Research

Human-Inspired Control of a Whip

Manipulating flexible, underactuated objects like a whip remains a challenge in robotics, despite humans’ ability to skillfully handle such objects for tasks ranging from hitting targets to precise demonstrations like extinguishing a cigarette. This study modeled a 25-degree-of-freedom whip to investigate human strategies for striking a target. A human-inspired controller was developed, incorporating two movement strategies: “striking only” and “preparing and striking.” While the latter required more complex trajectory planning, it enabled successful strikes at greater distances. These findings provide insights into preparatory movements humans employ when manipulating objects, offering valuable principles for robotic control of underactuated systems.

Presented at the IEEE RAS EMBS International Conference on Biomedical Robotics and Biomechatronics (BioRob 2024): Conference Paper

Human Adaptation in Physical Human-Robot Interaction

Effective human-robot collaboration requires movement compatibility, especially in physical interactions. Building on prior findings that humanlike velocity profiles facilitate smoother interaction, this study examined whether humans can adapt to robots with non-biological trajectories. Participants tracked a robot tracing an ellipse with biological and non-biological velocity profiles, aiming to minimize interaction forces. Six participant groups practiced over three days; half received real-time visual feedback on force errors. Results showed that feedback-enabled groups reduced interaction forces for non-biological profiles, while no-feedback groups showed no improvement. Biological profiles saw no change, indicating a performance floor. These findings underscore the importance of human-like trajectories and augmented feedback in physical human-robot interaction, with implications for applications like surgical and industrial robotics.

Presented at the IEEE International Conference on Robotics and Automation (ICRA 2023): Conference Paper

A Novel 3D-Printed Device for Unconstrained Grip
Force Measurements

EA novel measurement device was developed that can measure user-applied grip force all around a cylindrical handle without constraining the grasp posture. The device used off-the-shelf and 3D-printed components, making it easy and low-cost to build and use. Through the use of flexures, the device mechanically sums the radial grip forces, up to 200N, applied to any locations along any of its load plates. It exhibited less than 1.5% of full-scale error after calibration and less than 1.5N variation when load was applied to different plates. The accuracy and versatility of the device affords numerous applications in research on human motor control and human robot interactions. A human experiment on force and position tracking was presented to demonstrate the utility of the device.

This work is currently under review for publication. For more information, check out the project’s Github repository.

Predicting missing motion capture kinematic data for complex nonlinear systems using a NARX network

Motion tracking is a vital tool in robotics and human motor control research. One of the common issues faced by motion capture systems is caused by occlusions of the area being tracked. This results in gaps in the recorded kinematic data. The missing kinematic data in small gaps (less than 1 second) could in most cases be recovered through simple interpolation techniques. However, as the gaps get longer, more sophisticated algorithms are needed to accurately recover the missing data. In this project, I developed a machine learning based approach to recover missing kinematic data. Using a NARX network, I was able to more accurately recover missing kinematic data compared to existing algorithms.

Presented at the International Symposium on Adaptive Motion of Animals and Machines (AMAM 2021): Extended Abstract

Effect of dimensionality in complex object manipulation

Manipulating a deformable nonlinear object is an extremely difficult feat for modern robots, yet humans can manipulate such an object (a bull whip) with their eyes closed. In this project, we examine the strategies human use to control highly complex and nonlinear objects. Using a dimensionality reduction technique known as principal component analysis (PCA), I found that humans manipulate the whip in a way that results in a very low dimensional system. Humans use a control strategy that results in a 2-dimensional system, which is significantly simplifies the manipulation task.

Presented at the Society for Neuroscience Conference (SfN 2021): Poster