S. De La Rosa, C. Guertler and A. M. Dollar, "Grasp Frequency and Usage in Daily Household and Machine Shop Tasks," in IEEE Transactions on Haptics, vol. 6, no. 3, pp. 296-308, July-Sept. 2013, doi: 10.1109/TOH.2013.6.