The On-board Computer

We are almost done!

The last step before running policy on the real robot is to set up the environment on the on-board NUC computer.

Install Ubuntu 22.04

We find that Ubuntu 22.04 works well with our software and hardware stack on the robot. Follow Ubuntu tutorial to install Ubuntu 22.04 on the NUC computer.

Note

To enter BIOS on the BeeLink N95 NUC computer, press XXXX during bootup.

Install Dependencies

sudo apt install build-essential cmake net-tools can-utils python3-pip

Test connection with joint actuators

As the first step, we need to initialize the CAN transport on the Linux side. To do this, we have prepared a script:

source ./scripts/start_can_transports.sh

Note

Naturally, to stop the transports, you can do:

source ./scripts/stop_can_transports.sh

To test connection to individual actuator, run this command.

python ./berkeley_humanoid_lite_lowlevel/motor/ping.py --port can0 --id 1

--port corresponds to the port of the CAN device on linux. With the two arms and legs all connected, it should be can[0,1,2,3]

--id corresponds to the CAN ID of the device. It should be in range [1..14].

Alternatively, to test connection to all the motors, run this script.

python ./berkeley_humanoid_lite_lowlevel/robot/anyonehere.py

Test connection with IMU

To test connection to the IMU, run this command.

python ./berkeley_humanoid_lite_lowlevel/robot/test_imu.py

Test connection with Joystick

Run this script to launch the Python program that reads the joystick and broadcast the reading on UDP port.

python ./berkeley_humanoid_lite_lowlevel/policy/udp_joystick.py

Run everything together

There are two ways to run the lowlevel code: C codebase and Python codebase.

Note

Before running the program, make sure that the CAN transports are started correctly:

source ./scripts/start_can_transports.sh

C codebase

For locomotion tasks, we recommend to use the C codebase for better realtime gaurantee. The C codebase is under ./csrc/ directory.

On the on-board computer, do the following:

make
./build/real_humanoid

Real-time kernel performance

If running into performance issues where the USB-CAN adapter cannot maintain the required communication frequency, we can use the https://xanmod.org/ real-time kernel instead.

Python codebase

Go to the ./source/berkeley-humanoid-lite-lowlevel/ directory, and then run

uv run ./scripts/python/run_locomotion.py

The robot will follow the same state transition as in the Mujoco sim2sim simulation.

To visualize the robot, run

uv run ./scripts/sim2real/visualize.py