A Composable Specification Language for Reinforcement Learning Tasks

Abstract

Reinforcement learning is a promising approach for learning control policies for robot tasks. However, specifying complex tasks (e.g., with multiple objectives and safety constraints) can be challenging, since the user must design a reward function that encodes the entire task. Furthermore, the user often needs to manually shape the reward to ensure convergence of the learning algorithm. We propose a language for specifying complex control tasks, along with an algorithm that compiles specifications in our language into a reward function and automatically performs reward shaping. We implement our approach in a tool called SPECTRL, and show that it outperforms several state-of-the-art baselines.

Cite

Text

Jothimurugan et al. "A Composable Specification Language for Reinforcement Learning Tasks." Neural Information Processing Systems, 2019.

Markdown

[Jothimurugan et al. "A Composable Specification Language for Reinforcement Learning Tasks." Neural Information Processing Systems, 2019.](https://mlanthology.org/neurips/2019/jothimurugan2019neurips-composable/)

BibTeX

@inproceedings{jothimurugan2019neurips-composable,
  title     = {{A Composable Specification Language for Reinforcement Learning Tasks}},
  author    = {Jothimurugan, Kishor and Alur, Rajeev and Bastani, Osbert},
  booktitle = {Neural Information Processing Systems},
  year      = {2019},
  pages     = {13041-13051},
  url       = {https://mlanthology.org/neurips/2019/jothimurugan2019neurips-composable/}
}