Integration and Continual Learning-Based Modeling of a Soft Robotic Sensor for Social Robot Proprioception
Pak Chuen Hau, Seshagopalan Thorapalli Muralidharan, Randy Gomez, Georgios Andrikopoulos
AI summary
Problem
Soft robotic actuators lack reliable, long-term proprioceptive feedback due to sensor drift and the difficulty of embedding flexible sensing elements without compromising compliance or external form.
Approach
The authors fabricate a multi-material 3D-printed strain sensor array integrated directly into a soft continuum actuator and map its resistance signals to joint angles using linear regression, a static neural network, and an online continual-learning framework that updates parameters in real-time.
Key results
- Gauge-pattern geometry selected for optimal linearity and repeatability
- Continual-learning model achieves R² > 0.97 and MAE < 1° across motion profiles
- Online updates consistently outperform static linear and neural network baselines
- Validates embedded sensing and adaptive modeling under predefined and randomized trajectories
Why it matters
Provides a practical pathway for integrating robust, self-calibrating proprioception into soft social robots, enhancing safety, expressivity, and long-term reliability in human-robot interaction.
Abstract
This paper presents an embedded soft sensor for proprioceptive feedback in a soft continuum actuator (SCA) forming the neck of the social robot HARU. The sensor is fabricated in a single-step multi-material additive manufac- turing process, co-extruding conductive and non-conductive thermoplastic polyurethane to form an integrated structure. Several sensor geometries are evaluated, with a gauge-type con- figuration selected based on linearity and repeatability criteria. The design is embedded in a cross-configuration to measure the actuator’s two dominant degrees of freedom, pitch and roll. Sensor signals are mapped to angle estimates using linear regression, a static neural network, and a continual-learning framework that updates parameters online. Experiments involv- ing predefined trajectories, randomized motions, and repeated test cycles show that the continual-learning model achieves R2 > 0.97 and mean absolute errors below 1◦, consistently improving upon the baseline models. The results demonstrate the feasibility of directly embedding 3D-printed soft sensors into functional actuators and highlight the role of adaptive learning in supporting long-term soft robotic proprioception.