Mechanical Engineering
Graduate Seminar
NOTICE OF SEMINAR PRESENTATION
CANDIDATE: Farzaneh Yousefi
DEGREE SOUGHT: MASc
DATE: 4/6/2026
TIME: 10:30am
PLACE: Room 1102 CEI
TITLE: Online Robotic 3D Bin Packing via Reinforcement Learning and Learning-from-Demonstration
Abstract\
This research addresses the Online Three-Dimensional Bin Packing Problem by proposing a hybrid framework that combines Deep Reinforcement Learning (DRL) with Learning from Demonstration (LfD). By developing a custom 3D simulation, I captured human packing strategies to pretrain the agent, bypassing the slow and unstable early stages of traditional machine learning. The system learns to execute discrete action primitives inspired by robotic manipulation—such as alignment, stability-aware stacking, and compact placement, which are then refined using the Proximal Policy Optimization (PPO) algorithm. This approach achieves higher space utilization than pure reinforcement learning baselines, providing an efficient, stable, and collision-free pipeline for automated robotic packing.