Digital biomechanics and ergonomics apply computational methods, optimization, inverse modeling, and machine learning, to quantify and interpret human movement. At ETF Robotics, we combine musculoskeletal simulation, motion capture, and optimal control theory to reconstruct the mechanical and physiological underpinnings of human motion. Understanding how humans move is a prerequisite for building robots that can work safely alongside them: predicting skeletal-level motion trajectories during collaborative tasks enables robots to anticipate human intent and plan collision-free paths, while modeling muscle forces and joint loading is necessary to estimate the physical interaction forces a robot will exchange with a human partner. Our research addresses both clinical questions, such as understanding pathological gait and muscle coordination, and applied ones, such as objectively quantifying physical workload in manufacturing environments. Across all of these directions, the common thread is the same: accurate digital models of the human body are the foundation on which safe, human-centered robotic systems must be built.
Digital Biomechanics and Ergonomics
Research Focus Areas
We operate across the full pipeline from raw motion data to interpretable models of human movement strategy. Our research is built on a shared motion analysis framework and specifically features the following:
- Inverse optimal control for human motion analysis: Recovering cost functions that explain observed movement, applied to healthy and pathological gait, manual lifting, and jumping tasks. This approach treats the human motor system as an optimizer and recovers its implicit objectives from measurement data, enabling model-based predictions of motion across subjects and conditions.
- Inverse reinforcement learning for motion prediction: Learning reward functions from minimal motion observations to predict human movement trajectories. This data-efficient approach reduces the number of trials required for motion model identification and is applicable to anticipative robotic control.
- Musculoskeletal simulation and muscle force estimation: Solving the muscle redundancy problem to estimate individual muscle force contributions during complex motor tasks, with direct applications in clinical gait analysis, physical rehabilitation assessment, and wearable robotic device design.
- Physical ergonomics and workload quantification: Measuring muscle co-contraction, postural loading, and physical strain during repetitive industrial tasks using surface EMG and 3D body pose estimation. Results provide objective, quantitative criteria for workstation design and safety certification, with validated studies covering manual handling, drilling, and human-robot collaborative polishing.
- Computer vision for automated workplace safety: Developing and validating deep learning pipelines for automated recognition of unsafe worker postures, personal protective equipment (PPE) compliance monitoring, and robotic waste sorting. These systems process standard monocular video or IP camera footage and have been validated across real industrial tasks, including pushing/pulling, two-handed drilling, and cobot-assisted polishing, achieving classification accuracies between 85% and 98%.
- Motion capture and multi-sensor data acquisition: Integrating marker-based and markerless motion capture, including video-based 3D body pose and shape reconstruction via VIBE and SMPL-X parametric body models, instrumented force platforms, and surface electromyography for synchronized, high-fidelity human movement datasets. Markerless approaches enable ergonomic pose assessment directly from conventional IP camera footage, without specialist capture equipment.
Research Infrastructure
Our motion analysis and computational modeling facility supports data collection across all research areas.
- Marker-based optical motion capture systems and multi-axis instrumented force measurement platforms are not held in-house; instead, we access this infrastructure regularly through a strong collaborative relationship with the University of Belgrade Faculty of Sport and Physical Education (FSFV). Key collaborators include Prof. Dr. Dragan Mirkov and Jelena Aleksić, Research Assistant.
In-house equipment and software include:
- Surface electromyography (sEMG) acquisition system
- MuJoCo physics simulation environment (primary platform for musculoskeletal and optimal control modeling)
- OpenSim musculoskeletal modeling environment (tutorial and educational level)
- VIBE and SMPL-X pipelines for video-based markerless 3D body pose and shape estimation
- Custom motion analysis software pipelines (Python, MATLAB)
Related course: 13E054BMH – Biomechanics (undergraduate elective, School of Electrical Engineering, University of Belgrade). The course covers musculoskeletal kinematics and dynamics, motion capture analysis, bipedal gait, and the application of human biomechanics to humanoid robotics.