Research Analyzer
← Back ICRA 2026

Learning Controlled Separation of Small Objects between Two Fingers with a Tactile Skin

Ulf Kasolowsky, Berthold Bäuml

PDF

AI summary

Key figure (auto-extracted from paper)
Robotic hands can learn to drop small objects to a precise count using only tactile feedback and reinforcement learning, eliminating the need for vision.
tactile sensing reinforcement learning robotic grasping sim-to-real transfer fine manipulation blind bin picking

Problem

Blindly grasping many small objects and selectively dropping them to a target count is a complex fine manipulation skill that humans perform intuitively but robots struggle to execute without spatial touch feedback.

Approach

A reinforcement learning policy is trained in simulation to control two robotic fingers, using spatially-resolved tactile skin data and joint angles to learn a sparse reward for reaching a desired object count.

Key results

  • Successfully learned to separate 6 mm pellets to target counts of 1–3 in simulation
  • Spatially-resolved tactile feedback improved task success by up to 20% over joint-only control
  • An auxiliary estimator accurately predicted pellet contact positions from low-resolution tactile data
  • Achieved robust sim-to-real transfer on the DLR-Hand II with a physical tactile skin

Why it matters

Enables vision-free fine manipulation for multi-fingered robotic hands, expanding their capability for precise industrial and medical bin-picking tasks.

Abstract

We introduce and solve the novel task of controlled separation of small objects with two fingers of a multi-purpose robotic hand: after grasping into a box of small objects, the task is to drop as many of them until a desired number remains between the fingers. The objects are small compared to the width of the fingers but also in absolute terms. In our case little pellets with a diameter of only 6 mm are handled. We show that the task can be performed purely tactile (no vision) using a spatially-resolved tactile skin on a fingertip. The separation policy is trained in simulation via reinforcement learning using a straightforward sparse reward, which basically checks if the desired number of objects is reached. In simulation experiments, we provide an exhaustive analysis of the benefits of using spatially-resolved tactile feedback: while an ideal (high-resolution) tactile sensor allows solving the task almost perfectly, a sensor with lower spatial resolution (here 4 × 4 taxels) still leads to an improvement of up to 20 % compared to using only the fingers’ joint sensors. For this analysis, we further train an estimator alongside the policy that predicts the ground truth contact positions. Finally, we demonstrate the successful sim-to-real transfer for the DLR-Hand II equipped with a tactile skin. Website: aidx-lab.org/skin/icra26

Index terms

In-Hand Manipulation Force and Tactile Sensing Multifingered Hands

Related papers