OCEAN: Online Task Inference for Compositional Tasks with Context Adaptation
OCEAN is a principled task inference framework for meta-reinforcement learning. It provides accurate task inference by modeling tasks with global and local latent context variables.
Method
Real-world tasks often exhibit a compositional structure that contains a sequence of simpler sub-tasks. For instance, opening a door requires reaching, grasping, rotating, and pulling the door knob. Such compositional tasks require an agent to reason about the sub-task at hand while orchestrating global behavior accordingly.
This can be cast as an online task inference problem, where the current task identity, represented by a context variable, is estimated from the agent's past experiences with probabilistic inference. Previous approaches have employed simple latent distributions, e.g., Gaussian, to model a single context for the entire task. However, this formulation lacks the expressiveness to capture the composition and transition of the sub-tasks.
We propose a variational inference framework OCEAN to perform online task inference for compositional tasks.
OCEAN models global and local context variables in a joint latent space, where the global variables represent a mixture of sub-tasks required for the task, while the local variables capture the transitions between the sub-tasks. Our framework supports flexible latent distributions based on prior knowledge of the task structure including Dirichlet distribution, Gaussian distribution, Categorical distribution and Logit-normal distribution. The whole framework can be trained in an unsupervised manner without the need of labels for task or sub-task transition.
We created several new environments based on Mujoco. Each environment requires finishing a sequence of sub-task with HalfCheetah and Humanoid agents. OCEAN outperforms previous methods in both sample efficiency and asymptotic performance.
Please refer to our paper for detailed explanations and more results.
Code
A reference implementation of
OCEAN in Python is available on
GitHub.
Contributors
The following people contributed to OCEAN:
Hongyu Ren
Yuke Zhu
Jure Leskovec
Anima Anandkumar
Animesh Garg
References
OCEAN: Online Task Inference for Compositional Tasks with Context Adaptation. H. Ren, Y. Zhu, J. Leskovec, A. Anandkumar, A. Garg.
Conference on Uncertainty in Artificial Intelligence (UAI), 2020.