A flexible framework for connecting legged robots, input devices, and locomotion algorithms, especially learning-based ones. Out-of-the-box policies and configurations can be found in stepit_zoo.

Caution

Disclaimer: User acknowledges that all risks and consequences arising from using this code shall be solely borne by the user, the author assumes no liability for any direct or indirect damages, and proper safety measures must be implemented prior to operation.

Note: This project is under active development, which means the interface is unstable and breaking changes are likely to occur frequently.

Quick Start

Setup the workspace, build StepIt, and run the demo configuration:

mkdir stepit_ws && cd stepit_ws
bash -c "$(curl -fsSL https://raw.githubusercontent.com/chengruiz/stepit/main/scripts/setup.sh)"
./scripts/build.sh
./scripts/run.sh ./configs/demo.conf.sh

Manual Setup

Prerequisites

Tested on Ubuntu 20.04, 22.04 and 24.04.

sudo apt install cmake build-essential
sudo apt install libboost-dev libboost-filesystem-dev libboost-program-options-dev \
                 libeigen3-dev libfmt-dev libyaml-cpp-dev
mkdir -p stepit_ws/src && cd stepit_ws
git clone https://github.com/chengruiz/stepit.git src/stepit

Build

# In the stepit_ws directory
cmake -Bbuild -Ssrc/stepit -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=install
cmake --build build -j && cmake --install build

You can configure the build by passing CMake variables via -D flags:

• STEPIT_WHITELIST_PLUGINS (string): semicolon-separated list of plugins to build. Default is all available plugins.
• STEPIT_BLACKLIST_PLUGINS (string): semicolon-separated list of plugins not to build. Default is "".
• STEPIT_PLUGIN_DIRS (string): semicolon-separated list of directories to search for plugins.

StepIt searches for plugins in directories plugin and the specified STEPIT_PLUGIN_DIRS. All subdirectories with CMakeLists.txt and without STEPIT_IGNORE are considered plugins. Built-in plugins are listed below. Read the corresponding README.md first if you use any of the plugins.

• control_console: Controlling with standard input.
• debugging_helper: Helper utilities for debugging.
• field_base: Field registration and generic field operators.
• joystick_base: Interface for general joystick and joystick control.
• joystick_udp: Controlling with retroid gamepads.
• joystick_usb: Controlling with USB joysticks, e.g. Xbox 360 controller.
• nnrt_base: Interface for neural networks.
• nnrt_ascendcl: Neural network inference on Ascend AI processors, e.g. OrangePI AIpro.
• nnrt_onnxruntime: Neural network inference on general x86_64 processors.
• nnrt_rknnrt: Neural network inference on Rockchip platforms, e.g. RK3588.
• nnrt_tensorrt: Neural network inference on NVIDIA GPUs and Jetson platforms, e.g. Jetson Orin NX.
• nnrt_torchjit: Neural network inference with TorchScript (PyTorch JIT) models.
• policy_neuro: Neural network-based control policy.
• policy_neuro_motion_tracking: Motion tracking related modules for plugin policy_neuro.
• policy_neuro_ros: ROS extensions for plugin policy_neuro.
• policy_neuro_ros2: ROS2 extensions for plugin policy_neuro.
• publisher_csv: Publisher for writing robot data to csv file.
• pyutils: Python utilities for C++ modules.
• robot_deeprobotics_lite3: Controlling the DeepRobotics Lite3 robot.
• robot_deeprobotics_x30: Controlling the DeepRobotics X30 robot (Deprecated).
• robot_unitree_aliengo: Controlling the Unitree Aliengo robot.
• robot_unitree_go1: Controlling the Unitree Go1 robot.
• robot_unitree_b1: Controlling the Unitree B1 robot.
• robot_unitree2: Controlling Unitree Go2, B2 and G1 robots, and with Unitree joysticks.
• robot_unitree2_ros2: Controlling Unitree robots with unitree_ros2.
• ros_base: ROS extensions, e.g. subscribing joysticks and publishing states.
• ros2_base: ROS2 extensions, e.g. subscribing joysticks and publishing states.

Run

# Run StepIt to control the robot
./install/bin/stepit

Command line arguments:

Option	Type	Description
-c / --control	string	The control input type.
-f / --factory	string	The default factory for a specified type.
-P / --publisher	string	The publisher type.
-p / --policy	string	The policy type and directory.
-r / --robot	string	The controlled robot type.
-v / --verbosity	int	Verbosity level ranging from [0, 3]. Defaults to 2.
-- [arg1 arg2 ...]	—	Additional arguments passed to plugins' entry function.

Run ./install/bin/stepit --help for more information.

Notes

• Environment Variable: StepIt reads environment variables listed in environment.sh. Environment variables have lower precedence than their corresponding command-line arguments.
• Control Input: Generally, StepIt accepts strings as control inputs, and returns a response for each one. These inputs can be provided via the console, joysticks, or topics/services in ROS1/ROS2. See doc/control.md for more details.
• State Publishing: StepIt publishes the robot states to CSV files, or via ROS1 or ROS2 topics.
• Plugin Mechanism: StepIt provides a flexible plugin architecture to enable integration of new modules without modifying existing code. See doc/plugin.md for more details.
• ROS Intergrations: StepIt works well with ROS1 and ROS2. Make sure you read the README.md of ROS1 or ROS2 before you start.
• Simulation: StepIt supports unitree_mujoco-based simulation. See robot_unitree2 or stepit_sim for more details.