Force and Speed in a Soft Stewart Platform

TL;DR

This work addresses the challenge of achieving fast, large- displacement six-DoF motions with soft robots by integrating Handed Shearing Auxetics into a 6DoF Stewart-Gough platform. A data-driven learned inverse kinematics model complements a cascaded PID control strategy to deliver precise tracing and robust disturbance rejection, while frequency response and workspace analyses show competitive performance with rigid platforms. The system supports a $2$ kg payload with an open-loop bandwidth above $16$ Hz and demonstrates a $4.0$ mm/$1.15$ degree IK precision, along with a $<12$ cm natively reachable workspace. Collectively, the results indicate that soft actuators can deliver rigid-platform-like force, speed, and workspace with reduced hardware, enabling safer human-robot interaction and scalable soft-positioning for tasks like manipulation, assembly, and tactile sensing.

Abstract

Many soft robots struggle to produce dynamic motions with fast, large displacements. We develop a parallel 6 degree-of-freedom (DoF) Stewart-Gough mechanism using Handed Shearing Auxetic (HSA) actuators. By using soft actuators, we are able to use one third as many mechatronic components as a rigid Stewart platform, while retaining a working payload of 2kg and an open-loop bandwidth greater than 16Hz. We show that the platform is capable of both precise tracing and dynamic disturbance rejection when controlling a ball and sliding puck using a Proportional Integral Derivative (PID) controller. We develop a machine-learning-based kinematics model and demonstrate a functional workspace of roughly 10cm in each translation direction and 28 degrees in each orientation. This 6DoF device has many of the characteristics associated with rigid components - power, speed, and total workspace - while capturing the advantages of soft mechanisms.

Figures (7)

• Figure 1: Photo and Dimensions of the soft Stewart platform: The soft Stewart platform at a near-rest position. The platform has a diameter of 40.6cm and a strut-length of 25cm. The platform-to-platform distance ranges from 25.3cm to 33.2cm depending on the system's state.
• Figure 2: Waypoint Letter Tracing: We show that cascading PID control can be used to track waypoints in the shapes of letters spelling "HELLO WORLD". Across five trials for each letter, we show that the rolling puck consistently and accurately reaches each waypoint. Letters are traced sequentially, where each letter spans the area of the hexagonal platform.
• Figure 3: Mechanism Workspace: We performed a raster scan with 90-degree resolution over all combinations of motor angles from $[0, 270]$ degrees. The resulting positions and orientations are displayed with a convex hull for clarity. We found the X-Y cross-section of the position data to be approximately triangular with the extreme points (shown) opposite the corners where two servo motors are mounted. The maximum Z position is also displayed. For orientation, we display the poses in the workspace with maximum roll, minimum pitch, and maximum yaw.
• Figure 4: System Control Diagram: The control architecture used for our manipulation experiments. The inner PID loop handles ball velocity and the outer control loop handles ball position. $dx$ represents the velocity term used to compute error for the inner loop. Each axis $x$ and $y$ has a separate cascade controller.
• Figure 5: Learned Kinematics Model Comparison: We compare sample platform poses produced by our learned inverse kinematics model (purple) to a target pose (blue) and a rigid approximation (green). We tested 100 randomly generated poses which are not in the training set and found that in each dimension the learned IK model is accurate to within 4.0mm and 1.15 degrees on average.