About REVEL

REVEL is an AI robotics company developing physical intelligence for general-purpose humanoid robots. We capture the force, dexterity and intent of human work with our Neural Gambit wearable, and use it to train RAI, the intelligence that powers our robots. Our R&D and engineering facilities are in Prague, Czech Republic. This role is on-site with our team in Prague.

The intent layer is REVEL's core differentiator: reading neuromuscular signals 30 to 100 ms before motion. We're looking for a scientist to decode high-density surface EMG into force and intent.

The role

We are looking for a scientist to develop methods for decoding high-density surface EMG (HD-sEMG) into continuous measures of human movement, force and motor intent.

A central research question is how much information about forthcoming movement and mechanical output can be extracted from neuromuscular activity before observable movement occurs. You will investigate this temporal relationship experimentally and develop models that can exploit it under realistic conditions.

This is a highly experimental role. You will work with real human data, noisy physiological signals, wearable hardware and evolving data-collection protocols, while translating scientific findings into models that can ultimately support robotic intelligence.

Responsibilities

3+ years of relevant professional experience

Develop signal-processing and machine-learning methods for decoding HD-sEMG into continuous force, joint kinematics and motor intent.

Characterise the temporal relationship between neuromuscular activity and mechanical movement, including the potential 30 to 100 ms anticipatory window.

Design experiments to quantify the reliability, information content and limits of anticipatory information in sEMG.

Develop preprocessing and signal-quality methods for multi-channel EMG, including filtering, artefact handling, normalization and channel-quality assessment.

Investigate the effects of sensor placement, electrode contact, motion artefacts, fatigue, physiological variability and session-to-session drift.

Evaluate model robustness across operators/subjects, sessions, movements and sensor configurations.

Develop methods for cross-operator, cross-session and cross-condition generalisation, including efficient calibration where necessary.

Design capture and labelling protocols together with the data-collection team.

Define data-quality criteria and experimental metrics that enable reliable downstream ML.

Collaborate closely with the robotics ML team to translate biosignal representations and decoded intent into inputs suitable for RAI.

Prototype and evaluate models in real-time or near-real-time settings where appropriate.

Communicate experimental results clearly and use them to guide the next iteration of the sensing, modelling and data-collection system.

Requirements

3+ years of relevant professional experience

Hands-on experience with EMG or electrophysiological biosignals, including signal acquisition, preprocessing and quantitative analysis.

Strong understanding of signal processing for noisy, non-stationary physiological signals.

Experience developing machine-learning models for time-series or sequential data.

Strong Python skills and practical experience with PyTorch or JAX.

Experience designing, executing and analysing controlled experiments.

Ability to reason quantitatively about temporal alignment, latency, prediction, signal quality and model generalisation.

Experience working with real-world biological data where signal characteristics vary across people and sessions.

Strong scientific judgement and the ability to distinguish genuine physiological information from artefacts, leakage and experimental confounds.

Bonus qualifications

PhD or equivalent research experience in biomedical engineering, neuroscience, electrical engineering, biomechanics, computer science, machine learning or a related field; alternatively, an MSc with 2 to 3+ years of highly relevant research or industry experience.

Experience with high-density sEMG / electrode arrays.