English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Deep reinforcement learning excels in various domains but lacksgeneralizability and interoperability. Programmatic RL methods (Trivedi et al.,2021; Liu et al., 2023) reformulate solving RL tasks as synthesizinginterpretable programs that can be executed in the environments. Despiteencouraging results, these methods are limited to short-horizon tasks. On theother hand, representing RL policies using state machines (Inala et al., 2020)can inductively generalize to long-horizon tasks; however, it struggles toscale up to acquire diverse and complex behaviors. This work proposes ProgramMachine Policies (POMPs), which bridge the advantages of programmatic RL andstate machine policies, allowing for the representation of complex behaviorsand the address of long-term tasks. Specifically, we introduce a method thatcan retrieve a set of effective, diverse, compatible programs. Then, we usethese programs as modes of a state machine and learn a transition function totransition among mode programs, allowing for capturing long-horizon repetitivebehaviors. Our proposed framework outperforms programmatic RL and deep RLbaselines on various tasks and demonstrates the ability to generalize to evenlonger horizons without any fine-tuning inductively. Ablation studies justifythe effectiveness of our proposed search algorithm for retrieving a set ofprograms as modes.

Addressing Long-Horizon Tasks by Integrating Program Synthesis and State Machines