Workflow for using learning based approach for placing boxes on pallets

Patent metadata

Field	Value
Publication number	US-12485544-B2
Application number	US-202217838045-A
Country	US
Kind code	B2
Filing date	Jun 10, 2022
Priority date	Jun 16, 2021
Publication date	Dec 2, 2025
Grant date	Dec 2, 2025

Abstract

A robotic system is disclosed. The system includes a memory that stores a machine learning-based model to provide a scoring function value for a candidate item placement on a pallet on which are plurality of items are to be stacked given a current state value of the pallet and a set of zero or more items placed previously. The system includes one or more processors that use the model to determine a corresponding score for each of a plurality of candidate placements for a next item to be placed and the current state value associated with the current state of the pallet and a set of zero or more items placed previously, select a selected placement based at least in part on the respective scores, control a robotic arm to place the next item according to the selected placement.

First claim

The invention claimed is:

1 . A robotic system, comprising:
a memory configured to store a machine learning-based model to provide a scoring function value for a candidate item placement on a pallet on which are plurality of items are to be stacked given a current state value of the pallet and a set of zero or more items placed previously; and one or more processors coupled to the memory and configured to:
use the model to determine a corresponding score for each of a plurality of candidate placements for a next item to be placed and the current state value associated with the current state of the pallet and a set of zero or more items placed previously,
wherein:
a search space of possible scenarios for placing the candidate item and zero or more of the plurality of items is pruned based at least in part on a determination that a score for a particular candidate placement of the plurality of placement is less than a predefined scoring threshold; and
the pruning of the search for the particular candidate placement includes removing from the particular candidate placement as a potential placement, and a set of placements that depend on the pruned candidate placement;

select a selected placement based at least in part on the respective scores; and
control a robotic arm to place the next item according to the selected placement.

2 . The robotic system of claim 1 , wherein an item placement comprises a placement of the item in a particular location.

3 . The robotic system of claim 1 , wherein the one or more processors are further configured to train the machine learning-based model.

4 . The robotic system of claim 1 , wherein the machine learning-based model is trained by observing a set of physical palletization processes.

5 . The robotic system of claim 1 , wherein the machine learning-based model is trained based on a set of predefined heuristics.

6 . The robotic system of claim 5 , wherein the set of predefined heuristics comprise one or more of (i) a bias or preference to place an item at an edge, (ii) a bias or preference to place an item up against another item among a stack of items, (iii) a bias or preference to place a large or heavy item at or near a bottom of the stack of items, (iv) a bias or preference to place a small or light item at or near a top of the stack of items, (v) a bias or preference to place an irregularly shaped item at or near the top of the stack of items, (vi) a bias or preference to place a deformable item at or near the top of the stack of items.

7 . The robotic system of claim 1 , wherein the plurality of candidate placements is limited to a predefined number of placements.

8 . The robotic system of claim 1 , wherein the plurality of candidate placements is limited to a set of placements that satisfy a criteria for possible placements.

9 . The robotic system of claim 8 , wherein the criteria for possible placements is a predefined scoring threshold according to a predefined scoring function.

10 . The robotic system of claim 1 , wherein the selecting the selected placement is based at least in part on performing a tree search of various scenarios.

11 . The robotic system of claim 10 , wherein the various scenarios respectively correspond to different combinations of (i) sequences of item placements, (ii) locations at which respective items are placed, and (iii) orientations in which the respective items are placed.

12 . The robotic system of claim 1 , wherein the pruning the search space comprises pruning branches or nodes for scenarios that have respective scores for a scoring function that is less than a predefined scoring threshold.

13 . The robotic system of claim 12 , wherein the pruning the branches or nodes for scenarios comprises:
use the scoring function to determine a score for a scenario corresponding to a first node; determining whether the score corresponding to the first node is less than the predefined scoring threshold; and in response to determining that the score corresponding to the first node is less than the predefined scoring threshold,
pruning the first node from the search space.

14 . The robotic system of claim 13 , wherein the pruning the branches or nodes for the scenarios further comprises:
in response to determining that the score corresponding to the first node is less than the predefined scoring threshold,
determining whether the search space has one or more downstream nodes that branch directly or indirectly from the first node; and
in response to determining that the search space has one or more downstream nodes that branch directly or indirectly from the first node,
pruning the one or more downstream nodes.

15 . The robotic system of claim 14 , wherein the one or more downstream nodes are pruned without computing respective scores according to the predefined scoring function for the scenarios corresponding to the one or more downstream nodes.

16 . The robotic system of claim 1 , wherein the pruning the search space comprises pruning branches or nodes for scenarios that have respective costs for a cost function that is more than a predefined cost threshold.

17 . The robotic system of claim 1 , wherein the current state reflects one or more of (i) a stability of the pallet or stack of items on the pallet, (ii) a packing density of items placed on the pallet, (iii) an efficient use of resources, (iv) a collision avoidance, and (vi) avoidance of awkward positioning of the robotic arm.

18 . The robotic system of claim 1 , wherein the respective scores reflect how the corresponding candidate placement would contribute to a current or future value of the state of the pallet or a stack of items placed on the pallet.

19 . A method to control a robot, comprising:
storing a machine learning-based model to provide a scoring function value for a candidate item placement on a pallet on which are plurality of items are to be stacked given a current state value of the pallet and a set of zero or more items placed previously; using the model to determine a corresponding score for each of a plurality of candidate placements for a next item to be placed and the current state value associated with the current state of the pallet and a set of zero or more items placed previously, wherein:
a search space of possible scenarios for placing the candidate item and zero or more of the plurality of items is pruned based at least in part on a determination that a score for a particular candidate placement of the plurality of placement is less than a predefined scoring threshold; and
the pruning of the search for the particular candidate placement includes removing from the particular candidate placement as a potential placement, and a set of placements that depend on the pruned candidate placement;
selecting a selected placement based at least in part on the respective scores; and controlling a robotic arm to place the next item according to the selected placement.

20 . A computer program product to control a robot, the computer program product being embodied in a non-transitory computer readable medium and comprising computer instructions for:
storing a machine learning-based model to provide a scoring function value for a candidate item placement on a pallet on which are plurality of items are to be stacked given a current state value of the pallet and a set of zero or more items placed previously; using the model to determine a corresponding score for each of a plurality of candidate placements for a next item to

Assignees

• Dexterity Inc

Inventors

• Pidaparthi Rohit Arka
• Clary William Arthur
• Abhyankar Neeraja
• Kuck Jonathan
• Pottayil Ben Varkey Benjamin
• Chavez Kevin Jose
• Kumar Shitij

Classifications

• B25J9/163

  learning, adaptive, model based, rule based expert control · CPC title

• G05B2219/39001

  Robot, manipulator control · CPC title

• G05B2219/42152

  Learn, self, auto tuning, calibrating, environment adaptation, repetition · CPC title

• B65G61/00

  Use of pick-up or transfer devices or of manipulators for stacking or de-stacking articles not otherwise provided for · CPC title

• G05B2219/40067

  Stack irregular packages · CPC title