Learning from Observation: A Survey of Recent Advances
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866912601809616896 |
|---|---|
| author | Burnwal, Returaj Mehta, Hriday Bhatt, Nirav Pravinbhai Ravindran, Balaraman |
| author_facet | Burnwal, Returaj Mehta, Hriday Bhatt, Nirav Pravinbhai Ravindran, Balaraman |
| contents | Imitation Learning (IL) algorithms offer an efficient way to train an agent by mimicking an expert's behavior without requiring a reward function. IL algorithms often necessitate access to state and action information from expert demonstrations. Although expert actions can provide detailed guidance, requiring such action information may prove impractical for real-world applications where expert actions are difficult to obtain. To address this limitation, the concept of learning from observation (LfO) or state-only imitation learning (SOIL) has recently gained attention, wherein the imitator only has access to expert state visitation information. In this paper, we present a framework for LfO and use it to survey and classify existing LfO methods in terms of their trajectory construction, assumptions and algorithm's design choices. This survey also draws connections between several related fields like offline RL, model-based RL and hierarchical RL. Finally, we use our framework to identify open problems and suggest future research directions. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_19379 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Learning from Observation: A Survey of Recent Advances Burnwal, Returaj Mehta, Hriday Bhatt, Nirav Pravinbhai Ravindran, Balaraman Machine Learning Artificial Intelligence Robotics Imitation Learning (IL) algorithms offer an efficient way to train an agent by mimicking an expert's behavior without requiring a reward function. IL algorithms often necessitate access to state and action information from expert demonstrations. Although expert actions can provide detailed guidance, requiring such action information may prove impractical for real-world applications where expert actions are difficult to obtain. To address this limitation, the concept of learning from observation (LfO) or state-only imitation learning (SOIL) has recently gained attention, wherein the imitator only has access to expert state visitation information. In this paper, we present a framework for LfO and use it to survey and classify existing LfO methods in terms of their trajectory construction, assumptions and algorithm's design choices. This survey also draws connections between several related fields like offline RL, model-based RL and hierarchical RL. Finally, we use our framework to identify open problems and suggest future research directions. |
| title | Learning from Observation: A Survey of Recent Advances |
| topic | Machine Learning Artificial Intelligence Robotics |
| url | https://arxiv.org/abs/2509.19379 |