Actuation and Control of Passive Tendon-Driven Soft Extensions Through Rigid Manipulator Motion
This preprint turns a passive soft extension into a controllable tool by using an industrial arm's own motion to pull tendons anchored upstream. Simulations reached every target classified as feasible, while six UR10 experiments produced tendon-span changes up to 35 mm, demonstrating a route to compliant access without dedicated tendon motors.
Can an industrial manipulator control a passive soft extension through its existing joints while respecting end-effector constraints and geometric safety limits?
Rigid arms provide reach and precision but risk damaging confined surroundings. Soft extensions offer compliance, yet their separate motors and control hardware add integration work and mass. When an access opening restricts end-effector motion, the remaining joint motions must both actuate the tendons and position the soft tip.
Existing rigid-soft systems typically use separate tendon motors, including designs with fewer actuators. Passive adaptive grippers commonly change configuration through arm motion or contact rather than continuously regulating shape. Jacobian-based shape servoing estimates deformation from grasp-point motion; this work instead models deformation caused by changing exterior tendon spans.
Two tendons connect the extension to anchors on the upper arm and forearm. Arm motion changes their exterior spans and bends the extension. The model separates rigid transport from deformation, combining analytic kinematics with a locally fitted deformation map and online Broyden updates. Physical identification uses 13 probe poses, 10 for fitting and 3 for validation, with an external camera tracking the tip and body markers. Task-priority control uses motion that preserves the end-effector constraint; a quadratic program with control barrier functions handles clearance, aperture geometry, insertion and model-domain limits. Simulation code and an interactive browser implementation are publicly available.
Simulation used a UR10 with a 320 mm Cosserat rod. Three scenarios contained 9, 5 and 6 targets, with two targets in the last scenario deliberately infeasible. Every target classified as feasible met the arrival criterion of staying within 4 mm for 8 s. In an obstacle-barrier ablation, disabling enforcement caused 18.8 mm penetration; enforcement kept clearance above 10.7 mm while still reaching the target. Physical testing comprised six experiments on one UR10 with two handmade 475 mm foam extensions, each costing less than US$10, using parallel or helical tendon routing. Exterior spans changed by up to 35 mm. Validation RMS prediction errors for the four marker positions were 9 to 14 mm. Real targets counted as reached below 10 mm steady-state tip error, but an aggregate reached-target count was not reported in the paper. All reported barrier values remained nonnegative in simulation and physical tests.
The deformation map is local: simulated tip-prediction RMS error rose from 1.5 mm within ±10 mm spans to 5.5 mm at ±22 mm. Physical evidence covers one arm and six experiments, with virtual spherical obstacles and mostly model-derived targets. Some targets remained unreached because of active barriers or stationary points with nonzero error. Tendon count and anchor placement were not optimized; routing variants lack a controlled comparison. The controller does not regulate tension, contact force or stiffness. Exponential convergence requires the stated nominal assumptions and rank conditions; the weighted quadratic program only approximates strict task priority.
Inspection-system integrators could explore prototype use now for access through restricted openings, using an existing arm to actuate an inexpensive compliant extension. Product deployment would require validation of reachability, sensing, model accuracy and clearance in the intended environment. Commercial readiness and a deployment timetable are not reported in the paper; the evidence does not establish a 1 to 3 year product window.
The full text is not republished here because the paper's license does not allow it. Read the original on arXiv.