Design of a Data-Oriented Kansei Feedback System

In the development of the aging society, it is important for patients with hemiplegia to introduce adaptive welfare equipment. However, it is difficult to determine the suitable reference signal for each person. In this study, the design of a data-oriented cascade control system based on Kansei is proposed. In the proposed control system, there are two controllers which are a data-driven controller and a fixed controller. In particular, a data-driven controller can calculate the suitable reference signal based on Kansei.


Introduction
In the development of aging society, it is important for patients with hemiplegia to introduce the adaptive welfare equipment.However, it is difficult to support them by using general welfare equipment because there are a lot of individual disabilities.Therefore, an adaptive welfare equipment is needed in near future.However, it is very difficult to determine the suitable reference signal for each person.In this study, the design of a dataoriented cascade control system based on Kansei is proposed.In the proposed control system, there are two controllers which are a data-driven controller 1 and a fixed controller.In particular, a data-driven controller 1 is for a human and it can calculate the suitable reference signal of a welfare equipment based on Kansei.

Schematic figure of proposed control system
The schematic figure of the proposed control system is shown in Fig. 1.It is very difficult for patient with hemiplegia to move their foot by only torque τ from their brain.Therefore, in the proposed scheme, the Ankle Foot Orthosis 2 : AFO supports them to torque τ .In this paper, Kansei signal is defined as walking comfortable y( ) whose maximum value is 1.Therefore, reference comfortable signal ( ) is set as 1.Note that it is important to estimate reference signal of brain ( ) because ( ) is unknown.Therefore, a data-base controller (primary controller) in outer loop is applied to calculate the estimated reference signalw( ).

Controlled object
The schematic figure of tow vertical joint manipulator is shown as leg model in Fig. 2. , are moment of inertia in ankle and knee respectively.
, , are weight of upper body, lower leg and femur., , , are whole length and center of gravity distance of lower leg and femur respectively.The torque and are corresponding to angle of and .The equation of walking motion is expressed as follows: where ( ) denotes .In addition, Kansei signal y( ) in Fig. 1 is expressed as following equation based on Weber-Fechner law 3 : where ( ) is reference signal of ( ).

Design of a data-driven controller in outer
loop as primary controller

Control law of primary PID controller
The primary controller in Fig. 2 is designed as a datadriven controller 1 because human characteristic is nonlinear.The primary controller is defined as follows: where , and respectively are proportional gain, integral gain and derivative gain.denotes a difference operator.Note that an inner loop controller of AFO is designed as the following fixed PD controller.
(5) where are proportional gain and derivative gain.

Design procedure of a data-driven control
[STEP 1] Create an initial database.The historical data is needed to use data-driven control scheme.The database is constructed by following information vector: ( = 1,2, , ⋯ , ( ) denotes the th element of query ( ). max ( is a maximum th element in database.In contrast, min ( is a minimum th element.In addition, the number of neighbors' data selected, which data are based on smallest distance .[STEP 3] Calculate control parameters.Control parameters are calculated by using the following linearly weighted average (LWA): where is the weight corresponding to the th information vector ( ) in the selected neighbors.It is calculated by following equation: In order to calculate effective control parameters, an off-line learning method is described in next section.P -430 Where ( ) is given by following equation: Therefore, equation ( 15) and (18) show that control parameters can be learned off-line by using closed-loop data.

Numerical Example
In this section, the effectiveness of the proposed scheme is verified.The physical parameters 5,6 in Fig. 2

P -431
The dotted red line denotes the reference walking trajectory of brain.Walking support is not well in Fig. 4 because the blue solid line does not follow red reference signal.Therefore, the Kansei signal ( ) is not kept around 1 in Fig. 5. Furthermore, fixed PID controller cannot estimate the reference signal well because ( ) does not follow ( ).
On the other hand, walking trajectories of Fig. 6 by using the proposed scheme is better than Fig. 7.The effectiveness of the proposed scheme is shown by keeping almost ( ) = 1 in Fig. 7.The estimated reference signal (t) is almost same as (t) using a data-driven controller and PID gains in Fig. 8 are changed effectively.

Conclusion
In this paper, the field of welfare equipment is focused and the scheme based on the data-oriented Kansei feedback has been proposed in order to support each people adaptively.In the proposed scheme, reference signal of a brain can be estimated by using a data-driven controller and it can support walking well.The proposed scheme has been verified by numerical example.In the future, experimental result will be considered.We are particularly grateful for the assistance given by COI STREAM (Center of Innovation    P -432 ( ) ≔ [ ( + 1), ( ), ( ), ⋯ , − 1 , ( − 1), ⋯ , ( − (7) ( ) ≔ [ ( ), ( ), ( )], (8) where denotes the number of data.[STEP 2] Calculate distance and select neighbors' data.Distance between query ( ) and ( ) is calculated by using the following L1-norm with some weights: ( ), ( ) = ( ) − ( ) max ( − min ( ) .

Fig. 1 .
Fig. 1.Schematic figure of the proposed control system.

Fig. 4
Fig.4shows the walking trajectories corresponding to Fig.2by using fixed PID controller as primary controller.