sparkmind 1.0.0rc4

Unified Robot Skill Learning Framework

SparkMind

Focused robot policy training recipes on top of LeRobot. The maintained training scripts live under examples/; older experimental demos are archived under legacy/examples/.

---

Maintained Training Recipes

Model	Recipe
ACT	Local LeRobot dataset training
SmolVLA	Local LeRobot dataset training / fine-tuning
PI0	Local LeRobot dataset training / fine-tuning
PI0.5	Local LeRobot dataset training / fine-tuning

See docs/train_guide_local.md for command-line usage. Historical demos for XVLA, Groot, Wall-X, RL, GeomAM, and simulator experiments are kept in legacy/examples/ but are not part of the maintained public recipe surface.

---

Installation

Requirements: Python 3.12 or 3.13, pip, and a clone of this repository.

cd /path/to/SparkMind
# pip install -e . 
pip install -e ".[all]" # install all extras

Optional simulation suites and VLA policy extras (e.g. envs, pi, xvla, all) are documented here:

docs/Installation.md

Docker build and runtime usage for the current IL / VLA image is documented here:

docs/DOCKER_USAGE_ZH.md