GELLO
GELLO is a low-cost, 6 degrees of freedom robotic interface designed for seamless integration with Universal Robots (UR) systems. Built with 3D-printed components, its arm geometry follows Denavit-Hartenberg (DH) parameters to ensure accurate kinematic modeling and control.
This systems improves on the existing GELLO by:
Physical Components
Gripper
Precise grip control
Arm link
3D-printed
Based on scaled-down Denavit–Hartenberg (DH) parameters
Base (UR5e)
Support the master arm
Dynamixel motor
Position encoding
Position and Current modes
Brackets and Washers
Improve rigidity for force feedback
U2D2 board
Power and communication interface
Single arm
Prototype
Bimanual arm
Prototype
Control Flow
GELLO >> UR arm
Joint positions θ from Dynamixel encoders transmitted to UR arm
UR arm >> GELLO
Force-torque sensor measures interaction forces F at end effector and calculates corresponding torques to apply at each Dyanixels
Feedback Enhancements
Reduces sensor noise and produces smoother and stable force feedback
Increase baud rate to improve communication speed and reduces latency
Eliminates unstable rotational torque components and downscale forces
Remove motor induced current to reflect true external interaction forces
GELLO Build Guide
Hardware iterations
User Test
For our user testing, we recruited 5 participants from our industry partner company. Each participant completed 3 tasks using our GELLO prototype.
The tasks are adapted from the original paper, and performance is evaluated using the following metrics:
- Completion rate
- Time taken to complete
Results are compared against other teleoperation devices from the paper, namely the 3D SpaceMouse and VR.
Tasks
- Task 1: Pick & Place
- Task 2: Item Handover
- Task 3: USB Insertion
Procedure
- Each participant is given 3 minutes to familiarise themselves with both single-arm and bimanual GELLO operation.
- Task 1 : 45 seconds, performed using single-arm control.
- Task 2 : 90 seconds, performed using bimanual control.
- Task 3 : 90 seconds, performed using bimanual control.
Results
- Completion Rate
- GELLO is better across all 3 tasks
- Completion Time
- GELLO is better in Tasks 1 and 2, underperforms for Task 3 (due to different physical setup)
Acknowledgements
Team S22 would like to thank our capstone instructors, Dr. Sumbul Khan, Dr. Perry Lam, and Professor Franklin Anariba, for their valuable advice and guidance. We are also grateful to Belinda from the Centre for Writing and Rhetoric (CWR) for her assistance with our presentation and report.
We would also like to extend our thanks to our industry mentor, Dr. Zhu Hai Yue, Senior Scientist in Robotics at A*STAR, for his guidance and support throughout our project.
