New Approaches in Robotic Grasping

This presentation introduces recent advancements in robotic grasping, focusing on approaches grounded in deep learning, reinforcement learning, and graph-based techniques. It covers state-of-the-art technologies for grasp pose detection and validates these methods through the development of a 3-DOF Delta robot and various robotic grippers designed in the Human-Robot Interaction Laboratory at the University of Tehran.

Additionally, the presentation explores learning-based and environment-search methods for scene understanding and task planning, including applications such as scene rearrangement, object retrieval, and sorting. Prominent AI models used in robotic applications, including OpenVLA, RT-2, and pi-0, will also be reviewed.

Presentation Outline:

- Introduction to Robotic Grasping
- Geometric Methods and Their Integration with Intelligent Techniques for Object Grasping
- Reinforcement Learning Approaches for Data Insufficient Scenarios
- Graph-Based Analysis and Modeling for Specific Problems
- Overview of Common Robotic Simulators
- Practical System Description and Implementation Results Using the Developed Delta Robot and Two-Finger Gripper
- Review of the Latest Research in Robotic Task Execution

Speakers:

Dr. Mehdi Tale Masouleh – Associate Professor of Artificial Intelligence and Robotics, University of Tehran
Dr. Hamed Hosseini – Ph.D. in Artificial Intelligence and Robotics, University of Tehran
Hamed Ghasemi – Ph.D. Candidate in Artificial Intelligence and Robotics, University of Tehran