enjeeneer/zero-shot-rl

VC-FB and MC-FB algorithms from "Zero-Shot Reinforcement Learning from Low Quality Data" (NeurIPS 2024)

30
/ 100
Emerging

This project helps machine learning researchers improve how they train reinforcement learning agents when they only have a small or limited dataset to learn from. It takes existing, low-quality datasets and applies special algorithms to produce agents that can perform new tasks they haven't seen before, reducing overestimation errors. Researchers working on robotic control, automated decision-making, or simulation-based training would find this useful.

No commits in the last 6 months.

Use this if you need to train intelligent agents to solve new problems without collecting massive, perfect datasets for every single task, especially for locomotion or goal-reaching robots.

Not ideal if your primary goal is to train a single agent for a single, well-defined task with an abundance of high-quality data.

robotics reinforcement-learning-research offline-rl locomotion-control goal-reaching
Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 7 / 25
Maturity 16 / 25
Community 7 / 25

How are scores calculated?

Stars

26

Forks

2

Language

Python

License

MIT

Last pushed

Jan 14, 2025

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/enjeeneer/zero-shot-rl"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.