End-to-End Design & Validation of a Low-Cost Stewart Platform with Nonlinear Estimation & Control
Benedictus C. G. Cinun, Tua A. Tamba, Immanuel R. Santjoko, Xiaofeng Wang, Michael A. Gunarso, Bin Hu · 2025
Read End-to-End Design & Validation of a Low-Cost Stewart Platform with Nonlinear Estimation & Control by Benedictus C. G. Cinun, Tua A. Tamba, Immanuel R. Santjoko, Xiaofeng Wang, Michael A. Gunarso, Bin Hu (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-22949.txt.
Introduction
Benedictus C G Cinun and five co-authors built a low-cost Stewart platform, a table held by six moving legs that can tilt and shift in six ways. Such platforms serve as test beds for control research, but good ones are often costly. Their build mixes shelf parts with printed and custom pieces.
Software on the platform handles modeling, data capture, and real-time control in one frame. A feedback-based controller with an added optimal scheme tracks paths despite nonlinear motion, while a Kalman filter merges sensor and encoder data. Tests confirm the cheap build can still move with care and steadiness.
Catalog record
| CREATOR | Benedictus C. G. Cinun, Tua A. Tamba, Immanuel R. Santjoko, Xiaofeng Wang, Michael A. Gunarso, Bin Hu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 54 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2510-22949
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
End-to-End Design & Validation of a Low-Cost Stewart Platform with Nonlinear Estimation & Control Benedictus C. G. Cinun, Tua A. Tamba, Immanuel R. Santjoko, Xiaofeng Wang, Michael A. Gunarso, Bin Hu Abstract This paper presents the complete design, control, and experimental validation of a low-cost Stewart platform prototype developed as an affordable yet capable robotic testbed for research and education. The platform combines off-the-shelf components with 3D-printed and custom-fabricated parts to deliver full six-degree-of-freedom (6-DOF) motions using six linear actuators connecting a moving platform to a fixed base. The system software integrates dynamic modeling, data acquisition, and real-time control within a unified framework. A robust trajectory tracking controller based on feedback linearization, augmented with an LQR scheme, compensates for the platform’s nonlinear dynamics to achieve precise motion control. In parallel, an Extended Kalman Filter fuses IMU and actuator encoder feedback to provide accurate and reliable state estimation under sensor noise and external disturbances. Unlike prior efforts that emphasize only isolated aspects such as modeling or control, this work delivers a complete hardware–software platform validated through both simulation and experiments on static and dynamic trajectories. Results demonstrate effective trajectory tracking and real-time state estimation, highlighting the platform’s potential as a cost-effective and versatile tool for advanced research and educational applications. I Introduction The Stewart platform is a six-degree-of-freedom (6-DoF) parallel manipulator renowned for its high stiffness, precision, and ability to support large payloads relative to its size [ 1 , 2 , 3 ] . These properties make it well-suited for applications requiring precise motion and load-bearing capability, including flight simulators [ 4 , 5 ] , wave compensation in maritime operations [ 6 , 7 , 8 ] , evaluating the performance of 3D-printed components under simulated ocean wave conditions [ 9 ] , surgical robotics [ 10 , 11 ] , telescope positioning [ 12 , 13 ] , and more recently, fine adjustment of segmented mirrors on the James-Webb Space Telescope (JWST) [ 14 ] . The breadth of these applications highlights the platform’s versatility and ongoing relevance in both industrial and research domains [ 15 , 16 , 17 , 18 , 19 ] . (a) (b) (c) Fig. 1 : Examples of commercially-developed Stewart platforms (sources: [ 20 , 21 , 22 ] ). Access to Stewart platforms for experimental research and education purposes is often limited by high costs, as commercial systems typically use precision actuators, high-resolution sensors, and advanced controllers, placing them beyond the reach of many academic or small-scale laboratories. For example, the Acrome Stewart platform in Fig. 1(a) costs around USD \$8,899 [ 20 ] . Larger systems for flight simulation or heavy-duty testing purposes are even more expensive, such as the PT-Actuator model in Fig. 1(b) whose cost starts at USD \$12,000 [ 21 ] , and the Motion Systems platform in Fig. 1(c) whose cost reaches up to EUR 50,000 [ 22 ] . To address this gap, several low-cost prototypes have been developed using alternative actuation and sensing strategies, including an educational proof-of-concept platform [ 23 ] , a surgical motion simulator that replaced linear actuators with servo motors to reduce costs to USD \$250 [ 11 ] , and a haptic controller driven by hobby servos [ 24 ] . However, hobby servos often have limited performance due to low-cost potentiometers and restricted torque. A cost-conscious Stewart platform was proposed in [ 25 ] that uses industrial servomotors for simulator applications to emphasize portability, performance, and the ability to handle human-sized payloads. In contrast, our design employs linear actuators, which provide higher load capacity and stiffness [ 26 ] , while integrated potentiometer feedback enables position sensing for closed-loop length control. It is worth noting that an addition of elastic springs to the linear actuator has also been proposed, which could act as a low-pass filter to reduce disturbances from actuator backlash and external forces [ 27 ] . A related low-cost design also used linear actuators [ 28 ] ; however, it relied on an Arduino controller, which limited computational capacity and lacked the design and implementation of advanced control and estimation strategies. In contrast, our developed Stewart platform incorporates nonlinear control and estimation methods to enable a comprehensive exploration of the functionality of the platform, making it more suitable for both education and advanced experimental research. For the purpose of implementing advanced control strategies and ensure reliable operation of Stewart platforms, it is essential to develop an accurate system model and state estimation algorithm. Early approaches for solving forward kinematics problems in parallel manipulators often relied on closed-form geometric solutions [ 29 , 30 , 31 , 32 , 33 , 34 ] or the Denavit–Hartenberg convention [ 35 ] . A modeling approach using Conformal Geometric Algebra method has also been proposed in [ 36 ] to characterize the relationship between the platform joints and the states of the end-effector. Although these analytic approaches can provide an exact solution, they are often sensitive to measurement noise and yielded multiple solution configurations. To resolve the limitation of analytical approach, iterative numerical methods such as Newton–Raphson and Levenberg–Marquardt methods have been proposed in [ 37 , 38 , 39 , 40 , 41 , 42 ] , but these methods require high-performance machines and accurate initial guesses. The works in [ 43 , 44 ] proposed the use of additional external sensors to improve the accuracy of the model and the state estimation, respectively, to provide simplified kinematics and enhanced robustness of the model at the expense of additional hardware complexity. Other uses of additional sensors such as indoor GPS [ 45 ] and camera-based systems [ 46 ] have also been explored to improve the state estimation algorithm, although they are often difficult to set up and calibrate. The use of observer-based techniques, such as the Unscented Kalman Filter [ 47 ] method, has also been proposed to integrate sensor measurements with kinematic models to reduce measurement noise and improve the reliability of the estimation results. Finally, machine learning methods, such as artificial [ 48 ] and feed-forward [ 49 ] neural networks, have recently been investigated to solve the forward kinematics and pose estimation problems of the Stewart platform. In addition to modeling and state estimation approaches, various control strategies have been proposed to optimize the Stewart platform under different operational requirements. These control strategies range from classical PID control [ 50 , 51 , 52 ] to advanced control methods such as model predictive control and adaptive control based on kinematic or dynamic models of the platform [ 53 , 54 , 55 ] . Other designs have also incorporated sliding mode control with feedforward velocity compensation [ 7 ] as well as the combination of optimization-based control [ 56 , 57 ] . More recently, data-driven methods based on neural networks [ 58 ] and reinforcement learning [ 59 ] have also been explored to improve the performance of the Stewart platform. Furthermore, deep Reinforcement Learning has also been used in [ 60 ] to determine optimal PID gains and validated in a Gazebo simulation environment that is integrated with the Robot Operating System (ROS). In [ 61 ] , a modal space