Research Analyzer
← Back ICRA 2026

Replicating Painting Strokes: Shape Aware Dynamic Motion Primitives for Robotic Manipulation

Jelena Vuleti ́c, Pero Drobac, Bruno Mari ́c and Matko Orsag

PDF

AI summary

Key figure (auto-extracted from paper)
A shape-aware DMP framework with real-time magnetic sensing enables robots to accurately replicate human painting strokes while precisely controlling deformable tool curvature.
Shape-aware DMPs magnetic sensing soft robotics programming by demonstration tool deformation robotic painting

Problem

Replicating delicate tasks like painting requires simultaneous control of motion and tool deformation, which is difficult with simple deformable tools and existing sensing methods.

Approach

The authors combine shape-aware Dynamic Motion Primitives with a custom magnetic gradiometer to estimate tool curvature at 1 kHz, feeding this into a linear PD/P controller that maintains shape while tracking demonstrated trajectories.

Key results

  • Shape-aware DMPs enable simultaneous motion planning and real-time curvature servoing at 1 kHz
  • Custom magnetic gradiometer accurately estimates scraper bending angle without external calibration
  • Synthesized linear control algorithm maintains tool shape while replicating human-demonstrated trajectories
  • System successfully reproduces complex paintings under altered experimental conditions

Why it matters

Provides a cost-effective, high-speed solution for soft robotic manipulation in delicate industrial and artistic tasks requiring precise tool deformation control.

Abstract

This work proposes derivation and experimental validation of the robotic manipulation framework based on the shape aware Dynamic Motion Primitives (DMPs) that enables precise shape servoing relying on magnetic based sensing. The magnetic sensing system, with the sensor sampling rate of 1 kHz, outputs measurements that can be used as an estimate of the bending angle of the deformable plastering scraper, approx- imated as a constant curvature segment. Synthesized linear control algorithm encompasses PD controller that accounts for deviations in the scraper bending angle and P controller that enables maintaining end effector pitch angle reference. The proposed control system is combined with the Programming by Demonstration (PbD) approach. Consequently, the presented robotic painting system can follow the demonstrated scraper bending angle and end effector pitch angle reference values while replicating the demonstrated trajectory, even in signifi- cantly altered experimental conditions.

Index terms

Modeling Control and Learning for Soft Robots Art and Entertainment Robotics Soft Robot Applications

Related papers