Two-Finger Soft Gripper Force Modulation via Kinesthetic Feedback

TL;DR

The hypothesis is that once the contact between a finger and an object is detected, a controller that keeps a desired difference between the finger bending measurement and its bending at the moment of contact is sufficient to maintain and modulate the contact force.

Abstract

We investigate a method to modulate contact forces between the soft fingers of a two-finger gripper and an object, without relying on tactile sensors. This work is a follow-up to our previous results on contact detection. Here, our hypothesis is that once the contact between a finger and an object is detected, a controller that keeps a desired difference between the finger bending measurement and its bending at the moment of contact is sufficient to maintain and modulate the contact force. This approach can be simultaneously applied to both fingers while getting in contact with a single object. We successfully tested the hypothesis, and characterized the contact and peak pull-out force magnitude vs. the desired difference expressed by a multiplicative factor. All of the results are performed on a real physical device.

Figures (10)

• Figure 1: Two-finger soft gripper experiment setup: Fingers F1 and F2 are actuated by feedback control using their bending measurements and outputting a control variable for the corresponding finger inflation pressure. The cylindrical object positioned at $(X_c,Y_c)$ is mounted on a force sensor which we use to measure contact forces depicted along the $X$ and $Y$ directions. The finger design is based on 6696549.
• Figure 2: The soft finger feedback control: $R_{i}^{ref}$ is the bending reference for the bending measurement $y_i$ of the control loop with the tracking controller $D_{t,i}(z)$ (orange). The contact detection, which is based on magnitudes of $e_{t,i}$ and $d_i$, switches to the force modulating control with the controller $D_{f,i}(z)$ (purple). The signals $y_{@con,i}$ and $d_{@con,i}$ are bending and duty cycle at the time of contact, $d_{bias,i}$ compensates for the dead zone in actuation, and $P$ depicts the constant air pressure source.
• Figure 3: Bending measurements $y$: $(U_k,V_k)$, $k=1,2,3$ are pixel coordinates of the centers of the circular markers, $(U_0,V_0)$ is the center of a circle fitted to the three markers. Vector magnitudes $|r_k|$, $k=1,2,3$ are identically equal.
• Figure 4: Bending reference tracking: (A) Triangular reference inputs $R^{ref}_1$(green) for Finger 1 (F1) and $R^{ref}_2$(magenta) for Finger 2 (F2). The bending measurements $y_1$(red) for F1 and $y_2$(blue) for F2. (B) The tracking errors $e_{t,1}$(red) for F1 and $e_{t,2}$(blue) for F2 . (c) The duty cycle $d_1$(red) for F1 and $d_2$(blue) for F2 control outputs. During our experiments, we found that a dead zone of control action was approximately $20$%.
• Figure 5: Two-finger gentle force contact: (A) the solid green line is $R^{ref}_1$ for the $D_{t,1}(z)$ controller and $y_{con,1}$ for the $D_{f,1}(z)$ controller . The solid magenta line is $R^{ref}_2$ for the $D_{t,2}(z)$ controller and $y_{con,2}$ for the $D_{f,2}(z)$ controller. (B) Plots of controller tracking errors. The switch from $D_{t,1}(z)$ to $D_{f,1}(z)$ is when $e_{t,1}\ge e_{tr,1}=3.88$ and the switch from $D_{t,2}(z)$ to $D_{f,2}(z)$ is when $e_{t,2}\ge e_{tr,2}=4.71$. (C) Duty cycles $d_1$(red) and $d_2$(blue) for the inflation pressure of $F_1$ and $F_2$, respectively. (D) and (E) are plots of $X$ and $Y$ force components for $e_{des,1}=e_{des,2}=0$ on the cylindrical object of a radius $r=2 cm$ and positioned at the location ($X_c, Y_c$) = (0, 15) cm.