Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2506.08630 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866914025183379456 |
|---|---|
| author | Engwegen, Laurens Brinks, Daan Böhmer, Wendelin |
| author_facet | Engwegen, Laurens Brinks, Daan Böhmer, Wendelin |
| contents | A universal controller for any robot morphology would greatly improve computational and data efficiency. By utilizing contextual information about the properties of individual robots and exploiting their modular structure in the architecture of deep reinforcement learning agents, steps have been made towards multi-robot control. Generalization to new, unseen robots, however, remains a challenge. In this paper we hypothesize that the relevant contextual information is partially observable, but that it can be inferred through interactions for better generalization to contexts that are not seen during training. To this extent, we implement a modular recurrent architecture and evaluate its generalization performance on a large set of MuJoCo robots. The results show a substantial improved performance on robots with unseen dynamics, kinematics, and topologies, in four different environments. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_08630 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Modular Recurrence in Contextual MDPs for Universal Morphology Control Engwegen, Laurens Brinks, Daan Böhmer, Wendelin Artificial Intelligence Robotics A universal controller for any robot morphology would greatly improve computational and data efficiency. By utilizing contextual information about the properties of individual robots and exploiting their modular structure in the architecture of deep reinforcement learning agents, steps have been made towards multi-robot control. Generalization to new, unseen robots, however, remains a challenge. In this paper we hypothesize that the relevant contextual information is partially observable, but that it can be inferred through interactions for better generalization to contexts that are not seen during training. To this extent, we implement a modular recurrent architecture and evaluate its generalization performance on a large set of MuJoCo robots. The results show a substantial improved performance on robots with unseen dynamics, kinematics, and topologies, in four different environments. |
| title | Modular Recurrence in Contextual MDPs for Universal Morphology Control |
| topic | Artificial Intelligence Robotics |
| url | https://arxiv.org/abs/2506.08630 |