scienceworld-target-identifier

To identify specific objects in ScienceWorld, this skill filters raw room observations against descriptive criteria like 'living thing' or 'container' to extract actionable lists of potential targets.

Object prioritization in ScienceWorld works by ranking identified items based on accessibility and relevance, enabling agents to systematically navigate and interact with the most suitable targets first.

Yes, you can use this skill to find a living thing in a room by processing environmental context and returning a prioritized list of candidates such as 'baby wolf' or 'turtle egg'.

The best way to filter observable items from text-based game environments is to apply bundled classification scripts and taxonomies that categorize objects by matching descriptive criteria.

No additional dependencies are required to categorize targets in ScienceWorld, as the skill utilizes bundled classification scripts and taxonomies to accurately identify and rank objects.

Agents struggle to find specific targets in ScienceWorld when raw room observations are not filtered, a problem solved by transforming environmental context into prioritized lists of accessible objects.