Faster Than Muscle

Faster Than Muscle: Executing Action based on Voluntary Nervous Signals

The collaboration project with the leading author, Gabriela Vega, and Paul Strohmeier (Max Planck Institute for Informatics).

A voluntary motor command travels from the brain to the forearm muscle; EMG onset triggers on-body, virtual and robotic actions 51 ms, 82 ms and 19 ms faster — A voluntary motor command takes up to 100 ms to produce visible movement (the electromechanical delay, EMD). Faster Than Muscle detects EMG onset and executes the action within that window, on three targets.

Abstract

Human augmentation enables actions that surpass biological limits, but earlier intervention in the action loop can undermine user agency. We introduce human-in-the-loop action acceleration, a design approach that anchors action acceleration to voluntary initiation. Using forearm EMG onset, the system triggers actuation within the electromechanical delay (EMD), enabling earlier execution without relying on predictive inference. We implement this approach in Faster Than Muscle, a system with three instantiations: (1) on-body finger actuation achieving a 51 ms speedup, (2) EMG-triggered software input for a shooting task (82 ms speedup), and (3) bionic-hand actuation for catching a falling object (19 ms speedup). Across two studies (temporal binding experiment and qualitative interview study) and accompanying system evaluations, we show that EMG-triggered acceleration improves performance in time-critical tasks, while users retain a sense of agency. These results show that actions can be accelerated beyond biological limits while remaining grounded in voluntary initiation, and outline a design space spanning execution targets and feedback timing.

Three acceleration targets

A forearm EMG sensor detects the onset of voluntary muscle activation, and the same pipeline sends the action to one of three targets:

On-body (about 51 ms faster): an electromagnet moves the user’s own finger to press a key, so that the feel of the press, its sound, and what the user sees all arrive earlier. It builds on EMAS.
Virtual (about 82 ms faster): a key event reaches the game Shoot or Die before the physical press registers, while the user still presses the key as usual.
Robotic (about 19 ms faster): a bionic hand starts closing before the user’s own fingers visibly move, in time to catch a falling pen.

System overview: forearm EMG, a signal amplifier and a microcontroller drive an electromagnet, a software key event, or a robotic hand — Forearm EMG is amplified and sampled at 2 kHz; on onset, the microcontroller drives (a) an electromagnet under the finger, (b) a software key event, or (c) a robotic hand.

Two photos, labelled Normal and Faster Than Muscle: on the left a pen falls past a hand that has not closed yet; on the right a robotic hand has grasped it — Pen-catching task. Left: with a normal hand, the pen falls before the hand can close. Right: with Faster Than Muscle, EMG onset makes the robotic hand grasp the pen before biological movement completes.

Charts of the performance gains: on-body 74.0 to 23.0 ms, virtual 324.9 to 243.3 ms, robotic finger leading the human finger, summary 51, 82 and 19 ms — Performance gains on the three targets.

Sense of agency

In a temporal binding study with the Libet clock (27 participants), the accelerated presses the participants initiated themselves were rated far higher in agency than the same movement triggered at random, though lower than unassisted presses. In a gaming study (8 participants), EMG-triggered input raised the hit rate from about 54% to 83%, and participants reported a stronger sense of control, even though what they saw and heard came before the feel of their own key press.

Faster Than Muscle: Executing Action based on Voluntary Nervous Signals
Gabriela Vega, Johanna K. Didion, Reshma Ann Daniel, Nihar Sabnis, Daisuke Tajima, Shunichi Kasahara, and Paul Strohmeier. In The 39th Annual ACM Symposium on User Interface Software and Technology (UIST ’26), November 2–5, 2026, Detroit, MI, USA.

Figures from the paper (Vega et al., UIST ’26), licensed under CC BY 4.0.

News: Paper Accepted to UIST 2026
Related projects: Preemptive Action, Wired Muscle, EMAS, Whose touch is this?