Doctoral Student in Neuromotor Interfaces for Dexterous Robot Teleoperation (multimodal egocentric vision + EMG)

Doctorat / PhD Conditions : 100%, Zurich, fixed-term

The Sensing, Interaction & Perception Lab invites applications for a fully funded PhD position at ETH Zürich at the intersection of robotics, wearable sensing, signal processing, machine learning, and human-computer interaction. The goal of this PhD is to develop neuromotor interfaces for dexterous robot teleoperation using two sensing modalities: 1) surface electromyography (sEMG) measured at the wrist or forearm, with the aim of decoding subtle hand and finger activity, continuous movement, and motor intent in real time. 2) egocentric vision as a complementary input channel. While EMG provides information about human motor intent, egocentric cameras can provide context about the surrounding scene, manipulated objects, hand–object interactions, and task state. Combining these modalities creates opportunities for interfaces that understand both what the user intends to do and what is happening in the environment. The PhD will focus on the computational and sensing methods required to make such systems robust, generalizable, and effective in real interactive settings, including multimodal learning and reasoning across neuromotor and egocentric visual signals , while exploring applications in Mixed Reality and other interactive systems. Particularly beneficial: You have a background in (or experience with) prototyping sensing systems ( embedded systems , signal acquisition, processing, PCBs) Note: This is not a position in biomedical engineering and there is no health focus.

Wearable EMG provides a direct and unobtrusive way to sense muscle activity underlying hand and finger movements. This creates opportunities for interfaces that can recognize subtle actions, continuously estimate movement, and infer motor intent before or without large observable movements. Making such interfaces work reliably outside controlled settings remains a substantial research problem. EMG varies across users and recording sessions and is sensitive to electrode placement, contact conditions, movement, and other sources of noise. This motivates research in signal processing, temporal machine learning, representation learning, adaptation, and real-time inference. A second challenge is that neuromotor signals alone provide limited information about what the user is interacting with and why a particular movement is occurring. The project will therefore combine EMG with egocentric vision to capture objects, hands, contacts, affordances, and task state. This creates a multimodal research problem: learning representations that combine neuromotor and visual information and developing methods for reasoning about human intent, hand–object interaction, and the state of an ongoing manipulation task. For robotic teleoperation, these multimodal signals can support systems that jointly reason about what action the user intends, which object or target the action refers to, and how that intent should be translated to a robot with…