ARISTO Hand: Sensing-Driven Distal Hyperextension for Fine-Grained Manipulation

Technical Summary: ARISTO Hand: Sensing-Driven Distal Hyperextension for Fine-Grained Manipulation

Problem Statement
The manipulation of thin, small, and fragile objects (e.g., peeling tape, lifting plastic tabs, handling SD cards) remains a significant barrier to deploying robots in everyday human environments. Existing anthropomorphic robotic hands often lack the specific mechanical and sensing capabilities required for these interactions. Current designs typically prioritize either robust power grasping or soft tactile exploration, creating a functional gap for tasks requiring both rigid edge engagement and active fingertip adaptation. Furthermore, reliance on single sensing modalities (either purely proprioceptive or purely soft tactile) limits the ability to distinguish contact modes (e.g., fingernail vs. pad) or perceive incipient shear, particularly when contact geometry approaches kinematic singularities.

Methodology and System Design
The authors introduce the ARISTO Hand (Anthropomorphic, Robotic, Integrated-Sensing, Tendon-Operated), a tendon-driven robotic hand designed to unify advanced fingertip sensing with purposeful mechanical adaptation. The system is built on three core pillars:

1. Mechanical Architecture:

   • Active Distal Hyperextension: Unlike standard flexion-only hands, the distal joint supports active hyperextension. This allows the fingertip pad to flatten against planar surfaces and rotate past conventional limits, increasing the effective contact area for wedging and sliding.
   • Configuration-Independent Transmission: The hand utilizes a fully wrapped antagonistic tendon routing that preserves a constant effective moment arm ($r$) across the joint range. This ensures a linear relationship between actuator effort and joint torque ($\tau_j = r(f_{flex} - f_{ext})$), independent of configuration, which is critical for controlled distal interaction.
   • Fingernail Geometry: The rigid fingernail is angled downward and slightly curved. This pre-orientation allows the nail tip to engage edges and slide beneath thin objects without requiring excessive proximal interphalangeal (PIP) flexion, centralizing objects during pinch grasps.

2. Hybrid Sensing Architecture:

   • Rigid Nail-Mounted Sensor: A miniature six-axis force-torque (FT) sensor is mounted directly to the rigid fingernail. This provides high-frequency, localized force measurements for edge interactions, unaffected by the attenuation of compliant materials.
   • Soft Capacitive Tactile Array: A NARI-Touch 2×4 capacitive array is embedded in an elastomeric ventral pad. This monitors distributed surface pressure and incipient slip, supporting stable grasp acquisition.
   • Proprioceptive Estimation: The hand employs quasi-direct-drive (QDD) actuation with low reduction ratios (8:1) to minimize reflected inertia and friction, enabling reliable joint torque estimation from motor current. However, the authors note this is insufficient for distal wrench estimation near singularities.

Key Contributions
The paper presents three primary contributions:

1. A tendon-driven distal hyperextension mechanism enabling active fingertip alignment for stable thin-object manipulation.
2. An integrated rigid-soft fingertip sensing architecture that combines nail-mounted FT sensing for edge interaction with compliant tactile slip monitoring.
3. Experimental validation demonstrating that fingertip geometry and sensing modality jointly influence force transmission and proprioceptive observability.

Experimental Results
The authors conducted quantitative experiments to validate the system's performance:

• Transmission Linearity: The hand demonstrated a strong linear relationship between motor effort and applied force across MCP configurations ($-60^\circ$ to $+60^\circ$), with a mean coefficient of determination $R^2 = 0.982 \pm 0.007$.
• Hyperextension Efficacy: In pull-out force ablation studies across object thicknesses of 0.75 mm to 100 mm, active distal hyperextension increased the mean pull-out force for thin objects (1–20 mm) by 2.76× compared to standard flexion. For objects as thin as a credit card, hyperextension retained at least 98% of the nominal pull-out force, whereas standard flexion retained only 34–40%.
• Sensing Fidelity:
  • Edge Contact: The rigid nail-mounted FT sensor provided clean, step-like contact detection with a low noise floor ($\sigma = 0.0004$ Nm), whereas motor effort signals were noisy and obscured by stiction at low velocities.
  • Geometry Invariance: As the finger straightened (approaching kinematic singularities), motor-based force estimation degraded significantly, underestimating applied force by approximately 300% at shallow angles. The distal FT sensor maintained geometry-invariant performance, consistently tracking ground truth within $\pm 0.2$ N.
• Task Demonstration (SD Card Manipulation): The system successfully performed a multi-stage SD card extraction and insertion task. The rigid nail accessed recessed slots to trigger eject mechanisms and provided force feedback when the soft pad was geometrically occluded. Conversely, the hyperextending pad and soft sensor enabled stable acquisition and compliant alignment during insertion.

Significance and Claims
The paper argues that reliable fine-grained manipulation requires more than just dexterous kinematics; it demands the ability to actively shape and sense contact at the fingertip. The authors claim that:

• Sensing without geometric control cannot ensure proper contact conditions, while geometric control without detailed feedback is unsafe.
• Proprioceptive force estimation alone is insufficient for fine manipulation, particularly when fingers straighten into "poking" configurations where external forces become weakly observable through joint torque.
• The complementary nature of the rigid nail (for edge engagement and geometric constraints) and the hyperextending soft pad (for stable acquisition and alignment) is essential for overcoming the limitations of existing anthropomorphic hands.

The work concludes that the ARISTO Hand's hybrid approach effectively addresses the gap in manipulating small, thin objects by integrating mechanical adaptation with multi-modal sensing.