Stow: Robotic Packing of Items into Fabric Pods

TL;DR

The paper tackles dense, heterogeneous item packing into fabric pods in production warehouses where elastic bands occlude visibility. It presents an end-to-end robotic system with specialized hardware (band separator, extendable EoAT plank, dual-conveyor jaws), perception that sees through translucent bands via learned bin representations, and a risk-aware match planner to optimize stow rate and storage density. Deployment results show over 500,000 stows with high success and competitiveness to human stow rates, while maintaining safety through overhead-shelf operation. The work demonstrates a practical, scalable solution and highlights future directions in reducing defects and advancing learned planning and perception for even higher reliability.

Abstract

This paper presents a compliant manipulation system capable of placing items onto densely packed shelves. The wide diversity of items and strict business requirements for high producing rates and low defect generation have prohibited warehouse robotics from performing this task. Our innovations in hardware, perception, decision-making, motion planning, and control have enabled this system to perform over 500,000 stows in a large e-commerce fulfillment center. The system achieves human levels of packing density and speed while prioritizing work on overhead shelves to enhance the safety of humans working alongside the robots.

Figures (10)

• Figure 1: Storage Pod: Items are stored densely and heterogeneously within each bin on this pod face. Elastic bands are used to prevent objects from falling out as pods are moved throughout the building by mobile robots. Bins are continuously picked from and stowed into during operations. Items arrive at stow stations in plastic totes, bottom right.
• Figure 2: In the Robotic Stow Process, a human checks inbound items for quality and feeds three robotic workcells. The workcells store items in buffer, then match and store items to inbound pods using a set of manipulation behaviors.
• Figure 3: Top: Plan view of system with A) induct station, B) transfer conveyance, C) Item Buffer, D) Reorientation, E) Band Manipulator, F) Stow Manipulator. Bottom: Robot station where items are vended by C) Item Buffer then reoriented by D) SCARA robot before being grasped and stowed by F) robot arm on gross positioning gantry. E) linear gantry robot opens elastic bands.
• Figure 4: Orientation EoAT and Hand-off: A SCARA robot with a foam disk is used to orientate items before they are conveyed into the robot EoAT.
• Figure 5: a) The robot EoAT uses parallel jaws with built-in conveyors. With this strategy, items are fed into the jaws by simple conveyor transfer and inserted into bins without requiring arm motions A retractable aluminum plank is used to manipulate items already in the bin to create available space to stow. b) The band EOAT uses a compound hook to grasp and pull the elastic bands out of the way for stowing. With a cantilevered profile, collision volumes are minimized, and with a six-axis force-torque sensor, contact can be made safely with the pod bin.