Model-less Active Compliance for Continuum Robots using Recurrent Neural Networks

TL;DR

The paper addresses safe, active compliance for continuum robots without a detailed mechanical model by training an RNN to predict internal tendon forces from historical actuator signals, enabling a feed-forward controller in actuator space. The relation $\mathbf{F}_{ext}=\mathbf{F}_{meas}-\mathbf{F}_{int}$ is used to drive compliant behavior, with $\mathbf{F}_{int}=rnn(\mathbf{Q})$, and a deadband $\lambda$ ensures stability. On a 3-tendon single-segment platform, the approach outperforms non-recurrent baselines (e.g., CNN) in tension prediction, enables rapid response to external perturbations (within ~$0.2$ s), and supports compliant insertion in unknown environments (e.g., curved tubes). The results suggest a viable, data-driven, model-less path to safe manipulation with continuum robots and point toward scalable adoption for more complex multi-segment systems.

Abstract

Endowing continuum robots with compliance while it is interacting with the internal environment of the human body is essential to prevent damage to the robot and the surrounding tissues. Compared with passive compliance, active compliance has the advantages in terms of increasing the force transmission ability and improving safety with monitored force output. Previous studies have demonstrated that active compliance can be achieved based on a complex model of the mechanics combined with a traditional machine learning technique such as a support vector machine. This paper proposes a recurrent neural network based approach that avoids the complexity of modeling while capturing nonlinear factors such as hysteresis, friction and delay of the electronics that are not easy to model. The approach is tested on a 3-tendon single-segment continuum robot with force sensors on each cable. Experiments are conducted to demonstrate that the continuum robot with an RNN based feed-forward controller is capable of responding to external forces quickly and entering an unknown environment compliantly.

Figures (13)

• Figure 1: A 3-tendon single-segment continuum robot with force sensors on each cable. Active compliance allows the robot to move compliantly by adjusting the actuating cables when experiencing external forces along the segment.
• Figure 2: Cable tension throughout linear motion. The first, second, and third waveform corresponds to cable motion speed at 1mm/s, 2.5mm/s, and 10mm/s, respectively.
• Figure 3: An RNN based tension predictor is supplied with $n=200$ discrete time points of control signal for each actuator, and predicts the tension in each cable at the next timestep $t+1$. Recorded tension training data (black) is shown with prediction tension (red).
• Figure 4: Diagram of the RNN-based compliant motion controller in the actuator space.
• Figure 5: Experimental setup for tip-cable force calibration. Known weights (coins) were applied to the tip at various orientations and positions.