Cable-driven Continuum Robotics: Proprioception via Proximal-integrated Force Sensing

This paper proposes a novel proprioception method for micro-scale cable-driven continuum robots that integrates proximal cable tension and six-axis force/torque sensing with biomechanically inspired nonlinear modeling to enable accurate three-dimensional contact force perception and shape estimation, thereby overcoming limitations in miniaturization and sensor integration for safer clinical applications.

The Gist

Imagine you are trying to thread a needle while wearing thick, clumsy winter gloves. Now, imagine that needle is a tiny, flexible robot snake, only a few millimeters wide, and it needs to thread that needle inside a human body without a surgeon being able to see it directly.

This is the challenge of continuum robotics. These robots are like flexible, bendy straws used for minimally invasive surgery. They are amazing because they can wiggle through tiny holes, but they have a major blind spot: they can't "feel" what they are touching.

If a surgeon pushes this robot against a delicate organ, the robot doesn't know how hard it's pushing or exactly where it's touching. It's like trying to play a piano with your eyes closed and no sense of touch in your fingers.

The Problem: The "Tiny Robot" Dilemma

To give these robots a sense of touch, engineers usually try to glue tiny sensors all over the robot's body. But here's the catch: these robots are often smaller than a pencil lead (under 4mm). There is literally no room to stick sensors on them without making them too thick to fit inside the body.

Previous attempts to solve this were like trying to guess the weight of a suitcase by only looking at the handle. They could guess the shape, but they couldn't tell if the suitcase was heavy or light, or if it was being pushed from the side.

The Solution: The "Human Finger" Trick

This paper introduces a brilliant new way to give the robot a sense of touch, inspired by how your own fingers work.

Think about your finger when you pick up a cup:

1. Your Joints: Your joints have sensors that tell your brain how bent your finger is.
2. Your Tendons: Your muscles pull on tendons. Your brain feels the tension in those tendons to know how hard you are gripping.
3. The Result: Your brain combines the "joint angle" and the "tendon tension" to instantly know exactly how heavy the cup is and where your finger is touching it.

The researchers built a robot that mimics this biological teamwork.

How It Works (The "Back-of-the-Hand" Strategy)

Instead of putting sensors all over the robot's "body" (which is too small), they put all the sensing equipment at the base (the "wrist" or "hand" of the robot).

1. The "Joint" Sensor: At the base, they installed a super-sensitive 6-axis force sensor. This acts like the joint sensors in your finger, feeling the total push and pull from the robot's tip.
2. The "Tendon" Sensors: They also measure the tension in the cables that pull the robot to bend. This is like feeling the tension in your tendons.

The Magic Math:
The robot's "brain" (a computer algorithm) takes these two pieces of information—the tension in the cables and the force at the base—and runs a complex calculation. It's like a detective solving a puzzle:

• "If the cable is pulling this hard, and the base is feeling that much resistance, then the robot must be touching something here, with this amount of force."

They even added a clever trick inspired by human behavior: The "Hefting" Motion.
When you pick up a heavy box, you often wiggle it back and forth a little to get a feel for its weight. The robot does the same thing! If it's unsure about the friction or the force, it wiggles slightly. This "wiggling" helps smooth out the confusing friction inside the robot, allowing it to pinpoint the exact location of the touch with incredible accuracy.

The Results: A Super-Sensitive Snake

The team tested this on a tiny robot (3.5mm wide, about the size of a thick pen tip).

• Accuracy: It could tell where it was touching within 1 millimeter (about the thickness of a credit card).
• Force: It could measure the force of a touch as light as 1 gram (roughly the weight of a small paperclip).
• Speed: It does all this in real-time, fast enough for a surgeon to use during an operation.

Why This Matters

This is a game-changer for medical robotics.

• No More "Blind" Surgery: Surgeons will finally be able to "feel" through the robot, knowing exactly how hard they are pressing on a patient's tissue.
• Smaller is Better: Because they don't need to stuff sensors inside the robot, these robots can stay tiny and flexible, fitting into even the smallest spaces in the human body.
• Safer: It reduces the risk of accidentally poking or tearing delicate organs because the robot "knows" when it's touching something.

In short: This paper teaches a tiny, blind robot snake how to "feel" by listening to the tension in its muscles and the pressure at its base, just like your own finger does. It turns a clumsy tool into a delicate, sensitive instrument that could revolutionize how we perform surgery.

Technical Summary

Here is a detailed technical summary of the paper "Cable-driven Continuum Robotics: Proprioception via Proximal-integrated Force Sensing."

1. Problem Statement

Micro-scale continuum robots are increasingly vital for minimally invasive surgeries (e.g., neurosurgery, laparoscopy) due to their ability to navigate confined spaces. However, achieving human-like proprioception (the ability to perceive one's own state, including shape, position, and contact forces) remains a critical bottleneck.

• Limitations of Current Methods:
  • Sensor-based methods: Integrating sensors (e.g., FBG, strain gauges) directly onto the robot body is difficult in sub-millimeter diameters due to space constraints and structural integrity issues.
  • Model-based methods: Relying solely on internal actuation data (cable tension/displacement) often fails in complex scenarios with unknown contact locations or orientations, as the system becomes underdetermined.
  • Hybrid methods: While promising, many existing hybrid approaches still struggle with model inaccuracies, friction uncertainties, and the need for external tracking systems (e.g., vision), which limit their use in unstructured clinical environments.
• Core Challenge: How to achieve simultaneous, real-time estimation of 3D contact force, contact location, and robot shape in compact, micro-scale robots without embedding sensors along the robot's body.

2. Methodology

The authors propose a proximal-integrated force sensing framework inspired by the biological proprioception of human fingers (tendon-joint collaboration). The system relies entirely on sensors located at the robot's proximal end (base) and internal cable tension data, avoiding distal sensors.

A. Bio-Inspired Sensing Architecture

• Joint Proprioception: A six-axis Force/Torque (F/T) sensor is mounted at the robot's base to simulate joint receptors, measuring global forces and moments.
• Tendon Proprioception: Tension sensors are integrated into the drive cables to mimic muscle spindles, measuring local cable forces.
• Active/Passive Classification: The system distinguishes between "active" contact (robot moving into an obstacle) and "passive" contact (obstacle moving against the robot) to handle friction uncertainties.

B. Mathematical Modeling & Optimization

The core of the method is a quasi-static mechanical model combined with an optimization framework:

1. Force Decoupling: The contact force ($F_C$) is calculated by decoupling the base F/T sensor reading ($F_M$) and the input cable forces ($F_{in}$):
   $$F_C(t) = F_M(t) - F_{in}^{Ca}(t)$$
2. Mechanical Equilibrium: The model uses the Beam Constraint Model (BCM) and Capstan friction theory to describe the relationship between cable tension, beam deformation, and contact forces. It accounts for:
   • Mechanical nonlinearity (beam bending).
   • Material nonlinearity (Nickel-Titanium shape memory alloy properties).
   • Friction dynamics (static vs. kinetic friction based on motion direction).
3. Optimization Formulation: Instead of solving a complex multivariate nonlinear system, the problem is reformulated as a single-variable optimization problem.
   • Unknown: The contact location along the backbone ($s_c$).
   • Objective: Minimize the discrepancy between the set cable length and the calculated cable length required to satisfy the mechanical equilibrium given a specific contact location.
   • Process: The algorithm iterates through possible contact locations, calculates the resulting shape and forces, and finds the location that minimizes the cable length error.

C. Friction Uncertainty Mitigation

To address errors caused by internal friction and dynamic effects (acceleration/deceleration), the authors introduce a recalibration strategy:

• The robot performs a short reciprocating motion (back-and-forth) under load.
• This redistributes internal friction into a stable state, effectively "resetting" the friction model and significantly improving the accuracy of contact location estimation.

3. Key Contributions

1. Bio-Inspired Proximal-Only Sensing: A novel framework that achieves sub-millimeter shape perception and 3D force estimation using only base-mounted F/T sensors and cable tension feedback, eliminating the need for distal sensors.
2. Duality-Based Optimization: A method that leverages the duality between cable tension and base force to transform a complex, underdetermined perception problem into a stable, single-variable optimization problem.
3. Friction-Uncertainty Mitigation: An active/passive contact classification and reciprocating motion strategy that reduces contact location errors from ~1.67 mm to 0.21 mm by stabilizing internal friction.
4. Universality: The framework is validated across robots of varying diameters (1.7 mm to 6 mm) and mechanical structures, proving its scalability.

4. Experimental Results

The method was validated on a notched continuum robot (NCR) and tested under various conditions:

• Shape Estimation:
  • Achieved an average tip position error of 0.29 mm (under displacement control) and 0.41 mm (under force control).
  • Computational efficiency: ~98 Hz (MATLAB), suitable for real-time control.
• Force Perception:
  • Tip Contact: Average force estimation error of 0.93 g (max 4.02 g) over the full motion cycle.
  • Body Contact (Active): Average force error of 1.82 g (X) and 1.64 g (Z); location error 0.63 mm.
  • Body Contact (Passive): Average force error of 1.23 g (X) and 2.20 g (Z); location error 0.43 mm.
• 3D Force & Universality:
  • Successfully estimated 3D forces on robots with diameters of 1.7 mm, 3.5 mm, and 6 mm.
  • Force estimation errors remained below 1.0 g across all axes and sizes.
• Stability & Dynamics:
  • The reciprocating motion strategy reduced contact location error by 87% (from 1.67 mm to 0.21 mm).
  • The system maintained accuracy during dynamic movements (acceleration/deceleration), with errors stabilizing once motion ceased.
  • The framework successfully estimated resultant forces in multi-point contact scenarios, though it converges to an equivalent single-point contact for shape reconstruction.

5. Significance and Impact

• Clinical Applicability: By removing the need for embedded distal sensors, this method enables the use of proprioception in sub-millimeter surgical instruments where space is strictly limited.
• Safety and Precision: Accurate 3D force and location feedback allows surgeons to operate with greater confidence in delicate tissues, reducing the risk of perforation or damage.
• Computational Efficiency: The single-variable optimization approach is significantly faster and more robust than traditional iterative nonlinear solvers or data-heavy learning-based models, making it suitable for real-time haptic teleoperation.
• Future Outlook: The authors plan to integrate this framework into surgical robots for suturing tasks, providing haptic feedback to surgeons via teleoperation devices, thereby bridging the gap between robotic precision and human tactile perception.