Smart Object Recognition & Adaptive Grasping with Mixture-of-Gaussians 🎯
RoboGrasp-MoG is an advanced object recognition system designed to empower robots with adaptive grasping capabilities. Leveraging the power of Mixture-of-Gaussians (MoG) models, this project enables robust classification of diverse objects using RGB-D images and 3D point clouds — helping robots decide how to grasp each item effectively. 🖼️📊
- 🎯 Mixture-of-Gaussians Clustering for precise object classification
- 🤝 Seamless integration with robotic grasping strategies
- 📷 Utilizes rich RGB-D imagery and 3D point cloud data
- 🔄 Model training powered by the Expectation-Maximization (EM) algorithm
- ⚙️ Designed for adaptability across varied objects and environments
Robotic grasping is tricky — different objects need different handling. Our approach uses probabilistic modeling to learn object categories dynamically, improving grasp success rates and efficiency in real-world tasks. 🦾✨
git clone https://github.com/ItsShriks/ML_Project.git
cd RoboGrasp-MoG
pip install -r requirements.txt
# Follow further setup instructions in docs/setup.md- Prepare RGB-D images and point cloud data from your sensors 📸
- Train the MoG model using the included EM algorithm implementation 🧠
- Run the recognition system to classify objects in real-time ⚡
- Integrate with your robot’s grasp controller to adapt grip accordingly 🤖✋
- /data — Sample datasets (RGB-D and point clouds)
- /src — Core implementation of MoG clustering and EM training
- /docs — Detailed documentation and setup guides
This project was completed as part of a university course at Hochschule Bonn-Rhein-Sieg under the guidance of Prof. Dr. Sebastian Houben