Abstract:Dense packing in pick-and-place systems is an important feature in many warehouse and logistics applications. Prior work in this space has largely focused on planning algorithms in simulation, but real-world packing performance is often bottlenecked by the difficulty of perceiving 3D object geometry in highly occluded, partially observed scenes. In this work, we present a fully-convolutional shape completion model, F-CON, which can be easily combined with off-the-shelf planning methods for dense packing in the real world. We also release a simulated dataset, COB-3D-v2, that can be used to train shape completion models for real-word robotics applications, and use it to demonstrate that F-CON outperforms other state-of-the-art shape completion methods. Finally, we equip a real-world pick-and-place system with F-CON, and demonstrate dense packing of complex, unseen objects in cluttered scenes. Across multiple planning methods, F-CON enables substantially better dense packing than other shape completion methods.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the dense packing problem in warehouse automation and logistics applications, especially how to achieve high - density item placement in partially - observed complex and novel object scenarios. Specifically, the paper focuses on how to overcome the challenges of perceiving 3D object geometries in the real world. These challenges include partial occlusion of objects, materials being difficult to be perceived by depth cameras, etc., which limit the performance of existing dense packing systems. To address these challenges, the authors propose a fully - convolutional shape completion model (F - CON), aiming to improve the dense packing ability of robotic systems when dealing with unseen complex objects. The main contributions of the paper include: 1. **Released a simulated dataset** (COB - 3D - v2) for training shape completion models. This dataset has a high degree of visual realism and is helpful for the transfer from the simulated environment to the real world. 2. **Proposed a 3D fully - convolutional model architecture** (F - CON) specifically for shape completion tasks and performs excellently in the field of robotics. 3. **Through a large number of practical experiments**, it is shown that when F - CON is combined with existing simple planning methods, it can achieve state - of - the - art dense packing performance when dealing with complex and unseen objects. The paper also describes in detail how F - CON works, the construction process of the dataset, and how to apply F - CON in actual dense packing tasks. Through these studies, the authors not only solve the limitations of existing technologies in real - world applications but also provide new solutions for future robotic dense packing systems.

Convolutional Occupancy Models for Dense Packing of Complex, Novel Objects

MoreFusion: Multi-object Reasoning for 6D Pose Estimation from Volumetric Fusion

One Model Packs Thousands of Items with Recurrent Conditional Query Learning

Towards Flexible 3D Perception: Object-Centric Occupancy Completion Augments 3D Object Detection

Stable bin packing of non-convex 3D objects with a robot manipulator

Tight Robot Packing in the Real World: A Complete Manipulation Pipeline with Robust Primitives

Task-driven Perception and Manipulation for Constrained Placement of Unknown Objects

A Method of Tight Placement for Robotic Dense Packing

OccLoff: Learning Optimized Feature Fusion for 3D Occupancy Prediction

Occ3D: A Large-Scale 3D Occupancy Prediction Benchmark for Autonomous Driving

LiDAR-based 4D Occupancy Completion and Forecasting

Shape Completion Enabled Robotic Grasping

Convex-hull Feature Adaptation for Oriented and Densely Packed Object Detection

Scene-level Pose Estimation for Multiple Instances of Densely Packed Objects

Robot Online 3D Bin Packing Strategy Based on Deep Reinforcement Learning and 3D Vision

6D object pose estimation based on dense convolutional object center voting with improved accuracy and efficiency

Learning the Sequence of Packing Irregular Objects from Human Demonstrations: Towards Autonomous Packing Robots

CabiNet: Scaling Neural Collision Detection for Object Rearrangement with Procedural Scene Generation

Category-level Shape Estimation for Densely Cluttered Objects

RoboPack: Learning Tactile-Informed Dynamics Models for Dense Packing