0
Article ? AI-assigned paper type based on the abstract. Classification may not be perfect — flag errors using the feedback button. Tier 2 ? Original research — experimental, observational, or case-control study. Direct primary evidence. Sign in to save

Study on efficient planning method of dynamic sorting of waste plastic bottles based on deep reinforcement learning

Computers & Industrial Engineering 2026
Shilong Xie, Xinning Li, Hu Wu, Xiaoyu Wang, Yuzhe Zhang, Shanshan Yu

Summary

Scientists have developed a smarter AI-powered robot system that sorts plastic bottles for recycling far faster and more accurately than current methods, correctly identifying plastic types 96% of the time. This matters because better sorting means cleaner, higher-quality recycled plastic, which helps keep more plastic out of landfills and oceans—reducing the microplastic pollution that's been linked to potential health risks in humans.

The accumulation of plastic waste and its derivative microplastics poses serious environmental and health risks, drawing global concern. Therefore, efficient and precise sorting of waste plastic bottles is critical for improving the purity of recycled materials, promoting resource circulation, and mitigating ecological hazards. However, conventional methods primarily manual operations or First-Come, First-Served (FCFS) sorting robots, existed suffer from low efficiency, high cost, and limited adaptability, making them inadequate for dynamic industrial scenarios involving randomly distributed and diverse plastic types. To address these challenges, this study proposes a dynamic sorting efficient planning method based on the Guided Learning Proximal Policy Optimization (GL-PPO) algorithm. A masking mechanism is introduced to restrict invalid action exploration, enhancing policy search efficiency in complex environments. In addition, a robust feature extraction and noise suppression strategy is designed to improve high-dimensional dynamic state representation and stabilize training. Experimental results demonstrate that the proposed method achieves a 98.6% success rate in dynamic sorting planning, 96.1% sorting accuracy, and an average execution time of 7.4 s, significantly outperforming FCFS and PPO algorithms, with nearly a threefold increase in throughput. This study presents a scalable and intelligent solution for plastic waste sorting, contributing to green recycling and sustainable environmental governance.

Share this paper