ManiPose: A Comprehensive Benchmark for Pose-aware Object Manipulation in Robotics
المفاهيم الأساسية
ManiPose introduces a benchmark to advance pose-aware object manipulation research, offering simulation environments, a dataset with 6D pose labels, and a baseline method for enhanced manipulation skills.
الملخص
I. Introduction
Robotic manipulation in unstructured environments requires pose-aware object manipulation (POM).
ManiPose aims to enhance POM research with simulation environments, a dataset, and a baseline method.
II. Related Works
Benchmarks like CoppeliaSim and MuJoCo provide dynamic robot simulations.
Existing datasets lack support for generalizable 6D pose estimation methods.
III. Pose-aware Manipulation Environments
ManiPose offers tasks from single objects to cluttered scenes and articulated objects.
Three main types of tasks: single object with pose variation, multi objects in cluttered scene, articulated object interaction.
IV. Object Pose Estimation Dataset
ManiPose dataset includes real-world scanned rigid objects with standardized 6D pose labels.
Types of objects aligned based on geometry and function for accurate pose estimation.
V. Pose-aware Manipulation Baseline
Baseline method includes grasp pose prediction and action primitive planning for successful trajectories.
VI. Applications
Experiments demonstrate the effectiveness of ManiPose benchmark in simulated tasks and real-world robot applications.
ManiPose
الإحصائيات
"ManiPose encompasses Simulation environments for POM feature tasks ranging from 6D pose-specific pick-and-place of single objects to cluttered scenes."
"A comprehensive dataset featuring geometrically consistent and manipulation-oriented 6D pose labels for 2936 real-world scanned rigid objects."
"Our benchmark demonstrates notable advancements in pose estimation, setting new standards for POM research."