Online Learning for Vibration Suppression in Physical Robot Interaction using Power Tools

2025-08-05

AI Summary

To address the challenge of external vibration interference degrading operational precision and stability of collaborative robots performing power-tool tasks (e.g., grinding) in complex construction environments, this paper proposes a feedforward force control method integrating online learning and active vibration suppression. The core contribution is an enhanced Damping-based Multiple Fourier Linear Combiner (BMFLC) algorithm, featuring a logic-function-driven adaptive step-size mechanism that simultaneously ensures rapid convergence and significantly improves noise robustness. Comparative simulations demonstrate superior vibration suppression performance and lower computational overhead relative to conventional BMFLC and its Recursive Least Squares (RLS)/Kalman Filter (KF) variants. Real-world grinding experiments further validate the method’s effectiveness and engineering practicality in dynamic physical interaction tasks.

Technology Category

• Intelligent Robots: Behavior Learning & Control
• Machine Learning: Active Learning
• Planning, Routing, and Scheduling: Learning for Planning and Scheduling

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Vibration suppression is an important capability for collaborative robots deployed in challenging environments such as construction sites. We study the active suppression of vibration caused by external sources such as power tools. We adopt the band-limited multiple Fourier linear combiner (BMFLC) algorithm to learn the vibration online and counter it by feedforward force control. We propose the damped BMFLC method, extending BMFLC with a novel adaptive step-size approach that improves the convergence time and noise resistance. Our logistic function-based damping mechanism reduces the effect of noise and enables larger learning rates. We evaluate our method on extensive simulation experiments with realistic time-varying multi-frequency vibration and real-world physical interaction experiments. The simulation experiments show that our method improves the suppression rate in comparison to the original BMFLC and its recursive least squares and Kalman filter-based extensions. Furthermore, our method is far more efficient than the latter two. We further validate the effectiveness of our method in real-world polishing experiments. A supplementary video is available at https://youtu.be/ms6m-6JyVAI.

Problem

• Suppress vibration in robots using power tools
• Improve convergence and noise resistance online
• Validate method in simulations and real-world experiments

Innovation

• Uses BMFLC algorithm for online vibration learning
• Proposes damped BMFLC with adaptive step-size
• Implements logistic damping to reduce noise impact