Trajectory First: A Curriculum for Discovering Diverse Policies
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866911674637746176 |
|---|---|
| author | Braun, Cornelius V. Auddy, Sayantan Toussaint, Marc |
| author_facet | Braun, Cornelius V. Auddy, Sayantan Toussaint, Marc |
| contents | Being able to solve a task in diverse ways makes agents more robust to task variations and less prone to local optima. In this context, constrained diversity optimization has become a useful reinforcement learning (RL) framework for training a set of diverse agents in parallel. However, existing constrained-diversity RL methods often under-explore in complex tasks such as robot manipulation, resulting in limited behavioral diversity. We address this with a two-stage curriculum that introduces a spline-based trajectory prior as an inductive bias to produce diverse, high-reward behaviors in an initial stage, and then distills these behaviors into reactive, step-wise policies in a second stage. In our empirical evaluation, we provide novel insights into challenges of diversity-targeted training and show that our curriculum increases the diversity of learned skills while maintaining high task performance. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_01568 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Trajectory First: A Curriculum for Discovering Diverse Policies Braun, Cornelius V. Auddy, Sayantan Toussaint, Marc Machine Learning Robotics Being able to solve a task in diverse ways makes agents more robust to task variations and less prone to local optima. In this context, constrained diversity optimization has become a useful reinforcement learning (RL) framework for training a set of diverse agents in parallel. However, existing constrained-diversity RL methods often under-explore in complex tasks such as robot manipulation, resulting in limited behavioral diversity. We address this with a two-stage curriculum that introduces a spline-based trajectory prior as an inductive bias to produce diverse, high-reward behaviors in an initial stage, and then distills these behaviors into reactive, step-wise policies in a second stage. In our empirical evaluation, we provide novel insights into challenges of diversity-targeted training and show that our curriculum increases the diversity of learned skills while maintaining high task performance. |
| title | Trajectory First: A Curriculum for Discovering Diverse Policies |
| topic | Machine Learning Robotics |
| url | https://arxiv.org/abs/2506.01568 |