ICL Wins the First Prize ($60k) in ICRA Competition on Bimanual Packing

🎉 Exciting news from ICRA 2025! Our team won First Prize ($60k) in the Packing Track of the What Bimanual Teleoperation and Learning from Demonstration Can Do (WBCD) competition!

We developed a neuro-symbolic framework—called Skill Graph—that strategically decomposes the manipulation problem to leverage the best-suited method for each component:

🔹 Data-driven models for complex perception

🔹 Human knowledge for intuitive priors (e.g., “stacked bowls should be grasped from the interior”)

🔹 Assured and robust model-based robotics techniques, including fast motion planning and safe control

This hybrid approach enabled:

✅ Strong generalization across tasks (e.g., object types) and embodiments (e.g., robotic platforms and sensor configurations)—we trained on Unitree G1 and only accessed the competition setup 3 days before the event;

✅ Data efficiency—perception models were adapted with just one hour of retraining on a RTX 4060 laptop; skills are learned through one-shot key frame demonstrations.

✅ High reliability, achieving a >90% task success rate.

💡 We were also the only team to run in fully autonomous mode in the packing track during the competition!

Huge thanks to the amazing team that made this possible! 🙌