CUHK's PackLab: an MLLM framework for closed-loop robotic bin packing beats RL
CUHK-CSE · hf · 2026-09-24
- PackLab is a full framework for robotic bin packing, a long-horizon sequential decision task where each placement affects space for subsequent packing.
- PackLab-Suite: physics-based simulation platform for scalable training trajectory generation and physical outcome evaluation.
- PackLab-VLM: a packing-specialized MLLM that understands evolving object/container states and jointly selects objects and predicts placements in closed loop.
- PackLab-Bench: standardized scenarios at multiple difficulty levels.
- Experiments show PackLab-VLM beats geometric heuristics, traditional RL, and general-purpose MLLMs on average. Code, model, dataset and benchmark are open-sourced on GitHub.
More from Embodied
- Zuck reveals Muse-powered 'tamagotchi' charm, says it ships before the holidays — cyrus_zei · 2026-09-24
- Robots fold laundry and write code, but camera calibration still takes hours — RexDouglass · 2026-09-24
- Liquid AI partners with Qualcomm to bring on-device 'Liquid Context' to Snapdragon — JosephJacks_ · 2026-09-24
- 36B Agent Model Runs Fully Local on Snapdragon X2 Elite With Just 32GB RAM — Kyrannio · 2026-09-24
- EgoSuite and continuous learning for robotics shown at Humanoids Summit Seoul, next at IROS — jonstephens85 · 2026-09-24
- Alexandr Wang Shows Two Robots Playing Tennis — alexandr_wang · 2026-09-24