Automating Smart Pick-and-Place with Intrinsic Flowstate and NVIDIA Isaac Manipulator

Automating Smart Pick-and-Place with Intrinsic Flowstate and NVIDIA Isaac Manipulator

1. Introduction

Collaboration between Intrinsic.ai and NVIDIA on learning foundation skill models for industrial robotics tasks, focusing on machine-tending scenarios.

2. Workflow Overview

Utilizing NVIDIA Isaac Manipulator for grasp pose and motion generation, while employing Intrinsic Flowstate for perception tasks such as object pose estimation.

3. Isaac Manipulator for Grasp and Motion Generation

• Synthetic data generation for vacuum grasping using CAD models.
• Simulation for testing thousands of grasps and ensuring stability during transport.
• Transformation of grasp poses to the robot frame using object pose information.
• Collision-free trajectory planning with NVIDIA cuMotion for robot motion.

4. Real-World Execution with Intrinsic Flowstate

• Direct transfer of workflow from simulation to real-world workcell.
• Object pose detection using Intrinsic Flowstate's object pose estimation package.
• Computation of grasp positions and trajectory planning with Isaac Manipulator.
• Achieved cycle time of around 8s/pick in the demonstration.

5. Conclusion

Demonstrating a challenging smart pick-and-place application for metallic parts in a cluttered bin, showcasing seamless integration between simulation and real-world execution for efficient industrial automation.