Sparse Bayesian Learning-based Adaptive Impedance Control

(Physical Human–Robot Interaction)

“Adaptive compliance through learning — robots that move in harmony with humans.”

Challenge physical effort and delay in human–robot collaboration, Change impedance dynamically through intention prediction, and create Impact in safe, responsive human–robot interaction.

Motivation

In physical human–robot interaction (pHRI), the robot must not only follow commands but also understand and adapt to the human partner’s intentions. Excessive operating force or delayed responses increase fatigue and reduce safety. To minimize human effort in both force and time, the robot should continuously adjust its impedance parameters — stiffness and damping — based on the partner’s motion intention.

Framework Overview

This research introduces a data-driven adaptive impedance control (AIC) framework that integrates two learning-based modules:

Sparse Bayesian Learning-based Human Intention Predictor (SBL-HIP) — predicts the human partner’s intended motion from time-series data of position, velocity, and force.
Variable Impedance Controller (VIC) — dynamically modulates the robot’s impedance parameters using the predicted intention, enabling compliant, coordinated behavior.

The SBL-HIP captures the human’s next action using a sparse probabilistic model, ensuring efficiency and robustness to noise. The predicted intention is then passed to the VIC, which tunes the robot’s stiffness and damping in real time. As a result, the robot can anticipate human motion and respond proactively — achieving smooth and natural collaboration.

Application: Compliant Control

Compliant control diagram (Data-driven AIC Scheme)

Robot manipulator equipped with a force sensor and compliant end-effector.
Human operator interacts through contact force \( f_h \).
Impedance parameters \( (K, B) \) adapt online based on predicted intention.
Improves cooperation smoothness, safety, and energy efficiency.

Validation

The proposed SBL-AIC scheme was validated through simulation and experiments:

2-DOF planar robot simulation — demonstrates adaptive impedance response under varying motion patterns.
6-DOF manipulator experiment — confirms compliant motion and reduced operator effort in physical collaboration.

Results demonstrate that the SBL-AIC framework allows robots to adjust their dynamic behavior in real time, achieving safe, compliant, and human-centered interaction.

This research bridges probabilistic learning and adaptive control — teaching robots to feel, anticipate, and act in synchrony with human partners.

Illustrations