csmile-1006/ARP

Guide Your Agent with Adaptive Multimodal Rewards (NeurIPS 2023 Accepted)

27
/ 100
Experimental

This project helps reinforcement learning researchers and practitioners train AI agents more effectively, especially in complex virtual environments like games or simulations. It takes expert demonstrations (recorded actions and observations) and multimodal reward signals (like those from images or language) as input. The output is a trained agent that can perform tasks in new, unseen environments with greater adaptability. This is ideal for those developing AI agents for gaming, robotics simulation, or complex control tasks.

No commits in the last 6 months.

Use this if you need to train AI agents to adapt to varied and novel scenarios using visual or linguistic cues, particularly in environments where traditional reward functions are hard to define.

Not ideal if your AI agent's environment is simple and has a clear, easily definable numerical reward system without the need for complex visual or linguistic understanding.

reinforcement-learning agent-training robotics-simulation game-AI multimodal-AI
Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 7 / 25
Maturity 16 / 25
Community 4 / 25

How are scores calculated?

Stars

33

Forks

1

Language

Python

License

MIT

Last pushed

Sep 25, 2023

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/agents/csmile-1006/ARP"

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