Self-Predictive Representations for Combinatorial Generalization in Behavioral Cloning
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911604967211008 |
|---|---|
| author | Lawson, Daniel Hugessen, Adriana Cloutier, Charlotte Berseth, Glen Khetarpal, Khimya |
| author_facet | Lawson, Daniel Hugessen, Adriana Cloutier, Charlotte Berseth, Glen Khetarpal, Khimya |
| contents | While goal-conditioned behavior cloning (GCBC) methods can perform well on in-distribution training tasks, they do not necessarily generalize zero-shot to tasks that require conditioning on novel state-goal pairs, i.e. combinatorial generalization. In part, this limitation can be attributed to a lack of temporal consistency in the state representation learned by BC; if temporally correlated states are properly encoded to similar latent representations, then the out-of-distribution gap for novel state-goal pairs would be reduced. We formalize this notion by demonstrating how encouraging long-range temporal consistency via successor representations (SR) can facilitate generalization. We then propose a simple yet effective representation learning objective, $\text{BYOL-}γ$ for GCBC, which theoretically approximates the successor representation in the finite MDP case through self-predictive representations, and achieves competitive empirical performance across a suite of challenging tasks requiring combinatorial generalization. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_10137 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Self-Predictive Representations for Combinatorial Generalization in Behavioral Cloning Lawson, Daniel Hugessen, Adriana Cloutier, Charlotte Berseth, Glen Khetarpal, Khimya Machine Learning Artificial Intelligence While goal-conditioned behavior cloning (GCBC) methods can perform well on in-distribution training tasks, they do not necessarily generalize zero-shot to tasks that require conditioning on novel state-goal pairs, i.e. combinatorial generalization. In part, this limitation can be attributed to a lack of temporal consistency in the state representation learned by BC; if temporally correlated states are properly encoded to similar latent representations, then the out-of-distribution gap for novel state-goal pairs would be reduced. We formalize this notion by demonstrating how encouraging long-range temporal consistency via successor representations (SR) can facilitate generalization. We then propose a simple yet effective representation learning objective, $\text{BYOL-}γ$ for GCBC, which theoretically approximates the successor representation in the finite MDP case through self-predictive representations, and achieves competitive empirical performance across a suite of challenging tasks requiring combinatorial generalization. |
| title | Self-Predictive Representations for Combinatorial Generalization in Behavioral Cloning |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2506.10137 |