Deep Learning–Powered Automotive Parts Bin Picking with UR5e

#BinPicking#UR5e#3DVision#DeepLearning#InstanceSegmentation#ReinforcementLearning#6DPoseEstimation#RoboticPicking

Process Overview

This solution automates product sorting, assembly, and packing by utilizing 3D vision, deep learning algorithms (instance segmentation), reinforcement learning, and motion planning (OMPL) algorithms.

Since it performs tasks by recognizing the real-time product position and type as well as the box location using 3D vision, it can automatically compensate and operate even when the product position or box position changes.

Even if vision/robot/work jigs become worn or conditions change, the system is corrected through learning, making operation and maintenance easier and minimizing downtime.

Components

Robot	UR5e 6-axis articulated robot Payload: 5 kg Working radius: 850 mm Weight: 20.6 kg
Peripherals	Robot base (custom-made) Pneumatic gripper (custom-made) 3D vision camera (with proprietary algorithm embedded) Vision stand Workpiece alignment jig

Workflow

STEP 1.	Workpiece input
STEP 2.	Optimized picking plan
STEP 3.	Picking and alignment
STEP 4.	Aligned and loaded into the discharge box

※ The content provided by Marosol is protected under U.S. and Canadian copyright and intellectual property laws. Unauthorized reproduction, distribution, or use of this content is strictly prohibited.

Key Features

Smooth robot teaching

Vision calibration and teaching time reduced through deep learning
No need for recalibration or re-teaching even when work conditions change
Excellent recognition capability for various objects based on deep learning–based 3D object recognition
Learns optimal pick positions according to product shape
Reduced training time by leveraging simulation data
Performs picking by sweeping or moving the product when it is not suitable for gripping
Fast product pose estimation within 0.2 seconds (6D pose estimation)

Compact and simple installation

No need to fix the input box in place, eliminating the need for a separate custom feeder
Enables automation by adding a robot to the operator’s existing workspace

Fast ROI – optimized operation and reduced maintenance cost

Shorter cycle time through continuous learning and optimization of picking pose and path
Automatic compensation and task execution even if the layout shifts or the robot becomes worn over time
Stable production by minimizing downtime

Implementation Results

ROI Period
Key Metrics
Client Feedback