Motion Control of the CyberCarpet Platform

Alessandro De Luca, R. Mattone, Paolo Robuffo Giordano, Heinz Ulbrich, Martin Schwaiger, Michael Van den Bergh, Esther Koler-Meier, Luc Van Gool

TL;DR

This paper introduces the CyberCarpet, a novel platform for unconstrained virtual reality locomotion, and presents a robust motion control system. The system uses a combination of feedback and feedforward control, informed by a disturbance observer, to keep users centered on the platform while counteracting their natural movements, ensuring a more immersive VR experience.

What Problem Does This Paper Address?

The core challenge addressed is enabling users to walk freely and naturally within virtual reality environments without physical constraints, while keeping them within the confines of a limited physical space. Existing solutions often involve cumbersome harnesses or restricted movement. The CyberCarpet aims to provide omnidirectional locomotion by using a linear treadmill with a ball-array carpet mounted on a turntable. The key difficulty lies in controlling this system, which has a nonholonomic constraint, to precisely counteract a user's voluntary motion and maintain their position at the platform's center, all while ensuring the experience feels natural and avoids perceptual disturbances.

Methods

The paper details a two-part control strategy:

  • Velocity-Level Control: This is the primary control approach. It employs an input-output feedback linearization technique to decouple and linearize the system's kinematics. To handle singularities and ensure global stabilization, a modified, singularity-free control law is proposed. Crucially, a disturbance observer is integrated to estimate the user's intentional velocity, which is then used in a feedforward term to actively counteract the user's movements. This feedforward action is key to keeping the user centered.
  • Acceleration-Level Control: For a more advanced approach, the authors extend the velocity-level control to the acceleration level. This allows for explicit consideration of acceleration limitations imposed by actuators and user perception, and also enables the computation of apparent accelerations felt by the user, which is important for dynamic analysis. This is achieved using cascaded systems theory.

The system also relies on a markerless visual tracking system using an overhead camera and particle filters to accurately determine the walker's position and orientation on the platform.

Datasets

The paper does not explicitly mention or introduce new datasets for training. Instead, it focuses on experimental validation using a small-scale prototype of the CyberCarpet. The "dataset" for evaluation consists of recorded trajectories and control commands generated during various motion tasks performed by a mobile robot mock-up of a walking user on the platform.

Key Insights and Ideas

  • Nonholonomic Constraint as an Analogy: The paper highlights an interesting duality between the CyberCarpet's control problem and that of nonholonomic wheeled mobile robots, where the user's movement on the platform is analogous to the ground, and the platform itself acts as the controlled robot.
  • Disturbance Observer for Naturalness: The integration of a disturbance observer to estimate and counteract the user's intentional velocity is a critical insight. This feedforward compensation is what allows the platform to "follow" the user's natural movements, making the experience feel more intuitive and less like being passively moved.
  • Singularity-Free Control: Addressing the kinematic singularities inherent in the system is crucial for robust control, especially near the platform's center. The proposed singularity-free control law is a key technical contribution.
  • Acceleration-Level Control for Dynamics: Extending to acceleration-level control is important for understanding and mitigating the dynamic forces a user might feel, which is vital for a truly immersive and comfortable VR experience, especially in a full-scale system.

Key Results

  • The implemented velocity-level control, combining feedback and feedforward, successfully kept the user (emulated by a mobile robot) close to the platform center during various motion tasks, including straight lines, circular paths, and square paths.
  • The feedforward action based on the disturbance observer significantly improved performance compared to pure feedback control, especially for smoother motions, by effectively compensating for intentional user velocity.
  • The visual localization system, using an elliptical model with a head region, was able to track the user's position and orientation with reasonable accuracy.
  • Experimental results demonstrated that the system could handle intentional velocities up to 0.25 m/s.

Limitations and Future Work

  • Measurement Noise and Sampling Frequency: The authors acknowledge that noise in image processing and the relatively low sampling frequency of the visual localization system (up to 17 Hz, but often tested at 10 Hz or lower) led to discrepancies in trajectory tracking and control performance.
  • Walker Orientation Not Used in Control: While the visual system estimates the walker's orientation, it was not utilized in the control law. This prevented proactive compensation for intentional turns, leading to delays in re-centering.
  • Chattering near the Origin: While the singularity-free control law mitigates chattering, some residual effects were observed near the platform's center, particularly with the angular feedforward term.
  • Future Work: The paper suggests building a full-scale CyberCarpet to allow users to walk at normal speeds within VR. Further investigation into the perceptual effects of combined linear and angular motion during transients is also needed to refine control design constraints. The authors also mention that acceleration control, while explored, was not strictly necessary for the small-scale prototype but is crucial for a full-scale system.

Summary updated

Abstract

The CyberCarpet is an actuated platform that allows unconstrained locomotion of a walking user for Virtual Reality exploration. The platform consists of a linear treadmill covered by a ball-array carpet and mounted on a turntable, and is equipped with two actuating devices for linear and angular motion. The main control objective is to keep the walker close to the platform center in the most natural way, counteracting his/her voluntary motion while satisfying perceptual constraints. The motion control problem for this platform is not trivial since the system kinematics is subject to a nonholonomic constraint. In the first part of the paper we describe the kinematic control design devised within the CyberWalk project, where the linear and angular platform velocities are used as input commands and feedback is based only on walker's position measurements obtained by an external visual tracking system. In particular, we present a globally stabilizing control scheme that combines a feedback and a feedforward action, based on a disturbance observer of the walker's intentional velocity. We also discuss possible extensions to acceleration-level control and the related assessment of dynamic issues affecting a walker during his/her motion. The second part of the paper is devoted to the actual implementation of the overall system. As a proof of concept of a final full-scale platform, the mechanical design and realization of a small-scale prototype of the CyberCarpet is presented, as well as the visual localization method used to obtain the human walker's position on the platform by an overhead camera. To validate the proposed motion control design, experimental results are reported and discussed for a series of motion tasks performed using a small tracked vehicle representative of a moving user.
BibTeX
@article{luca2013motion,
  author = {Alessandro De Luca and R. Mattone and Paolo Robuffo Giordano and Heinz Ulbrich and Martin Schwaiger and Michael Van den Bergh and Esther Koler-Meier and Luc Van Gool},
  title = {{Motion Control of the CyberCarpet Platform}},
  journal = {IEEE Trans. Control. Syst. Technol.},
  year = {2013},
  doi = {10.1109/tcst.2012.2185051},
}
Figure