Mobile robotic rebar cage assembly via imitation learning

Manipulation remains a key bottleneck in achieving fully autonomous rebar cage assembly. Existing solutions based on rail-guided systems are expensive, poorly scalable, and limited in capability. This paper introduces a framework that leverages a mobile manipulator and uses visual servoing together with imitation learning (IL) to address complex rebar manipulation tasks. The framework enables autonomous execution of two challenging manipulation tasks: (a) tight-fit rebar slot insertion and (b) rebar tying at complex intersection nodes within cages. Using only low-cost RGB cameras, the proposed approach achieves over 90% success rate for over 20 rollouts on both tasks. A highlight is the integration of a segmentation module and a reinsertion strategy that improves tight-fit insertion performance by 41.7% over the baseline and significantly improves robustness to background changes. Notably, the system requires neither depth sensors nor explicit geometric modeling, and supports rapid deployment in novel environments. This paper establishes a foundation for extending autonomy to broader rebar manipulation scenarios. Qualitative results are available on the project website 1 1 https://rebarbot.github.io . . • Integrates imitation learning and visual servoing for autonomous rebar cage assembly. • Achieves the first tight-fit insertion of rebars into assembly slots. • Enables rebar tying at complex intersection nodes with only RGB cameras. • Improves success rate and robustness over a baseline imitation learning rebar policy.