SHORE: Stability-Hybrid On-Policy Reinforcement Learning for Online Bin Packing
Overview
This research presents SHORE, a reinforcement learning framework designed for online 3D bin packing tasks that must maintain stack stability under uncertain future arrivals. The method combines a learned stability probability with on-policy return estimates using a 3D U-Net and an actor-critic architecture. Evaluated in Isaac Gym, SHORE achieves 73% space utilization by 0.4M environment steps, outperforming the GOPT baseline (0.588 utilization) on a 60-box set, though with a higher placement-failure rate of 1.03% versus 0.63%. The approach is also demonstrated on a real robot to confirm its integration feasibility.
Generated from attributed reports · Updated 2 days ago
Event evidence and corrections
0 attributed source owners. Ownership does not establish independent confirmation. Quantities are reported separately and are never added together.
Reported quantity · box set count: 60 other · Basis not reported
“ions in both its action values and policy prior, with gates that limit the effect of an untrained or changing model. In deployment, SHORE does not require additional search or simulation overhead. In Isaac Gym, SHORE reaches 73\% space utilization by 0.4M environment steps, while the same learner without stability modeling reaches that level at about 1M. On the expanded 60-box set, SHORE achieves 0.734 space utilizat”
Exact source · revision 1Source owner not reported
Report timeline
Follow attributed reports and material updates.
- arXiv Robotics — research abstractsSignalSHORE: Stability-Hybrid On-Policy Reinforcement Learning for Online Bin Packing
This research presents SHORE, a reinforcement learning framework designed for online 3D bin packing tasks that must maintain stack stability under uncertain future arrivals. The method combines a learned stability probability with on-policy return estimates using a 3D U-Net and an actor-critic architecture. Evaluated in Isaac Gym, SHORE achieves 73% space utilization by 0.4M environment steps, outperforming the GOPT baseline (0.588 utilization) on a 60-box set, though with a higher placement-failure rate of 1.03% versus 0.63%. The approach is also demonstrated on a real robot to confirm its integration feasibility.
Timezone · UTC
Article dates follow your selected timezone. Briefing editions use Hong Kong time (UTC+8).