Control the Soft Robot Arm with its Physical Twin (2503.17227v1)

Abstract: To exploit the compliant capabilities of soft robot arms we require controller which can exploit their physical capabilities. Teleoperation, leveraging a human in the loop, is a key step towards achieving more complex control strategies. Whilst teleoperation is widely used for rigid robots, for soft robots we require teleoperation methods where the configuration of the whole body is considered. We propose a method of using an identical 'physical twin', or demonstrator of the robot. This tendon robot can be back-driven, with the tendon lengths providing configuration perception, and enabling a direct mapping of tendon lengths for the execture. We demonstrate how this teleoperation across the entire configuration of the robot enables complex interactions with exploit the envrionment, such as squeezing into gaps. We also show how this method can generalize to robots which are a larger scale that the physical twin, and how, tuneability of the stiffness properties of the physical twin simplify its use.

• The paper proposes a teleoperation method using a back-drivable physical twin to intuitively control the full configuration space of soft robot arms by mapping the twin's state (e.g., tendon lengths) to the executor's actuators.
• This approach simplifies controlling compliant robots by bypassing complex inverse kinematics models and has been demonstrated to work with scaling, allowing a smaller twin to control a larger executor.
• The method was shown effective for tasks leveraging compliance, such as squeezing through confined spaces, although successful implementation depends on accurate twin fabrication, sensing, and managing dynamic mismatches.

This work proposes a teleoperation methodology for soft robot arms centered around the utilization of a kinematically and dynamically similar 'physical twin' as the input interface (2503.17227). The primary objective is to enable intuitive control over the entire configuration space of compliant soft manipulators, moving beyond endpoint control paradigms common in rigid robotics, and facilitating tasks that leverage the inherent compliance of soft systems.

Physical Twin Teleoperation Mechanism

The core concept involves employing an identical, or closely matched, physical replica of the soft robot arm, designated as the 'demonstrator' or 'physical twin'. This twin is specifically designed to be back-drivable, particularly in the context of tendon-actuated soft robots. A human operator physically manipulates the twin robot into desired configurations. The key innovation lies in using the state of the twin directly to command the state of the primary 'executor' robot.

For tendon-driven systems, the back-drivability allows the operator's manipulation to directly alter the lengths of the tendons within the twin. Sensors integrated into the twin measure these tendon lengths in real-time. These measured lengths, representing the configuration of the twin, are then used as the command signals for the actuator controllers of the executor robot. This creates a direct mapping:

$L_{executor, i} = f(L_{twin, i})$

where $L_{executor, i}$ is the target length for the $i$-th tendon of the executor robot, $L_{twin, i}$ is the measured length of the corresponding $i$-th tendon in the physical twin, and $f$ is the mapping function. In the simplest case, where the twin and executor are identical, $f$ can be a direct identity mapping ($L_{executor, i} = L_{twin, i}$), potentially with scaling factors if the robots differ in size.

This approach bypasses the need for complex inverse kinematics models, which are often challenging to derive and computationally expensive for high-DOF, continuum soft robots. It intrinsically captures the full body configuration, as the tendon lengths collectively define the robot's shape.

Implementation Considerations

Implementing this physical twin strategy necessitates several key considerations:

1. Physical Twin Design: The twin must accurately replicate the kinematics and, ideally, the passive dynamics (stiffness, damping) of the executor robot. Crucially, it must be designed for low-friction back-drivability, allowing easy manipulation by a human operator without significant resistance from the actuation mechanism itself. For tendon robots, this implies careful routing and low-friction materials.
2. Sensing: Precise and high-frequency measurement of the twin's configuration variables (tendon lengths in this case) is critical. This could involve encoders on pseudo-actuators or take-up spools, potentiometers, or other length/displacement sensors integrated into the twin's structure. Sensor noise and resolution directly impact the fidelity of the teleoperation.
3. Control Loop: A control loop is required on the executor robot to drive its actuators (e.g., motors pulling tendons) to match the target tendon lengths received from the twin. Standard position controllers (e.g., PID) can be employed for each tendon actuator, taking $L_{executor, i}$ as the setpoint. Latency in sensing, communication, and actuation must be minimized to ensure stability and intuitive control feel.
4. Calibration: Initial calibration is required to establish the zero points and scaling factors for the tendon length measurements on both the twin and the executor, ensuring correspondence between their configurations. Mismatches in dimensions or material properties between the twin and executor can lead to tracking errors.
5. Stiffness Tuning: The paper mentions the benefit of tunable stiffness in the physical twin. This could be implemented via adjustable passive elements or potentially low-bandwidth active control on the twin itself. Lowering the twin's stiffness can reduce the physical effort required from the operator, enhancing usability, especially during prolonged operation or manipulation of complex configurations.

A potential system architecture could look like this:

+-------------------+ Operator +----------------+ Tendon Lengths +---------------------+ Tendon Commands +-----------------+ | Human Operator | Manipulates> | Physical Twin |----------------->| Communication Link |------------------>| Executor Robot | | | <Feedback | (Back-drivable,| | (e.g., ROS, UDP) | | (Actuator Ctrl) | | | (Proprio.) | Sensorized) | | | | | +-------------------+ +----------------+ +---------------------+ +-----------------+ | | | Measures Tendon Lengths L_twin | Executes L_executor = f(L_twin) V V +-----------------------+ +-----------------------+ | Tendon Sensor System | | Tendon Actuator System| +-----------------------+ +-----------------------+

Demonstrated Capabilities and Applications

The paper demonstrates the efficacy of this approach for tasks requiring whole-body interaction with the environment. A key example presented is guiding the soft executor arm to squeeze into confined gaps. This task leverages the robot's compliance and requires precise control over its entire shape, which is intuitively provided by manipulating the physical twin. The operator can physically feel the interaction constraints when manipulating the twin (if stiffness is matched or appropriately scaled) or visually guide the executor based on the twin's posture.

This method offers a potential pathway for intuitive programming of complex, compliance-driven tasks for soft robots in unstructured environments, such as:

• Manipulation in Clutter: Navigating and grasping objects within tightly packed spaces.
• Inspection Tasks: Conforming the robot body to surfaces or navigating through intricate structures.
• Human-Robot Collaboration: Allowing a human to directly guide a soft robot's compliant interactions.

Scaling and Stiffness Adaptation

The research indicates that the method is not strictly limited to identical 1:1 scale twins. It demonstrates generalization where the executor robot is larger than the physical twin. This requires incorporating appropriate scaling factors into the mapping function $f$:

$L_{executor, i} = S \cdot L_{twin, i}$

where $S$ is the scale factor (e.g., ratio of executor size to twin size). This capability significantly enhances the practicality of the approach, as it allows for smaller, potentially more manageable input devices to control larger operational robots. However, dynamic mismatches might become more pronounced with significant scaling.

Furthermore, the tunability of the physical twin's stiffness is highlighted as a feature that simplifies its use. By adjusting stiffness (e.g., lowering it relative to the executor), the operator effort can be reduced, making the system less fatiguing and potentially more sensitive for delicate manipulations. This decoupling of twin stiffness from executor stiffness offers an additional layer of ergonomic control design.

Advantages and Potential Limitations

Advantages:

• Intuitive Full-Body Control: Provides direct manipulation of the soft robot's entire configuration space.
• Implicit Compliance Handling: Leverages the operator's intuition to exploit the robot's physical compliance during interaction tasks.
• Model-Free Potential: Reduces reliance on accurate, complex dynamic or kinematic models of the soft robot.
• Scalability: Can potentially control robots of different scales using a single twin design.

Potential Limitations:

• Hardware Requirement: Necessitates the design, fabrication, and maintenance of a dedicated, potentially complex, physical twin.
• Calibration and Matching: Performance relies heavily on accurate kinematic matching and calibration between the twin and the executor. Wear and tear or environmental changes could necessitate recalibration.
• Dynamic Mismatch: Differences in dynamics (inertia, damping, friction) between the potentially simplified twin and the real executor, especially under rapid motion or load, can lead to discrepancies in behavior. Stiffness tuning can mitigate static effort but not necessarily dynamic mismatches.
• Latency: System latency (sensing, communication, actuation) can degrade performance and stability, particularly for fast movements.
• Limited Force Feedback: While the operator feels the resistance of the twin, this may not perfectly replicate the interaction forces experienced by the executor, especially if stiffness or scale differs significantly.

Conclusion

The physical twin teleoperation method presented offers a compelling approach for controlling the complex configurations of tendon-driven soft robot arms, particularly for tasks involving environmental contact and leveraging compliance. By using a back-drivable, sensorized replica for input, it provides an intuitive interface that maps directly to the executor's actuators, bypassing traditional model-based control complexities. The demonstrated ability to facilitate tasks like squeezing through gaps, along with adaptability through scaling and stiffness tuning, underscores its potential for practical applications in domains requiring nuanced soft robot manipulation. However, practical implementation hinges on careful design of the twin, robust sensing, low-latency communication, and managing potential dynamic mismatches.