JUICER: Data-Efficient Imitation Learning for Robotic Assembly
This addresses the problem of data-efficient imitation learning for robotic assembly, offering incremental improvements in performance for tasks requiring precise manipulation.
The paper tackles the challenge of achieving high-performance imitation learning for precise, long-horizon robotic assembly tasks with limited human demonstrations, resulting in a pipeline that outperforms baselines by enabling assembly of up to five parts over nearly 2500 time steps from RGB images.
While learning from demonstrations is powerful for acquiring visuomotor policies, high-performance imitation without large demonstration datasets remains challenging for tasks requiring precise, long-horizon manipulation. This paper proposes a pipeline for improving imitation learning performance with a small human demonstration budget. We apply our approach to assembly tasks that require precisely grasping, reorienting, and inserting multiple parts over long horizons and multiple task phases. Our pipeline combines expressive policy architectures and various techniques for dataset expansion and simulation-based data augmentation. These help expand dataset support and supervise the model with locally corrective actions near bottleneck regions requiring high precision. We demonstrate our pipeline on four furniture assembly tasks in simulation, enabling a manipulator to assemble up to five parts over nearly 2500 time steps directly from RGB images, outperforming imitation and data augmentation baselines. Project website: https://imitation-juicer.github.io/.