This project introduces DYNAMO-GRASP, a novel approach to the challenge of suction grasp point detection. By leveraging the power of physics-based simulation and data-driven modeling, DYNAMO-GRASP is capable of accounting for object dynamics during the grasping process. This significantly enhances a robot's ability to handle unseen objects in real-world scenarios. Our method was benchmarked against established approaches through comprehensive evaluations in both simulated and real-world environments. It outperformed the alternatives by achieving a success rate of 98% in diverse simulated picking tasks and 80% in real-world, adversarially-designed scenarios. DYNAMO-GRASP demonstrates a strong ability to adapt to complex and unexpected object dynamics, offering robust generalization to real-world challenges. The results of this research pave the way for more reliable and resilient robotic manipulation in intricate real-world situations.
This repository provides the codebase for collecting data through simulation. These scripts will help users to generate their own datasets, further enhancing the extensibility and usefulness of DYNAMO-GRASP.
For more information please refer to our project website
Follow these steps for the installation:
# Import the conda environment with Python version 3.8
conda env create -f environment.yamlFor more details see Installation
pip install -e .
# Create a folder inside the repo for saving the data(grasp point properties, depth image, segmentation mask image),
mkdir scenario_grasp_configurationsTo run the project, use the dynamo_grasp.sh bash script:
conda run -n dynamo-grasp ./data-collection.py --num-envs 50