Generation of Synthetic Logistic Data in Isaac Sim

Institut

Lehrstuhl für Fördertechnik Materialfluss Logistik

Typ

Bachelorarbeit / Semesterarbeit / Masterarbeit /

Inhalt

 

Beschreibung

Background

Training perception models for logistic robotics applications often require large amounts of annotated data, which is time-consuming and expensive to collect manually. Synthetic data generation using high-fidelity simulation environments such as Isaac Sim provides a powerful alternative, enabling precise control, full annotations, and rapid scalability.

Objective

The goal of this thesis is to develop a system for generating synthetic data of logistic environments using Isaac Sim, with a focus on generating diverse, realistic, and annotated scenes for machine learning and perception tasks. The synthetic scenes should feature non-overlapping object placement, dynamic lighting conditions, and varied environmental properties to simulate the complexity of real-world warehouse settings.

Requirements

• Isaac Sim Environment Setup:
  • Use Isaac Sim's scatter_2d function to place logistic objects in the environment with no spatial overlap.
• Domain Randomization:
  • Vary lighting conditions to simulate different times of day and visibility scenarios.
  • Randomize object poses (position and orientation) within defined boundaries.
  • Apply randomized textures and colors to objects to promote generalization of perception models.
• Use of Logistic Distractors:
  • Introduce logistic and non-logistic distractor objects to simulate real-world occlusions and clutter.
  • Ensure distractors vary in size, type, and placement to increase scene variability.
• Data Generation and Annotation:
  • Automatically export ground truth annotations for bounding boxes, segmentation masks, depth, and RGB data.
  • Organize datasets for use in training object detection and segmentation models.
• User Interface for Parametrization:
  • Develop a user interface to allow easy parametrization of the data generation process.
  • Enable options for selecting which logistic objects and distractors to include.
  • Include checkboxes or toggles for enabling/disabling lighting variation, object color randomization, and other domain randomization settings.
  • Ensure that users can intuitively adjust and launch data generation scenarios without modifying the source code directly.

Deliverables

• A fully automated pipeline in Isaac Sim for generating diverse and annotated synthetic logistic data.
• A documented codebase with clear instructions on launching scene generation and customizing environment parameters.
• A user interface for customizing the generation parameters.
• A final report describing the system design, implementation, and evaluation.

Voraussetzungen

Requirements

• Experience with Python and Isaac Sim (Omniverse)
• Familiarity with 3D transformations, perception models, and robotics simulation
• Interest in synthetic data, machine learning, and robotic logistics
   

Möglicher Beginn

sofort

Kontakt

Daniel Vidal, M.Sc.
Raum: 5505.01.590C
Tel.: +49 (89) 289 - 15955
daniel.vidaltum.de

Ausschreibung