Yonatan Hutabarat

When confronted with lethal threats, animals often display rapid, species-specific defensive behaviors. In contrast, human-specific defensive behaviors remain largely unknown and are challenging to investigate, partly due to ethical constraints in studying real life threat scenarios. In this project, we established a synchronized freely moving virtual reality (fm-VR) and motion-tracking system where participants forage in naturalistic environments while evading biologically relevant lethal threats. Across multi-site experiments, we identify a dominant human escape sequence: initial head turn toward the threat, whole-body rotation in the same direction, escape initiation using ipsilateral foot first, followed by forward locomotion toward shelter. Because the immersive nature of our virtual environment may reduce awareness of physical surroundings, we developed a validated kinematic stopping-distance model to predict collision risk and optimize physical safety buffers from the nearest walls. Together with low psychological distress and minimal cybersickness, this framework enables safe and ecologically valid investigations of human behavior, shifting psychological research toward studying behavior as continuous, embodied action.

Related Publications

• L. Kornemann, Y. Hutabarat, H. Liu, M. Pensel, U. D. Serratos Hernandez, J. K. Sporrer, S. Zabbah, D. R. Bach, “Running in Tight Spaces: A Large-scale Assessment of Physical and Psychological Safety Risks in Freely Moving Virtual Reality,” PsyArXiv (2026).
• Y. Hutabarat, J. K. Sporrer, J. Brookes, S. Zabbah, L. Kornemann, P. Domenici, D. R. Bach, “Human escape in freely moving virtual reality follows a structured movement pattern shaped by threat and context,” bioRxiv (2025).
• U. D. S. Hernandez, Y. Hutabarat, L. Kornemann, D. R. Bach, “Human movements during unsuccessful threat encounters,” Gait & Posture (2025).
• Y. Hutabarat, L. Kornemann, U. D. S. Hernandez, J. K. Sporrer, J. Brookes, S. Hall, S. Zabbah, D. R. Bach, “Biomechanical constraints on escape from threat in virtual reality: Preliminary findings,” Gait & Posture (2023).

Related Tools

• VR Safety model (PsyArXiv 2026)
• Machine vision camera capture (acquisition software, CAIAN 2023)

Continuous, out-of-lab gait assessment requires minimal sensor hardware capable of capturing complex biomechanical dynamics across varying locomotion modes. Here, we established and benchmarked a lightweight wearable framework that utilizes as few as two foot/shoe-mounted IMUs to reliably extract rich features of gait events, spatio-temporal characteristics and their derivatives as gait indices across various walking conditions. Further, we demonstrated the extent to which such a framework can be utilized not only work walking gait but also naturalistic running gait in out-of-lab setting, as well as potential clinical application in assessing dual-task gait for older adults. Finally, we further extent the framework to include machine learning-based classification that can predict GRF from sparse sensor configuration as well as for classification of different gait phases using LSTM and TCN, respectively. Taken together, this project provides an end-to-end framework for continuous gait assessment in out-of-lab settings using minimal wearable hardware.

Related Publications

• Review
• Y. Hutabarat, D. Owaki, M. Hayashibe, “Recent Advances in Quantitative Gait Analysis using Wearable Sensors: A Review,” IEEE Sensors Journal (2021).
• Benchmark study
• Y. Hutabarat, D. Owaki, M. Hayashibe, “Quantitative Gait Assessment with Feature-Rich Diversity Using Two IMU Sensors,” IEEE Transactions on Medical Robotics and Bionics (2020).
• Proof of concept: running and dual-task gait
• Y. Hutabarat, D. Owaki, M. Hayashibe, “Seamless Temporal Gait Evaluation during Walking and Running Using Two IMU Sensors,” IEEE EMBC (2021).
• Y. Hutabarat, D. Owaki, M. Hayashibe, “Temporal Variation Quantification During Cognitive Dual-Task Gait Using Two IMU Sensors,” IEEE EMBC (2022).
• Machine-learning extensions
• Y. Guo, Y. Hutabarat, D. Owaki, M. Hayashibe, “Speed-Variable Gait Phase Estimation During Ambulation via Temporal Convolutional Network,” IEEE Sensors Journal (2024).
• S. Sakamoto, Y. Hutabarat, D. Owaki, M. Hayashibe, “Ground Reaction Force and Moment Estimation through EMG Sensing Using LSTM Network during Posture Coordination,” Cyborg and Bionic Systems (2023).

Related Tools

• Temporal Gait Analysis (demo software, IEEE EMBC 2021)

Restoring natural locomotion in transfemoral amputees requires prosthetic control architectures capable of coordinating user intent with non-linear actuator dynamics. In this project, we propose a hierarchical control framework for a semi-active prosthetic knee driven by a magnetorheological (MR) damper. The architecture integrates a high-level intent recognizer to classify behavioral transitions, a mid-level feedforward neural network to generate continuous joint impedance and reference trajectories, and an adaptive low-level actuator controller. To overcome the non-linear hysteresis of MR fluid without explicit system dynamic models, the low-level controller employs model-free reinforcement Q-learning augmented with an exponential multi-horizon reward shaping function. Across variable walking speeds (2.4–5.4 km/h), this control structure achieves rapid convergence and sub-1% normalized root mean squared tracking error. This unified multilevel paradigm bridges biomechanical intent with adaptive, data-driven damping control for wearable bionic systems.

Related Publications

• Y. Hutabarat, K. Ekkachai, M. Hayashibe, W. Kongprawechnon, “Reinforcement Q-Learning Control with Reward Shaping Function for Swing Phase Control in a Semi-Active Prosthetic Knee,” Frontiers in Neurorobotics (2020).