sapien

What it is and what it does

SAPIEN is a GPU-accelerated physics simulator designed for robotic manipulation and vision research. It provides a large-scale dataset of articulated objects with part-level annotations, built on the ShapeNet and PartNet foundations as a collaborative effort between UCSD, Stanford, and SFU. The simulator uses PhysX 5 for physics and supports GPU-accelerated rendering via its SapienRenderer (formerly VulkanRenderer), including ray tracing and stereo depth sensor simulation.

Version 3.0 introduced a major refactor to an entity-component architecture, replacing the previous Actor-based API. The package is designed for tasks requiring detailed understanding of object parts and their interactions—typical use cases include training robotic grasping policies, simulating complex manipulation tasks, and generating synthetic training data for vision models. It supports both offscreen rendering on headless GPU servers and virtual desktop setups, making it suitable for large-scale distributed robot learning experiments.

Use it for:

• Training robotic grasping and manipulation policies on articulated objects with part-level understanding.
• Generating synthetic training data for robotic vision models using GPU-accelerated rendering and ray tracing.
• Simulating complex multi-step manipulation tasks requiring detailed physics and collision handling.
• Running large-scale distributed robot learning experiments on GPU servers without display.
• Prototyping and validating robot control algorithms in a realistic physics environment before deployment.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

SAPIEN is a physics-rich simulation environment for articulated objects that enables robotic vision and interaction tasks requiring detailed part-level understanding.

Yes, if you are developing robotic manipulation or vision systems and have access to a Linux GPU. SAPIEN is actively maintained, has no known vulnerabilities, uses permissive MIT licensing, and provides a mature physics simulation environment with a large dataset of articulated objects. The GPU requirement and Linux-only runtime are hard constraints; verify your hardware before installing. The medium install friction is typical for compiled simulation packages and is not a blocker for most research or production use.

Install

sapien on PyPI

pip install sapien

uv add sapien

poetry add sapien

Installing sapien

Before you install

Medium install friction due to compiled wheels and GPU requirement. The package is actively maintained with recent release (157 days ago) and supports modern Python versions (3.8–3.14). Pre-built wheels available for macOS, Linux, and Windows, but Linux with NVIDIA, AMD, or Intel GPU is required to run.

License in practice

MIT license permits commercial and private use with minimal restrictions, making it suitable for both research and production robotics projects.

Quickstart

pip install sapien

import sapien
from sapien.core import Scene

scene = Scene()
# Configure physics and rendering for your simulation

Requires Linux with NVIDIA, AMD, or Intel GPU. On headless servers, system dependencies libegl1 and libxext6 are needed for offscreen rendering.

Verify before relying

• Whether the 9 runtime dependencies (numpy, requests, transforms3d, lxml, networkx, pyperclip, opencv-python, setuptools, importlib-resources) are all required for basic usage or only for specific features.
• Performance characteristics and scalability limits for large-scale simulations with many articulated objects.
• Compatibility and integration details with popular robot learning frameworks beyond ManiSkill.

Package facts

Evidence: sapien-3.0.3-cp310-cp310-macosx_12_0_universal2.whl; sapien-3.0.3-cp310-cp310-manylinux_2_28_x86_64.whl; sapien-3.0.3-cp310-cp310-win_amd64.whl; sapien-3.0.3-cp311-cp311-macosx_12_0_universal2.whl; sapien-3.0.3-cp311-cp311-manylinux_2_28_x86_64.whl; sapien-3.0.3-cp311-cp311-win_amd64.whl; sapien-3.0.3-cp312-cp312-macosx_12_0_universal2.whl; sapien-3.0.3-cp312-cp312-manylinux_2_28_x86_64.whl; sapien-3.0.3-cp312-cp312-win_amd64.whl; sapien-3.0.3-cp313-cp313-macosx_12_0_universal2.whl; sapien-3.0.3-cp313-cp313-manylinux_2_28_x86_64.whl; sapien-3.0.3-cp313-cp313-win_amd64.whl; sapien-3.0.3-cp314-cp314-macosx_12_0_universal2.whl; sapien-3.0.3-cp314-cp314-manylinux_2_28_x86_64.whl; sapien-3.0.3-cp314-cp314-win_amd64.whl

Keywords: robotics, simulator, dataset, articulation, partnet

Tags

More Python Modules packages

Further reading