Autonomous Navigation for Laikago Quadruped

Developing an intelligent robotic system is multi-disciplinary work, integrating dynamics, control, computer vision, artificial intelligence, and more. It is hard to master all these fields. Even if you master them all, it takes a lot of time to get it right and robust.

To help roboticists accelerate the development of intelligent robots, NVIDIA Isaac SDK includes reference applications and platforms. One of these is platforms is Kaya, a three-wheeled holonomic autonomous robot. The Laikago application was built using Kaya as a reference to create an autonomous machine that can navigate and avoid obstacles.

Laikago is a quadruped robot made by Unitree Robotics. It features a locomotion control algorithm for walking and balancing in the Microcontroller unit (MCU). It also provides a mount as an optional package, which includes the NVIDIA Jetson TX2 Module for a brain, allowing users to develop custom software and access locomotion control and the sensor data. Out of the box, it does not have any sensor or software for mapping, localization, or obstacle avoidance.

This application uses a Velodyne VLP-16lidar for perception, and integrates the Isaac SDK navigation stack with the Unitree Robotics API. All computation is done inside the TX2. The locomotion controller runs at 500Hz, while the navigation stack requires 50% of the CPU. When running this application, the Laikago walks at a peak speed of 0.6 m/s.


You will need to mount both the lidar and the interface box to the Laikago. We designed 3D printed mounts based on the Laikago CAD files from the Unitree GitHub repo. Laikago provides 19V output, which is higher than the operational voltage of the VLP-16.


The newest VLP-16 supports up to 32V from its specs.

We suggest using a DC-DC converter to decrease the voltage and a USB-to-Ethernet to connect the lidar sensor to the TX2. You can also mount a compatible camera for object detection, though it is not applied in this example application. The figure below shows the overall setup.


The Isaac SDK navigation and perception stack is agnostic to the sensor brand and type of Jetson board. For instance, Kaya uses the same navigation stack but runs on Jetson Nano and uses a camera instead of a lidar sensor for localization. Many of the perception algorithms for Kaya also work on a simple webcam.

This application primarily uses the Isaac SDK navigation stack, which includes maps, localization, global path planning, control, obstacle avoidance, odometry, and path tracking. The Isaac SDK also includes a lidar driver and the Laikago SDK, so no extra libraries or dependencies are required.

The figure below shows the design hierarchy. All rounded boxes are included in Isaac SDK. The rectangular boxes specify the robot hardware. The Laikago driver is mostly used to pass the messages from Isaac SDK to the Laikago SDK.


  1. Make sure the Jetson device is set up as detailed in the Setup document.

  2. Build an ARM target and deploy it to the robot with the following command:


    bob@desktop:~/isaac$ ./engine/build/ --remote_user <username> -p //packages/laikago/apps:laikago_navigate-pkg -d jetpack43 -h <robot_ip_address>

    Where <username> is your username on the robot (nvidia by default), and <robot_ip_address> is the IP address of the robot.

  3. SSH into the robot with the following command:


    bob@desktop:~/isaac$ ssh <username>@<robot_ip_address>

    Where <username> is your user name on the robot (nvidia by default), and <robot_ip_address> is the IP address of the robot.

  4. Run the Laikago application on the robot with the following command:


    bob@jetson:~/$ ./packages/laikago/apps/laikago_navigate --config <map_config_json> --graph <map_graph_json>

    Where <map_config_json> and <map_graph_json> are the map files. Samples are provided in the apps/assets/maps folder.

  5. Connect to a Bluetooth joystick controller. We use the NVIDIA Shield controller in this example. This can be used to both send directional commands to the Laikago and trigger the autonomous navigation mode.


By default, the robot is in the “stance” mode. When the directional command passes a small threshold, the robot will start to walk. See the Joystick section for more information.

  1. Open Isaac Sight in a browser at <robot_ip>:3000. You should see the map where the Laikago is localized. Use the joystick to move the Laikago around and observe the map updates.

© Copyright 2018-2020, NVIDIA Corporation. Last updated on Oct 31, 2023.