|
SpawnNet: Learning Generalizable Visuomotor Skills from Pretrained Networks
Xingyu Lin*, John So*, Sashwat Mahalingam, Fangchen Liu, Pieter Abbeel
Conferences: ICRA 2024
Workshops: "GMPL" workshop at RSS 2023, "NeuRL-RMW" workshop at CoRL 2023
We adapt dense pre-trained representations to learn generalizable manipulation skills. Our method shows significant improvements in categorical generalization over more naive pre-training paradigms and well-designed from-scratch baselines.
|
Credits for this website template to Jonathan T. Barron. His source code is here.
|
|