PACT: Pose-Robust Actuation Co-Design for Tendon-Driven Robotic Hands

Prashant Kumar, Rachael Granberry, Taylor Niehues, Andrew A. Stanley, Gabe Pirie

November 2026

TL;DR PACT jointly picks tendon routing, moment arms and capstan sizes for a tendon-driven finger, so that one design covers the required fingertip forces even at its weakest pose while keeping 5 Hz joint bandwidth. For a human-sized four-DoF finger, it raises worst-pose task-force coverage from 0.24 to 0.88 over a hand-designed baseline. On the built prototype, measured force falls about 24.5% short of prediction because of tendon friction.

What they do

Tendon-driven hands move their actuators into the palm and forearm to fit a human-sized envelope. But routing, moment arms and capstan sizes together set both fingertip force and joint speed, and a design tuned at one pose can be weak at another. PACT selects all three jointly against a library of poses, and the authors build and test the resulting finger.

How it works

Exact force sets. At each pose, the feasible fingertip forces must satisfy every joint-equilibrium equation within actuator limits. They are solved as linear programs instead of with a pseudoinverse projection, which can overestimate capability.
Bandwidth constraint. A drivetrain model with torque and speed limits and reflected rotor inertia requires at least 5 Hz of isolated flexion/extension at each joint.
Max-min search. Routing signs and actuator assignments are enumerated. Moment arms and capstan radii are Sobol-sampled and refined with Nelder-Mead to maximize task-force coverage at the weakest of eight poses (pinches, tripods, extended reach, deep flexion, abducted pinch).
Resulting layout. An asymmetric N+1 design: one wrist-mounted high-torque QDD drives a shared flexion tendon, three small palm-mounted QDDs drive differential tendons, and a fifth actuator handles abduction.

Results

Worst-pose coverage rises from 0.24 (a baseline routed by hand for 15 N peak fingertip force) to 0.88, every pose improves, and the 5 Hz bandwidth holds at all eight poses. Single-pose designs reach full coverage at their own pose but lose it elsewhere.
On the prototype, measured fingertip force at one pose was about 24.5% below prediction, with a 14.82° direction error. Estimated torque losses grow from 14% at the MCP to 24% at the PIP and 32% at the DIP.
Re-scoring with these losses keeps all eight poses force-feasible (mean coverage loss 0.03), but worst-pose coverage drops to 0.75.

Limitations

The model ignores friction, tendon compliance, backlash and pretension, and the prototype shows that they matter. Force was validated at one pose at 50% of peak torque, and MCP extension bandwidth sits exactly at the 5 Hz limit. The search is heuristic. Only a single finger was built, so whole-hand grasping is untested.

Figures

From Figure 1: The PACT framework. A pose library, bandwidth constraints and actuator properties go in. Feasible forc... — From Figure 1: The PACT framework. A pose library, bandwidth constraints and actuator properties go in. Feasible force sets are shaped across the poses under worst-joint bandwidth limits, and the tendon routing, moment arms and capstan sizes come out.

From Figure 2: Actuation layout of the four-DoF finger. Flexion and extension of the MCP, PIP and DIP joints use an N... — From Figure 2: Actuation layout of the four-DoF finger. Flexion and extension of the MCP, PIP and DIP joints use an N+1 arrangement of three small tendons and one shared tendon M, while MCP abduction/adduction is actuated separately.

From Figure 3: Flexion/extension-plane cross-sections of the shared design's feasible fingertip-force sets at all eig... — From Figure 3: Flexion/extension-plane cross-sections of the shared design's feasible fingertip-force sets at all eight poses, drawn over the required task-force regions.

From Figure 6: The prototype finger with its low- and high-torque QDD actuators and capstan pulley, and the test setu... — From Figure 6: The prototype finger with its low- and high-torque QDD actuators and capstan pulley, and the test setup with a fixed finger base and a six-axis load cell at the fingertip.

Abstract

Humanoid robotic hands must provide sufficient force, speed, and backdrivability within a human-scale envelope. Tendon-driven transmissions address this challenge by relocating actuators to the palm and forearm, but routing paths, moment arms, and capstan dimensions jointly determine both force capability and joint speed, so selecting them sequentially satisfies one requirement at the expense of the other. This paper presents a pose-robust actuation co-design for tendon-driven robot hands (PACT) framework that optimizes tendon routing, moment arms, and capstan dimensions under pose-dependent force requirements and drivetrain-bandwidth constraints. A feasible-force formulation enforces actuator bounds and equilibrium at all joints, while a max-min objective maximizes task-force coverage at the weakest pose. Applied to a four-degree-of-freedom finger, the framework produces an asymmetric N + 1 tendon topology using quasi-direct-drive actuators. Across eight poses, the design achieves task-force coverage of 0.88-1.00 and a predicted 5 Hz minimum isolated-joint flexion/extension bandwidth within a 70th-percentile human-size envelope. Prototype measurements deviate from predictions by up to 24.5%, primarily due to unmodeled transmission losses. These results indicate that co-designing routing with the drivetrain can meet pose-dependent force requirements without relaxing the bandwidth constraint, paving the way for human-sized robotic hands with human-level performance.

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.

Matt Strong