NACLab/self-revising-active-inference

Official Implementation for the paper "SR-AIF: Solving Sparse-Reward Robotic Tasks from Pixels with Active Inference and World Models"

24
/ 100
Experimental

This project helps robotics engineers develop and train AI agents that can solve complex robotic tasks, especially when the robot receives very infrequent feedback (sparse rewards) and has to learn from raw visual information. It takes in visual input (pixels) and previous attempts from an "expert" agent, then outputs a more robust and effective control policy for the robot. Robotics researchers and engineers working on autonomous systems or robot control would find this useful.

No commits in the last 6 months.

Use this if you are developing AI for robots that need to learn intricate physical tasks from visual input and sparse rewards, aiming for better performance than initial expert demonstrations.

Not ideal if your robotic tasks have dense, frequent reward signals or if your agents don't rely on visual learning from pixels.

robot-control reinforcement-learning autonomous-robotics robot-vision sparse-rewards
No License Stale 6m No Package No Dependents
Maintenance 2 / 25
Adoption 8 / 25
Maturity 8 / 25
Community 6 / 25

How are scores calculated?

Stars

64

Forks

3

Language

Python

License

Last pushed

Sep 06, 2025

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/NACLab/self-revising-active-inference"

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