Environment-Independent Task Specifications via GLTL
This work addresses the challenge of creating environment-independent task specifications for reinforcement learning, though it appears incremental as it builds on existing LTL frameworks.
The authors tackled the problem of environment-dependent reward functions in Markov decision processes by proposing a new task-specification language called GLTL, a variant of Linear Temporal Logic extended for probabilistic specifications, which allows approximations to be learned in finite time and is demonstrated in small environments to specify reinforcement-learning tasks straightforwardly.
We propose a new task-specification language for Markov decision processes that is designed to be an improvement over reward functions by being environment independent. The language is a variant of Linear Temporal Logic (LTL) that is extended to probabilistic specifications in a way that permits approximations to be learned in finite time. We provide several small environments that demonstrate the advantages of our geometric LTL (GLTL) language and illustrate how it can be used to specify standard reinforcement-learning tasks straightforwardly.