Characterization, Analytical Planning, and Hybrid Force Control for the Inspire RH56DFX Hand

TL;DR

The approach is modular, designed for compatibility with external object detectors and vision-language models for width&force estimation and high-level planning, and provides an interpretable and immediately deployable interface for dexterous manipulation with the Inspire RH56DFX hand.

Abstract

Commercially accessible dexterous robot hands are increasingly prevalent, but many remain difficult to use as scientific instruments. For example, the Inspire RH56DFX hand exposes only uncalibrated proprioceptive information and shows unreliable contact behavior at high speed (up to 1618% force limit overshoot). Furthermore, its underactuated, coupled finger linkages make antipodal grasps non-trivial. We contribute three improvements to the Inspire RH56DFX to transform it from a black-box device to a research tool: (1) hardware characterization (force calibration, latency, and overshoot), (2) a sim2real validated MuJoCo model for analytical width-to-grasp planning, and (3) a hybrid, closed-loop speed-force grasp controller. We validate these components on peg-in-hole insertion, achieving 65% success and outperforming a wrist-force-only baseline of 10% and on 300 grasps across 15 physically diverse objects, achieving 87% success and outperforming plan-free grasps and learned grasps. Our approach is modular, designed for compatibility with external object detectors and vision-language models for width & force estimation and high-level planning, and provides an interpretable and immediately deployable interface for dexterous manipulation with the Inspire RH56DFX hand, open-sourced at this website https://correlllab.github.io/rh56dfx.html.

Figures (9)

• Figure 1: A. 3D workspace visualization of the Inspire RH56 hand, showing that the coupled finger linkages constrain motion to arcs. The red bubbles represent the workspace of the 2-DoF thumb yaw and pitch actuation, highlighting the limited region of intersection with the other fingers and subsequent constrained grasping. B. The coupled kinematics require significant translation and rotation up to $\ang{49}$ for different width, here 110, 55, and 0mm are shown.
• Figure 2: Left: Step response to reach position "500" for different speed settings, showing 66ms latency followed by linear growth. The fingers stop when the set-points are reached with significant overshoot in the force domain (see below). Right: Force overshoot as a function of commanded speed and force setpoint ($N=20$, error bars = variance). Overshoot increases with higher speed, while hybrid control yields overshoot close to that of constant speed 25.
• Figure 3: Completion time vs. commanded speed under identical force setpoint for various speeds and two different control parameters. In Direct Speed Mode, a single constant speed is commanded for the entire motion. In the Hybrid scheme, the contact speed is fixed at 25. Markers show the mean over trials ($N=20$) and error bars indicate variance.
• Figure 4: We formulate three grasp strategies: iterative (left), reflex (center), and naive (right) closure. In the iterative approach, the hand rotates downward from a pre-grasp pose to reach the desired grasp pose. In the reflex approach, fingers remain open until the thumb makes contact with the object. In the naive approach, the fingers are treated like a parallel jaw gripper prior to closure, leading to ground collisions for small objects.
• Figure 5: A. For all evaluated objects, we assume high-fidelity perception of the major axis and centroid for antipodal grasping. Then, we define 10 offset positions with randomized orientation from which to initiate grasping B. We evaluate on 10 YCB and YCB-like (bottle) objects, adapted from FromPowerToPrecision2024. C. We additionally grasp 5 delicate objects to evaluate adaptive grasping.