Setup¶

Prerequisites¶

OS: Linux x86_64
Python: 3.11
GPU: CUDA 12.x (for GraspGen, ManiSkill RT rendering, torch)
pixi: Install pixi package manager

Installation¶

git clone --recursive git@github.com:Robot-TLab/TyGrit.git
cd TyGrit
bash scripts/setup.sh

What `setup.sh` does¶

Step	Command	What it installs
1	`git submodule update --init --recursive`	VAMP, GraspGen, MomaViz
2	`pixi install`	Full conda + pypi environment
3	`pixi run install-vamp`	VAMP motion planner (C++ build)
4	`pixi run install-momaviz`	MomaViz visualization toolkit
5	Symlink gripper config	Register Fetch gripper with GraspGen
6	`pixi run install-graspgen`	pointnet2_ops + GraspGen package
7	`pixi run download-graspgen-weights`	Pretrained GraspGen model weights

ManiSkill assets¶

After setup, download simulation assets:

pixi run python -m mani_skill.utils.download_asset ReplicaCAD
pixi run python -m mani_skill.utils.download_asset ycb

Verify¶

pixi run test

Environment details¶

The environment is managed by pixi via pixi.toml.

Conda dependencies¶

scipy, numpy, loguru, cmake, scikit-build-core, nanobind, pybind11, eigen, nlopt, orocos-kdl, cuda-toolkit 12.8, ros-humble-desktop

PyPI dependencies¶

py-trees, torch 2.7, mani-skill 3, torchvision, hydra-core, trimesh, timm, and more — see pixi.toml for the full list.

Vendored C extensions¶

Extension	Source	Build system
`ikfast_fetch`	`ext/ikfast_fetch/`	setuptools (via `setup.py`)
`pytracik`	`ext/trac_ik/`	pybind11, deps: eigen, nlopt, orocos-kdl

Git submodules¶

Submodule	Path	Purpose
VAMP	`thirdparty/vamp_preview/`	Motion planning (`import vamp_preview`)
GraspGen	`thirdparty/GraspGen/`	6-DOF grasp prediction
MomaViz	`thirdparty/MomaViz/`	Episode visualization

Environments (pixi)¶

Environment	Purpose	Command
`default`	Full dev environment (GPU + ROS)	`pixi run test`
`lint`	Pre-commit hooks only	`pixi run -e lint pre-commit run`
`ci`	CI (no GPU, headless)	`pixi run -e ci test`