Collapse and collision aware grasping for cluttered shelf picking.

Area of Science:

• Robotics
• Artificial Intelligence
• Physics Simulation

Background:

• Automated shelf picking in smart factories requires high throughput, flexibility, and safety.
• Retrieving stacked objects presents challenges due to spatial and structural interdependencies.
• Traditional vision methods lack physical reasoning, causing collisions and collapses.

Purpose of the Study:

• To develop a collapse and collision-aware grasp planner for robotic shelf picking.
• To integrate dynamic physics simulations into robotic decision-making for retrieval tasks.
• To improve the efficiency and safety of automated object retrieval in smart factories.

Main Methods:

• Reconstructed approximate 3D scene representations from single images and depth maps.
• Integrated dynamic physics simulations for evaluating retrieval strategies.
• Proposed heuristic-based and physics-based approaches for single-box extraction and shelf clearance.

Main Results:

• Physics-aware method significantly improved efficiency and success rates compared to baseline heuristics.
• Demonstrated effectiveness on structured and unstructured box stacks through real-world experiments.
• Validated performance using existing dataset benchmarks.

Conclusions:

• Physics-integrated robotic decision-making enhances automated shelf picking safety and efficiency.
• The proposed grasp planner effectively addresses challenges in stacked object retrieval.
• This approach offers a significant advancement for human-robot collaboration in smart factories.