Skip to main content

A robotics and language benchmark

Project description

Language-World

A language and robotics benchmark based on Meta-World.

Language-World is packaged as a set of small tools which process Meta-World observations. Wrapping Meta-World environments directly is left up to the user. To avoid confusion, please always use the goal-observable, randomized goal and initial state variation of Meta-World (MT50-rand), and always sample length 500 episodes. If you choose not to, please prominently document what configuration you used instead.

Language-World can be installed via pip:

pip install git+https://github.com/krzentner/language-world.git@v0.1.0

Alternatively, copy this file into your project.

Language-World consists of three main components:

First, a set of natural language task descriptions available as a dictionary: language_world.TASK_DESCRIPTIONS: dict[str, str]

Second, a set of "scripted skills," that can be used to solve MT10 (and also function in other tasks).

language_world.SCRIPTED_SKILL_NAMES: list[str] lists the names of all scripted skills. language_world.run_scripted_skill(skill_name: str, obs: np.ndarray) -> np.ndarray produces an action given a skill name and observation.

Thirdly, a query answering function:

language_world.eval_queries(task_name: str, queries: List[str],
                            obs: np.ndarray, fuzzy: bool=False)

The query answering function can evaluate a large number of queries. Enabling the fuzzy flag will map unsupported queries to the nearest supported queries using string edit distance.

language_world.enumerate_base_queries(task_name: str) -> list[str] provides a list of the "base" queries for a task (i.e. those that do not contain conjunctions).

language_world.enumerate_all_queries(task_name: str) -> list[str] provides a list of all queries for a task.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

language_world-0.1.1.tar.gz (12.5 kB view hashes)

Uploaded Source

Built Distribution

language_world-0.1.1-py3-none-any.whl (11.1 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page