GELLO: A General, Low-Cost, and Intuitive Teleoperation Framework for Robot Manipulators
TL;DR Build a half-scale, kinematically equivalent copy of a robot arm from 3D-printed parts and hobby servos for under $300, and stream its joint angles straight to the real arm. In a 12-person bimanual user study GELLO averaged 0.92 task success, against 0.72 for VR controllers and 0.63 for 3D mice.
What they do
Cheap teleoperation devices such as 3D mice and VR controllers command the end effector, so the operator cannot feel the robot’s joint limits, singularities or self-collisions. GELLO is a miniature replica of the target arm with the same kinematic structure, built from 3D-printed links and off-the-shelf servos used as joint encoders. The authors build one for the UR5, xArm7 and Franka Panda, each under $300 in parts, and open-source the hardware and software.
How it works
- Kinematically equivalent controller. The target arm’s DH parameters are scaled by α = 0.5. GELLO’s joint angles go straight to the robot as joint commands, so no inverse kinematics is needed.
- Servos as encoders. Dynamixel XL330-288T servos provide 12-bit encoders (0.088° resolution) and a ready communication protocol. Backdriving their high gear ratio adds damping. They are passive and give no force feedback.
- Joint regularization. Springs or rubber bands on the joints that carry the most gravity load (joints 2 and 3 on the UR version) stop the arm from sagging. Without them, holding the device still takes about 1.9 N at every height. With them, the restoring force grows as the arm moves away from the table.
Results
- User study: 12 volunteers with no teleoperation training, a bimanual UR5 station, five tasks, and device order randomized.
- Average success was 0.92 for GELLO, 0.72 for VR and 0.63 for 3D mice. The largest gaps were on towel folding (0.92 vs 0.58 for both) and USB insertion (0.83 vs 0.50 for VR and 0.58 for 3D mice).
- Summed over all tasks, GELLO had 1 timeout and 1 self-collision. 3D mice had 9 and 5, and VR had 5 and 5. On successful trials GELLO matched VR on the hat task and was faster on the others.
Limitations
The servos are too weak for force feedback, which limits contact-rich teleoperation. The study only tested novices after 5 minutes of practice per device. The paper evaluates teleoperation only and trains no policies on the collected demonstrations.
Figures
Abstract
Humans can teleoperate robots to accomplish complex manipulation tasks. Imitation learning has emerged as a powerful framework that leverages human teleoperated demonstrations to teach robots new skills. However, the performance of the learned policies is bottlenecked by the quality, scale, and variety of the demonstration data. In this paper, we aim to lower the barrier to collecting large and high-quality human demonstration data by proposing a GEneraL framework for building LOw-cost and intuitive teleoperation systems for robotic manipulation (GELLO). Given a target robot arm, we build a GELLO controller device that has the same kinematic structure as the target arm, leveraging 3D-printed parts and economical off-the-shelf motors. GELLO is easy to build and intuitive to use. Through an extensive user study, we show that GELLO enables more reliable and efficient demonstration collection compared to other cost efficient teleoperation devices commonly used in the imitation learning literature such as virtual reality controllers and 3D spacemouses. We further demonstrate the capabilities of GELLO for performing complex bi-manual and contact-rich manipulation tasks. To make GELLO accessible to everyone, we have designed and built GELLO systems for 3 commonly used robotic arms: Franka, UR5, and xArm. All software and hardware are open-sourced and can be found on our website: https://wuphilipp.github.io/gello/.
Summaries are written automatically from each paper's text. Figures and videos come from the paper and its project page; all credit goes to the authors.