Installation#
RPent installs with a single pip install and provides several optional
dependency combinations.
Prerequisites#
Linux with an NVIDIA GPU.
A CUDA 12.x-compatible NVIDIA driver.
Python 3.10–3.12.
git,bash, and a working C toolchain.
You will also want:
An API key for at least one LLM provider — Anthropic, OpenAI, or an OpenAI-compatible chat endpoint — for the planner.
1. Install RPent with pip#
Clone RPent (for the CLI and run configs) and install with the extra for the stack you want:
git clone https://github.com/RLinf/RPent rpent && cd rpent
# Recommended default:
pip install -e ".[libero-pro]" # LIBERO-PRO
Other environment configurations are available when needed:
pip install -e ".[robocasa]" # RoboCasa
pip install -e ".[robotwin]" # RoboTwin
.[libero-pro] is the recommended default.
Available extras:
Extra |
Installs |
|---|---|
|
Standard LIBERO + openpi Pi0.5 VLA + SAM 3.0 + RLinf runtime |
|
LIBERO-PRO + openpi Pi0.5 VLA + SAM 3.0 + RLinf runtime |
|
LIBERO-plus + openpi Pi0.5 VLA + SAM 3.0 + RLinf runtime |
|
RoboCasa365 simulator + the RLDX-1 VLA; see RoboCasa |
|
RoboTwin simulation and LingBot inference dependencies; see RoboTwin |
|
RLinf runtime only |
|
SAM 3.0 only |
2. Download the assets required to run LIBERO#
The Python packages installed with pip do not include the large resource files required to run LIBERO. Choose one command based on the extra installed above:
libero-download-assets --skip-existing # .[libero]
liberopro-download-assets --skip-existing # .[libero-pro]
liberoplus-download-assets --skip-existing # .[libero-plus]
These resources usually need to be downloaded only once;
--skip-existing skips files that are already present.
Tip
If your connection to Hugging Face is slow, download through the
mirror by prefixing the command with HF_ENDPOINT:
HF_ENDPOINT=https://hf-mirror.com liberopro-download-assets --skip-existing
3. (Optional) Real-world robot dependencies#
Franka and SO-101 support is being rolled in; when it lands, each
robot’s robot package will live under robots/<name>/ with its own
README.md describing the SDK / firmware requirements. See
Franka and SO-101 for the current status.
Checking the installation#
The quickest way to confirm everything is wired correctly is to run one LIBERO task end-to-end — see Quick Start. If that succeeds, the env server, VLA server, SAM3 server, and agent are all healthy.
If something breaks:
The env server log is at
<output_dir>/env_server.log.The VLA server writes to
<output_dir>/vla_server.log.The SAM3 server writes to
<output_dir>/sam3_server.log.The agent’s own run log lives at
<output_dir>/run.log.
These logs are always in that per-run scratch directory, so a failed run is self-contained and easy to inspect.