Search papers, labs, and topics across Lattice.
This paper introduces a human-in-the-loop reinforcement learning framework for contact-rich robotic manipulation that distills a vision-enabled teacher policy into a vision-free student policy. The distillation leverages real-world data and pose, twist, and wrench sensing to train robust policies without domain randomization or data augmentation. Experiments on the NIST assembly benchmark show the vision-free student achieves 95% success after 50 minutes of training and generalizes to unseen task variants, outperforming baselines in robustness and adaptability.
Vision-free robotic manipulation policies can be trained in under an hour to achieve 95% success and generalize to unseen tasks by distilling knowledge from vision-enabled teachers using real-world reinforcement learning.
When using reinforcement learning (RL) for contact-rich robotic manipulation, vision can provide task-relevant information that accelerates learning beyond what proprioception alone can achieve. However, vision-enabled policies tend to overfit to the visual conditions seen during training, limiting their robustness and transferability. We present a human-in-the-loop RL framework that employs teacher-student distillation to achieve robust performance across multiple task variants, trained entirely in the real world without requiring domain randomization or data augmentation. A vision-enabled teacher distills its knowledge into a vision-free student that relies solely on pose, twist, and wrench sensing, combining fast training with strong task generalization. On the real-world NIST assembly benchmark board, our approach achieves 95\% overall success after approximately 50 minutes of training on 3 representative tasks, including robust generalization to 8 unseen task variants. Fine-tuning with distillation achieves full success on the most challenging task. We demonstrate that the resulting policies outperform baselines in both robustness and adaptability.
