Haptic fMRI : A Novel Five DOF Haptic Interface for Multi-Axis Motor Neuroscience Experiments Samir Menon 1 , Amaury Soviche 2 , Jananan Mithrakumar 1 , Alok Subbarao 1 , Hari Ganti 1 , and Oussama Khatib 1 . Abstract— We present a novel electromagnetically actuated Haptic fMRI Interface with five degrees-of-freedom (DOF), HFI-5. The interface uses two three DOF devices connected with a gimbal in a closed kinematic chain to achieve three- axis translation and two-axis rotation. To highlight the device’s design, we develop a taxonomy of similar devices and demon- strate why HFI-5’s design excels. We use a cross-correlation based novel analysis method to demonstrate that HFI-5 is transparent by showing that it does not affect unconstrained subject motions. This is due to its low friction (< 0.31 N) and lightweight design. Next, we show that HFI-5 can accurately track positions, velocities, and accelerations, in part due to a 7.5 kHz servo loop. Finally, we demonstrate that HFI-5 is fMRI- compatible and does not interfere with fMRI scans even while applying large torques with its electromagnetic motors. I. I NTRODUCTION Realizing the potential of functional magnetic resonance imaging (fMRI) for motor neuroscience requires the devel- opment of tools that enable the study of rich manipulation tasks in a controlled setting. Haptic interfaces are ideal to meet this challenge. They can recreate a variety of day-to- day motor tasks in virtual environments [1], [2], [3], [4] while allowing experimenters to precisely monitor motions and apply forces when required. Developing a functioning multi- axis haptic interface that works with high-resolution fMRI (mm, sec) [5], [6] promises to substantially expand the study of the brain’s sensorimotor regions [7]. Moreover, combining haptics with fMRI also promises dramatic improvements to motor rehabilitation [8], [9], [10], [11], [12] by allowing clinicians to observe how therapy affects neural circuits in real-time. Integrating such haptics with fMRI will allow us to step beyond classical motor neuroscience experiments, which have focused on non-interactive free-space motions. Research efforts have led to numerous haptic interfaces for fMRI. These rely on a variety of actuators such as elecro-active polymers [13], pneumatics [14], ultrasonics [12], [15], and hydraulics [16], [15]. Cables, with dynamic models to improve performance [17], are also promising. Functional interfaces include a shielded Phantom [18], [19], *This work was supported by BioX, Stanford University 1 S. Menon, J. Mithrakumar, A. Subbarao, H. Ganti, and O. Khatib are with the Stanford Robotics Laboratory, Department of Computer Science, Stanford University, Stanford, CA 94305, USA. 2 A. Soviche is with the Robotics and Autonomous Systems Department, Ecole Polytechnique F´ ed´ erale de Lausanne, Route Cantonale, 1015 Lausanne, Switzerland. Email: smenon@cs.stanford.edu, janamith@stanford.edu, hganti@stanford.edu, asubbarao@stanfordhealthcare.org, ok@cs.stanford.edu, amaury.soviche@epfl.ch A X Y Z B C 0 5% tNSR D Fig. 1. Haptic fMRI Interface 5. A, A subject operating HFI-5 in gimbal mode inside a 3T MRI scanner. B, The two planar parallelograms may be moved together to realise three-axis (X, Y, Z) motions. In addition, moving them anti-parallel along the X and Z axes produces Z and X rotations respectively (see axis frame). C, A volumetric rendering of a human’s brain demonstrates the volume spanned by an exemplar fMRI motor neuroimaging scan. D, Temporal noise measurements plotted on an inflated rendering of the subject’s brain. fMRI recordings were made while HFI-5 applied a series of random forces and torques. a wrist device [20], a pneumatic device [14], and a family of electromagnetically actuated devices [21], [22]. Present efforts are directed towards devices with more DOF, or towards achieving high-fidelity force control, natural motions [23] and uniform inertial properties across a large three- dimensional workspace. This paper presents a novel fMRI-compatible haptic inter- face, HFI-5, with five DOF. The device’s DOF are realized using two three degree-of-freedom device modules connected in a closed-chain kinematic structure (Fig. 1). It supports translations that span the MRI workspace, as well as two- axis rotations that span the human wrist’s ability to rotate. To highlight HFI-5’s design advances, we present a taxonomy of possible similar devices and discuss how its design is both pragmatic and effective. We demonstrate that HFI-5 is transparent to human operators: motions performed while operating the device are indistinguishable from free space motions. Moreover, HFI-5 can accurately monitor positions, velocities, as well as accelerations due to its fast control and sensing servo loop (at ~7.5 kHz). Finally, we demonstrate that HFI-5 is fMRI-compatible and does not affect the quality