Research Analyzer
← Back ICRA 2026

Squeezing the Last Drop of Accuracy: Hand-Eye Calibration Via Deep Reinforcement Learning-Guided Pose Tuning

Seunghui Shin, Daeho Kim, Hyoseok Hwang

PDF

AI summary

Key figure (auto-extracted from paper)
A deep reinforcement learning agent can automatically adjust a robot's end-effector pose to minimize estimation errors, significantly improving hand-eye calibration accuracy.
hand-eye calibration deep reinforcement learning pose estimation robotic manipulation markerless calibration SAC-Discrete

Problem

Hand-eye calibration accuracy depends heavily on the end-effector's pose, but pose estimation errors vary nonlinearly with pose and are difficult to optimize manually, especially when ground truth is unavailable.

Approach

The authors train a Soft Actor-Critic-Discrete agent in simulation to iteratively adjust the end-effector's position and orientation, using pose estimation error as a reward to guide it toward configurations that yield lower calibration errors.

Key results

  • Reduces hand-eye calibration error compared to initial poses
  • Improves accuracy for both markerless and marker-based calibration
  • Validated through simulation and real-world robotic experiments
  • Learns to navigate nonlinear pose-estimation error landscapes via DRL

Why it matters

Enables more reliable and precise robot manipulation in dynamic environments by making hand-eye calibration robust to pose-dependent estimation errors without requiring manual tuning or ground truth.

Abstract

Hand-eye calibration is a fundamental task in robotics, requiring high precision to ensure accurate manipulation. This is especially crucial for recent markerless methods, which depend on precise pose estimation for effective end-effector cali- bration. In this letter, we propose a novel approach that improves calibration performance by adjusting the end-effector’s pose to reduce prediction error. Our method utilizes a reward structure derived from trained pose estimation networks, enabling a Soft Actor-Critic-Discrete agent to learn in a simulated environment how to enhance calibration performance through action selection. Our experiments show that calibration results achieved with our method outperform those from initial poses alone in both marker- less and marker-based methods. Real-world experiments further validate the efficacy of our approach in actual robotic systems. These results demonstrate that our proposed method effectively enhances the performance of pose estimation-based hand-eye cali- bration.

Index terms

Calibration and Identification Reinforcement Learning Machine Learning for Robot Control

Related papers