Tidying System Using a Tidying Robot

Robot Tidying System Combining Image Segmentation and Deep Reinforcement Learning

This technology is a tidying system that controls an articulated robot to match the arrangement of multiple target objects in an unstructured environment to a target image. It uses an image segmenter to process current and target images, then sequentially applies deep reinforcement learning-based task sequencing and motion estimation models.

Existing cleaning robots are limited to simple suction tasks, and tidying objects in unstructured environments has historically been inefficient and inflexible, requiring operators to manually pre-define sequences and locations.

This technology proposes a method for planning final robot motions by combining AI-based image segmentation, deep reinforcement learning-based task sequencing and motion estimation, 3D point cloud processing, and generative adversarial networks. This allows the robot to autonomously establish tidying plans without pre-defined instructions. It can be applied to home service robots and automated retail display systems, significantly expanding the range of household and service tasks that robots can perform.

‍

‍

Key Features:

• An image capture unit that generates a current image by capturing the target area where multiple objects are distributed
• An AI-based image segmenter that segments the target image and the current image, respectively
• A tidying planner that receives the segmented target and current images to establish a tidying plan
• A configuration including a task sequencing model that estimates the next object to be tidied