---
license: mit
tags:
- reinforcement-learning
- stable-baselines3
- sac
- robotics
- gymnasium
- mujoco
---

# SB3 SAC on gym_xarm/XarmLift-v0

This repository contains a Stable-Baselines3 SAC policy trained on `gym_xarm/XarmLift-v0`.

## Files
- `best_model.zip` / `final_model.zip`: SB3 model checkpoint (policy + critics).
- `vecnormalize.pkl`: VecNormalize statistics (required for correct inference).
- `metadata.txt`: simple metadata about the run.

## Inference (important)
You must load the VecNormalize stats before running inference.

Example (pseudo):
1) Create env (same wrappers).
2) Load VecNormalize with `VecNormalize.load(...)`.
3) Load model with `SAC.load(...)` and set env.