On Pre-Training for Visuo-Motor Control: Revisiting a Learning-from-Scratch Baseline

Nicklas Hansen, Zhecheng Yuan, Yanjie Ze, Tongzhou Mu, Aravind Rajeswaran, Hao Su, Huazhe Xu, Xiaolong Wang
Proceedings of the 40th International Conference on Machine Learning, PMLR 202:12511-12526, 2023.

Abstract

In this paper, we examine the effectiveness of pre-training for visuo-motor control tasks. We revisit a simple Learning-from-Scratch (LfS) baseline that incorporates data augmentation and a shallow ConvNet, and find that this baseline is surprisingly competitive with recent approaches (PVR, MVP, R3M) that leverage frozen visual representations trained on large-scale vision datasets – across a variety of algorithms, task domains, and metrics in simulation and on a real robot. Our results demonstrate that these methods are hindered by a significant domain gap between the pre-training datasets and current benchmarks for visuo-motor control, which is alleviated by finetuning. Based on our findings, we provide recommendations for future research in pre-training for control and hope that our simple yet strong baseline will aid in accurately benchmarking progress in this area. Code: https://github.com/gemcollector/learning-from-scratch.

Cite this Paper

BibTeX
@InProceedings{pmlr-v202-hansen23c, title = {On Pre-Training for Visuo-Motor Control: Revisiting a Learning-from-Scratch Baseline}, author = {Hansen, Nicklas and Yuan, Zhecheng and Ze, Yanjie and Mu, Tongzhou and Rajeswaran, Aravind and Su, Hao and Xu, Huazhe and Wang, Xiaolong}, booktitle = {Proceedings of the 40th International Conference on Machine Learning}, pages = {12511--12526}, year = {2023}, editor = {Krause, Andreas and Brunskill, Emma and Cho, Kyunghyun and Engelhardt, Barbara and Sabato, Sivan and Scarlett, Jonathan}, volume = {202}, series = {Proceedings of Machine Learning Research}, month = {23--29 Jul}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v202/hansen23c/hansen23c.pdf}, url = {https://proceedings.mlr.press/v202/hansen23c.html}, abstract = {In this paper, we examine the effectiveness of pre-training for visuo-motor control tasks. We revisit a simple Learning-from-Scratch (LfS) baseline that incorporates data augmentation and a shallow ConvNet, and find that this baseline is surprisingly competitive with recent approaches (PVR, MVP, R3M) that leverage frozen visual representations trained on large-scale vision datasets – across a variety of algorithms, task domains, and metrics in simulation and on a real robot. Our results demonstrate that these methods are hindered by a significant domain gap between the pre-training datasets and current benchmarks for visuo-motor control, which is alleviated by finetuning. Based on our findings, we provide recommendations for future research in pre-training for control and hope that our simple yet strong baseline will aid in accurately benchmarking progress in this area. Code: https://github.com/gemcollector/learning-from-scratch.} }
Endnote
%0 Conference Paper %T On Pre-Training for Visuo-Motor Control: Revisiting a Learning-from-Scratch Baseline %A Nicklas Hansen %A Zhecheng Yuan %A Yanjie Ze %A Tongzhou Mu %A Aravind Rajeswaran %A Hao Su %A Huazhe Xu %A Xiaolong Wang %B Proceedings of the 40th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2023 %E Andreas Krause %E Emma Brunskill %E Kyunghyun Cho %E Barbara Engelhardt %E Sivan Sabato %E Jonathan Scarlett %F pmlr-v202-hansen23c %I PMLR %P 12511--12526 %U https://proceedings.mlr.press/v202/hansen23c.html %V 202 %X In this paper, we examine the effectiveness of pre-training for visuo-motor control tasks. We revisit a simple Learning-from-Scratch (LfS) baseline that incorporates data augmentation and a shallow ConvNet, and find that this baseline is surprisingly competitive with recent approaches (PVR, MVP, R3M) that leverage frozen visual representations trained on large-scale vision datasets – across a variety of algorithms, task domains, and metrics in simulation and on a real robot. Our results demonstrate that these methods are hindered by a significant domain gap between the pre-training datasets and current benchmarks for visuo-motor control, which is alleviated by finetuning. Based on our findings, we provide recommendations for future research in pre-training for control and hope that our simple yet strong baseline will aid in accurately benchmarking progress in this area. Code: https://github.com/gemcollector/learning-from-scratch.
APA
Hansen, N., Yuan, Z., Ze, Y., Mu, T., Rajeswaran, A., Su, H., Xu, H. & Wang, X.. (2023). On Pre-Training for Visuo-Motor Control: Revisiting a Learning-from-Scratch Baseline. Proceedings of the 40th International Conference on Machine Learning, in Proceedings of Machine Learning Research 202:12511-12526 Available from https://proceedings.mlr.press/v202/hansen23c.html.

Related Material