Intelligent Grasping for Warehouse Cobotics

Autonomous Mobile Robots & Vehicles

A smart robot that uses regular or depth images to spot and grasp objects on a table with precision-guiding each pick with accuracy for smooth, hassle-free handling.

Vector (2)-da6325

Institute:
Prof. GC Nandi

Vector (3)-42d8ac
PI Name:
IIIT Allahabad

Technology Readiness Level (TRL)
4

Intellectual Property:
PA No. 202311009245

Problem
Addressed

The project tackles three key challenges: intelligent robotic grasping using model-free reinforcement learning. autonomous coordination of multiple mobile robots, and seamless human-robot interaction. These advancements aim to enhance cobot efficiency in dynamic environments like warehouses and healthcare.

Arrow 6

About the
Technology

  • It takes an RGBD image in one model and RGB image in another model of an object as input that is centrally placed 8 on the tabletop with a white background and captures top view of the object.
  • With our tool, in both the models a grasp rectangle is generated as an output, with parameters (x, y, and Ѳ), where (x, y) is the center of the rectangle and Ѳ is the orientation of the grasping rectangle. These parameters can be further utilized as an input to any robot system for executing robotic grasping through existing inverse kinematics and trajectory planning methods.
  • Both CNN and deep Generative models have been used for model development.
  • Both RGBD & RGB images have been used as inputs for grasp generations.
  • Works for seen and unseen objects

Application Areas & Use Cases

  • Input: RGBD & RGB images
  • Output: Grasp rectangle with parameters (x, y and Ѳ)
  • Models Used for our design: VQ-VAE & Pix2Pix GAN
  • Approach: Generative Based
Arrow 6
Warehouse
Assitive Robotics