Embedding Morphology into Transformers for Cross-Robot Policy Learning
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866914358702899200 |
|---|---|
| author | Suzuki, Kei Liu, Jing Wang, Ye Hori, Chiori Brand, Matthew Romeres, Diego Koike-Akino, Toshiaki |
| author_facet | Suzuki, Kei Liu, Jing Wang, Ye Hori, Chiori Brand, Matthew Romeres, Diego Koike-Akino, Toshiaki |
| contents | Cross-robot policy learning -- training a single policy to perform well across multiple embodiments -- remains a central challenge in robot learning. Transformer-based policies, such as vision-language-action (VLA) models, are typically embodiment-agnostic and must infer kinematic structure purely from observations, which can reduce robustness across embodiments and even limit performance within a single embodiment. We propose an embodiment-aware transformer policy that injects morphology via three mechanisms: (1) kinematic tokens that factorize actions across joints and compress time through per-joint temporal chunking; (2) a topology-aware attention bias that encodes kinematic topology as an inductive bias in self-attention, encouraging message passing along kinematic edges; and (3) joint-attribute conditioning that augments topology with per-joint descriptors to capture semantics beyond connectivity. Across a range of embodiments, this structured integration consistently improves performance over a vanilla pi0.5 VLA baseline, indicating improved robustness both within an embodiment and across embodiments. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_00182 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Embedding Morphology into Transformers for Cross-Robot Policy Learning Suzuki, Kei Liu, Jing Wang, Ye Hori, Chiori Brand, Matthew Romeres, Diego Koike-Akino, Toshiaki Robotics Artificial Intelligence Machine Learning Systems and Control Cross-robot policy learning -- training a single policy to perform well across multiple embodiments -- remains a central challenge in robot learning. Transformer-based policies, such as vision-language-action (VLA) models, are typically embodiment-agnostic and must infer kinematic structure purely from observations, which can reduce robustness across embodiments and even limit performance within a single embodiment. We propose an embodiment-aware transformer policy that injects morphology via three mechanisms: (1) kinematic tokens that factorize actions across joints and compress time through per-joint temporal chunking; (2) a topology-aware attention bias that encodes kinematic topology as an inductive bias in self-attention, encouraging message passing along kinematic edges; and (3) joint-attribute conditioning that augments topology with per-joint descriptors to capture semantics beyond connectivity. Across a range of embodiments, this structured integration consistently improves performance over a vanilla pi0.5 VLA baseline, indicating improved robustness both within an embodiment and across embodiments. |
| title | Embedding Morphology into Transformers for Cross-Robot Policy Learning |
| topic | Robotics Artificial Intelligence Machine Learning Systems and Control |
| url | https://arxiv.org/abs/2603.00182 |