1 DOF Robot Gripper Optimized for Haptic Feedback

• SO-101 Modification: Adapted the gripper to the SO-101 architecture to better suit high-fidelity haptic feedback requirements.
• Mechanical Specifications: The gripper weighs ~195 grams and features a maximum jaw opening of 140 mm. The transmission uses a low gear ratio of ~1.2.
• Quassi-Direct-Drive Actuation: Uses CubeMars GL40 II gimbal brushless motors without a gearbox, ensuring an instantaneous and crisp haptic response.
• Maximizing Backdriveability: The lack of a gearbox and the use of tendons, timing belts, and a linkage, minimizes friction and inertia. This is further enhanced by active Coulomb friction compensation in software to improve backdriving by the user.

• Dynamic Simulation: Developed an idealized dynamic model using Python (scipy.signal) to analyze the gripper's frequency response and optimize it for haptic feedback.
• Bandwidth Optimization: Evaluated transfer functions (Jaw Torque to Displacement, Velocity, and Acceleration) across various Gear Ratios (G) and Inertia Ratios (N) to maximize the system's velocity bandwidth.
• Design Validation: The Bode plot and contour analysis mathematically validated the choice of a quasi-direct-drive configuration (low G) to minimize reflected inertia (calculated at 1.98 × 10^-5 kg/m²) and mechanical damping, ensuring high transparency.

• Bilateral Teleoperation: Implemented a 1:1 Leader-Follower setup using MIT mode control. The leader mirrors the position of the follower while directly reflecting the follower's measured torque back to the user's hand.
• Force & Backlash Characteristics: Capable of delivering a pinch force of ~2.5 N before the tendons slip (an SO-101 hardware limitation). The measured static friction is ~1 N or ~0.06 Nm at the jaw, and total backlash is ~2.4 mm at the jaw tip (1.45 degrees at the pivot).
• Transparency: The low friction direct-drive system, combined with 1kHz SocketCAN control loops, allows the user to feel remote interactions with minimal interference from the hardware itself.
• Current-Based Homing: Designed a sensorless homing routine that detects physical hard-stops by monitoring current spikes when the jaws close.

• Control Loop: Operates a 1kHz control loop using MIT mode over a SocketCAN interface. This high-frequency loop applies position tracking with KD damping and feedforward torque commands for haptic feedback.
• ROS 2 Integration: The entire gripper and SO-101 arm system is wrapped in a unified ROS 2 node, allowing seamless integration with broader robotic software stacks and standardized message interfaces.

The project demonstrated a feasible way to add a gripper with usable haptics inexpensively to a 5 DOF robot arm. This can be used to grab both soft and hard objects and help train VLA models. Due to the use of low-cost manufacturing techniques like FDM 3D printing, there is more play in the system than is desirable. Future work would design a robot arm in tandem with the gripper to ensure the gripper transmission method is well supported.