CDF-Glove: A Cable-Driven Force Feedback Glove for Dexterous Teleoperation
Authors: Huayue Liang, Ruochong Li, Yaodong Yang, Long Zeng, Yuanpei Chen, Xueqian Wang
Organizations: Center for Artificial Intelligence and Robotics, Shenzhen International Graduate School, Tsinghua University, Shenzhen 518055, China · PKU-Psibot Joint Lab, Peking 100871, China · Department of Advanced Manufacturing, Shenzhen International Graduate School, Tsinghua University, Shenzhen 518055, China
High-quality teleoperated demonstrations are a primary bottleneck for imitation learning (IL) in dexterous manipulation. However, haptic feedback provides operators with real-time contact information, enabling real-time finger posture adjustments, and thereby improving demonstration quality. Existing dexterous teleoperation platforms typically omit haptic feedback and remain bulky and expensive. We introduce CDF-Glove, a lightweight and low cost cable-driven force-feedback glove. The real-time state is available for 20 finger degrees of freedom (DoF), of which 16 are directly sensed and 4 are passively coupled (inferred from kinematic constraints). We develop a kinematic model and control stack for the glove, and validate them across multiple robotic hands with diverse kinematics and DoF. The CDF-Glove achieves distal joint repeatability of 0.4 degrees, and delivers about 200 ms force feedback latency, yielding a 4x improvement in task success rate relative to no-feedback teleoperation. We collect two bimanual teleoperation datasets, on which we train and evaluate Diffusion Policy baselines. Compared to kinesthetic teaching, the policies trained in our teleoperated demonstrations increase the average success rate by 55% and reduce the mean completion time by approximately 15.2 seconds (a 47.2% relative reduction). In particular, the CDF-Glove costs approximately US$230. The code and designs are released as open source at https://cdfglove.github.io/.
Figures & tables
Fig. 1: The CDF-Glove and the whole machine teleoperation system. The CDF-Glove is worn on the operator’s hand to collect hand joint data and provide haptic feedback.
Fig. 2: Overview of the CDF-Glove system. (a) Overall structural diagram of the CDF-Glove system. (b) Kinematic structure of the human hand system. (c) Overall kinematic layout of the CDF-Glove system.
Fig. 3: Arrangement of encoders at the base of the fingers.
Fig. 4: Exploded view of the single-finger structure.
Force Sensor Readings (N)
Vibration Feedback
Force Feedback
<0.1
—
—
0.1 ~ 0.5
✓(Waveform 1)
—
0.5 ~ 1
✓(Waveform 2)
—
>1
—
✓
TABLE I: Haptic Feedback Strategy of CDF-Glove
Fig. 5: Kinematic model for a single finger joint in the flexed position. (a) Schematic of joint angle calculation for an individual finger. (b) Force-feedback cable following calculation based on fingertip position localization.
Fig. 6: Overview of the complete experimental test system.
Fig. 8: Schematic diagram of the overall haptic feedback process. (a) represents the vibration signal when force is applied to the index finger. (b) represents the generation of cable-driven feedback.
Fig. 9: Results of three repeated positioning experiments for the index finger DIP joint.
Fig. 10: Haptic feedback processes information for the index finger. (a) shows the current variation curve of the RY-H1. (b) presents the waveform of the LRA and the servo angle, including mode switching.
Fig. 11: Bottle grasping experiment based on haptic feedback.
Condition
Success Rate
Average Completion Time (s)
✓F✓E
5/10
8.52
×F✓E
1/10
18.30
✓F×E
9/10
2.51
×F×E
7/10
3.11
TABLE IV: Performance for Grasping a Water Bottle
Fig. 12: Demonstrations of various tasks using the CDF-Glove. (a) Cola transportation task: removing a cola bottle from a box and placing it upright on the table. (b) Gesture demonstration: performing hand gestures based on the SDH model. (c) Manipulation task: rotating a light bulb with one or two fingers to turn it on. (d) Transport task: moving a spoon and the block on top of it from one bowl to another. (e) Manipulation task: playing the piano.
Fig. 13: Bimanual data collection. (a) Teleoperate cup stacking. (b) Cup stacking using the trained model. (c) Teleoperate plastic film roll transfer. (d) Plastic film roll transfer using the trained model.
Condition
Success Rate
Average Completion Time (s)
CDF (Cup Stacking)
10/10
14.74
KT (Cup Stacking)
4/10
28.91
CDF (Film Roll Transfer)
8/10
19.49
KT (Film Roll Transfer)
3/10
35.70
TABLE V: Success Rate and Completion Time for the Two Tasks
Dept. of Artificial Intelligence, Sogang University, Korea · Dept. of Electronic Engineering, Sogang University, Korea · Dept. of Mechanical Engineering, Sogang University, Korea +1