Chronocooked: New RL Benchmark for Temporal Decision-Making
Chronocooked, a new benchmark suite for reinforcement learning (RL), has been developed by researchers to investigate implicit interval timing in artificial agents. Drawing inspiration from the cooperative cooking game Overcooked, this suite features cooking scenarios where agents must utilize temporal information that is not easily visible but essential for achieving optimal results. The design of the environment is deliberately straightforward, facilitating controlled experimentation and supporting biologically plausible models. Evaluation metrics have been crafted to reveal the timing limitations of RL agents. The researchers provide baseline results from three different model types: a non-recurrent model, a recurrent model, and a biologically plausible model. This study emphasizes the importance of integrating time perception and temporal processing in artificial agents, especially for human-robot interactions and their roles in time-sensitive human contexts. The paper can be found on arXiv with the identifier 2608.16666, under Computer Science > Artificial Intelligence.
Key facts
- Chronocooked is a new reinforcement learning benchmark suite.
- It is inspired by the game Overcooked.
- The suite focuses on implicit interval timing in RL agents.
- Temporal information is unobserved but critical for optimal performance.
- The environment is kept simple for controlled experiments.
- Evaluation metrics expose timing limitations in RL agents.
- Baselines include non-recurrent, recurrent, and biologically plausible models.
- The work aims to underscore the need for time perception in AI for human-robot interaction.
Entities
Institutions
- arXiv