[Paper Review] Touch-based object localization in cluttered environments

Huy Nguyen, Quang‐Cuong Pham|arXiv (Cornell University)|Sep 27, 2017

Robotics and Sensor-Based Localization7 references3 citations

TL;DR

This paper proposes a robust touch-based object localization method for cluttered environments by integrating RANSAC into a Bayesian estimation framework to handle outlier measurements. It introduces a fast face selection algorithm that accelerates Bayesian updating, enabling accurate and practical localization even with noisy data, achieving up to 6.5x speedup and sub-millimeter precision in experiments.

ABSTRACT

Touch-based object localization is an important component of autonomous robotic systems that are to perform dexterous tasks in real-world environments. When the objects to locate are placed within clutters, this touch-based procedure tends to generate outlier measurements which, in turn, can lead to a significant loss in localization precision. Our first contribution is to address this problem by applying the RANdom SAmple Consensus (RANSAC) method to a Bayesian estimation framework. As RANSAC requires repeatedly applying the (computationally intensive) Bayesian updating step, it is crucial to improve that step in order to achieve practical running times. Our second contribution is therefore a fast method to find the most probable object face that corresponds to a given touch measurement, which yields a significant acceleration of the Bayesian updating step. Experiments show that our overall algorithm provides accurate localization in practical times, even when the measurements are corrupted by outliers.

Motivation & Objective

• To address the challenge of inaccurate localization due to outlier touch measurements in cluttered environments.
• To improve the robustness of Bayesian touch-based localization against outliers from unintended contacts.
• To accelerate the computationally intensive Bayesian updating step in iterative estimation frameworks.
• To enable practical running times for real-world robotic manipulation tasks requiring sub-millimeter precision.
• To develop a method that does not require prior knowledge of surrounding objects, ensuring applicability in unknown clutter.

Proposed method

• Integrates RANSAC into a Bayesian estimation framework to iteratively hypothesize and verify sets of inlier measurements, rejecting outliers.
• Proposes a fast face selection algorithm that identifies the most probable object face corresponding to a touch measurement, reducing likelihood evaluation cost.
• Uses a threshold-based face selection method with angular deviation ($\delta_\alpha = 0.09$ rad) to filter candidate faces before Bayesian update.
• Employs the Scaling Series method for 6-DOF pose estimation with particle-based Bayesian inference.
• Applies forward kinematics and tactile sensing (position and normal) from end-effector contacts to generate measurements.
• Uses ray-tracing simulation to generate realistic touch measurements, including outliers from nearby objects.

Experimental results

Research questions

• RQ1 Can RANSAC be effectively adapted to a Bayesian framework to improve robustness against outlier touch measurements?
• RQ2 How can the computational cost of Bayesian updating be reduced in touch-based localization?
• RQ3 To what extent does the proposed face selection method improve runtime without sacrificing localization accuracy?
• RQ4 Can the method achieve high-precision localization in cluttered environments where unintended contacts are common?
• RQ5 How does the method perform under varying initial uncertainty and object complexity?

Key findings

• The proposed method achieved an average localization error of 0.81 mm in translation and 0.015 rad in rotation, significantly outperforming the baseline without outlier classification (6.64 mm, 0.1 rad).
• The face selection algorithm reduced average execution time by up to 6.5 times for complex objects (e.g., cash register model) compared to the baseline.
• The method maintained 100% success rate across 50 trials in cluttered scenes with 1–5 outliers among 15 measurements.
• The speedup was most pronounced for complex objects with more faces, demonstrating the scalability of the face selection approach.
• The method remained effective even with large initial uncertainty (10 mm in position, 0.3 rad in orientation), showing robustness to initialization.
• Experiments confirmed that outlier measurements alone severely degrade performance, validating the necessity of outlier classification.