We developed AutoGrasp™, an AI grasping technology based on techniques in deep learning and reinforcement learning. Our proprietary AI model continuously learns and improves with our robots - generating a variety of skills that enable operation in the real world, such as the fast-evolving landscape of e-commerce retail.


4 Core Elements to AutoGrasp™


1. Vision

Our robot is able to see and distinguish individual items in unstructured environments typical of commercial operations. Once items are identified, AutoGrasp™ will identify which item has the highest confidence score for a successful pick. 


2. Motion

Once an item is identified, AutoGrasp™ creates a motion plan to map out the best path for the robotic arm to approach the item, including obstacle avoidance.


3. Action

Once the robotic arm has successfully navigated to the desired item, AutoGrasp™ will calculate the best gripper position needed, select the appropriate force required to grasp the item, and verify successful action completion. 


4. Placement

With the single item picked,  AutoGrasp™ moves the robotic arm to the correct bin location based on information received from the WMS provider.


Our Data Model


Robot Pilot Center

If AutoGrasp™ reaches a minimum confidence threshold when attempting a pick, our human robot pilots can remote in and assist. The benefits of our human-in-the-loop approach are twofold: 1. Ensures we deliver a consistent and reliable service to our customers 2. Each human-assisted pick provides valuable data that is fed back into our algorithms.


Production Data

Operating in commercial production environments provides a continuous collection of data from our robots comprehensive sensor suite.



Read the latest in Kindred's AI research.