Walker C1 EDU: Simulation Platform

The Walker C1 EDU simulation is built on NVIDIA Isaac Sim (Isaac Lab 2.2.0). A ROS2-ZMQ bridge publishes simulation state and images on exactly the same ROS 2 topics as the real robot, enabling sim-to-real consistent teleoperation and data collection. The whole environment runs in a Docker container.

C1 shares the same SDK topic surface (mc_task_msgs / mc_state_msgs) with Walker S2 EDU, but the message packages are isolated and selected by UBT_SIM_ROBOT=walker_c1. Both robots use simulation bridge ports 5655-5658, so they cannot run at the same time.

Quick start

Commands are split into two parts: build/enter the container on the host; launch simulation and collect data inside the container (container root is /ubt_sim).

1. Build and start the container (host)

cd ubt_sim/docker
bash run.sh build      # build the ROS 2 + Isaac Sim image
bash run.sh start      # create/start the container
bash run.sh init       # build robot message packages (including walker_c1_msgs), first time only
bash run.sh check      # verify GPU / ROS 2 / message packages / numpy<2

See Simulation Platform · Container Setup for the build workflow.

⚠️ Must be changed before building (real-robot or cross-machine deployment): edit interfaceWhiteList in ubt_sim/docker/fastdds_no_shm.xml and set <address> to the IP of the direct-link NIC (Thor side 192.168.11.3, Walker peer 192.168.11.99). The file is copied into the image at build time; changes after the build have no effect. Pure local simulation over loopback can keep the default.

2. Launch simulation (container)

# Enter the container (ROS 2 and C1 message packages sourced automatically)
UBT_SIM_ROBOT=walker_c1 bash run.sh bash

# Option 1: C1-specific script (GPU rendering + CPU PhysX, 100 Hz physics step by default)
bash scripts/start_c1_pick_place_sim.sh

# Option 2: unified launcher
UBT_SIM_TASK=UBTSim-WalkerC1-Parlor-v0 bash scripts/start_sim.sh

Startup prints the ZMQ ports, action joint order, and a hand joint limit/gain summary.

• Press R to reset the robot/scene; Space requests saving the current episode (works with the ROS-side recorder).
• When the simulation window closes, the exit trap in start_sim.sh stops the bridge automatically.

Demo (picking an apple from the table):

Common environment variables/arguments (combinable):

Variable / argument	Purpose	Default
UBT_SIM_TASK	Task name (determines the robot)	UBTSim-TienkungPro-Parlor-v0; use UBTSim-WalkerC1-Parlor-v0 for C1
UBT_SIM_ROBOT	Message package selection inside the container	Set to walker_c1 when entering the container
UBT_SIM_WALKER_C1_USD_PATH	Override the robot USD	usd/walker_c1_merged.usd
UBT_SIM_C1_STEP_HZ	Physics step rate	100
UBT_SIM_LOAD_ONLY=1	Load the scene only, no bridge	Off
--headless	Run without a window	Common on headless servers
--device cpu\|cuda	Execution device	cpu (default in the C1-specific script)

3. Data collection (container, system Python 3.10)

# Single episode: reset scene -> ready pose -> one online-IK pick-and-place -> save only on success
bash teleoperation/control/walker_c1/save_data.sh

# Batch: 20 episodes, reset between episodes, save only successful ones
bash teleoperation/control/walker_c1/save_data.sh --batch 20

# Interactive recording (Enter/s/q)
bash teleoperation/control/walker_c1/save_data.sh --manual

• Episode flow: reset scene -> task ready pose -> one online-IK pick-and-place -> save only if the task succeeds.
• Sampling rate 15 Hz; one camera_head stream at 640×480 PNG, 31 body joints + 6 channels per hand.
• Output path: /ubt_sim/dataset/walker_c1/<epoch seconds>/trajectory.hdf5 (directories 0777, files 0666).
• Before collecting, verify /mc/sdk/robot_state is publishing (about 48 Hz) and keep hand gains at 10/1 with effort_limit_sim=2.

HDF5 layout for one episode:

Path	Shape	Meaning
joint_observations/joint_position	(T, 31)	L arm 0-6 / R arm 7-13 / head 14-15 / waist 16-18 / legs 19-30
hand_observations/hand_left_position	(T, 6)	Left hand SDK 6 channels
hand_observations/hand_right_position	(T, 6)	Right hand SDK 6 channels
action_observations/*	Same	Most recently commanded values
camera_observations/color_images/camera_head	(T,)	PNG byte stream, 640×480
timestamp	(T,)	Collection timestamp

4. Reset (container)

# Task ready pose (three stages: raise shoulder -> fold elbow -> ready pose)
/usr/bin/python3 teleoperation/control/walker_c1/reset.py --mode task

# Scene reset (equivalent to pressing R in the simulation window)
/usr/bin/python3 teleoperation/control/walker_c1/reset.py --mode sim-scene

reset.py also supports sim-task (scene + task ready pose) and sdk-home (all SDK joints at zero).

5. Dataset preview and playback (container)

# Preview an HDF5 episode (Rerun)
/usr/bin/python3 teleoperation/tools/playback_walker_c1.py --mode preview

# Action playback (same topics in simulation and on the real robot)
/usr/bin/python3 teleoperation/tools/playback_walker_c1.py \
    --episode dataset/walker_c1/<epoch seconds>/trajectory.hdf5 --mode control

See Playback for all options and safety notes.

Simulation modules

Module	Path	Description
Docker orchestration	ubt_sim/docker/run.sh	Image build, container lifecycle, bridge start/stop
Environment variables	ubt_sim/docker/env.sh	Container name, image, ROS_DOMAIN_ID
Simulation launcher	ubt_sim/scripts/start_sim.sh	Detects the robot from UBT_SIM_TASK and starts the bridge
C1-specific launcher	ubt_sim/scripts/start_c1_pick_place_sim.sh	GPU rendering + CPU PhysX, 100 Hz step
Task definition	ubt_sim/source/ubt_sim/task/walker_c1_parlor/	Parlor apple-picking scene
Simulation config	ubt_sim/config/walker_c1/parlor.yaml	Scene/camera/device parameters
ROS2-ZMQ bridge	ubt_sim/teleoperation/bridges/walker_c1/	Bridge script + walker_c1_bridge_config.yaml
Control and collection	ubt_sim/teleoperation/control/walker_c1/	reset.py / pick_place_controller.py / save_data.sh / online IK
Playback	ubt_sim/teleoperation/tools/playback_walker_c1.py	Preview / action playback
Data output	ubt_sim/dataset/walker_c1/<timestamp>/trajectory.hdf5	Collected episodes

ZMQ ports

Port	Purpose
5655	Control commands (bridge PUB)
5656	Robot state (bridge SUB)
5657	Raw camera images (RGB + depth, C++ bridge)
5658	JPEG stream (debugging)

The ubt_IL C1 plugin uses 5661/5662/5663, which do not conflict with the simulation bridge ports 5655-5658, so both can run together. See Architecture · ZMQ ports for the cross-robot overview.

ROS 2 topics

Direction	Topic	Type
sub	/mc/sdk/robot_command	mc_task_msgs/RobotCommand
sub	/mc/left_hand/command, /mc/right_hand/command	mc_task_msgs/JointCommand
pub	/mc/sdk/robot_state	mc_state_msgs/RobotState (31 joints)
pub	/mc/{left,right}_hand/joint_states	sensor_msgs/JointState (SDK 6-joint order)
pub	/sim/object_state	std_msgs/String
pub	/sim/episode_complete	std_msgs/Bool
pub	/sensor/camera/head/color/raw	sensor_msgs/Image (BEST_EFFORT)

Troubleshooting

• bash run.sh check reports errors: follow the prompt and rerun bash run.sh init; verify numpy<2 and walker_c1_msgs.
• Collection script fails to import rclpy / message types: make sure you are using /usr/bin/python3, not Isaac Sim Python, and that bash run.sh init has been run.
• Bridge exits with RMW implementation not installed: the image only ships rmw_cyclonedds_cpp. start_sim.sh / save_data.sh detect it automatically; export it manually when starting the bridge yourself.
• Address already in use (tcp://*:5656): a previous bridge did not exit cleanly. kill -9 the remaining sim_runner / ros2_zmq_bridge, confirm ports 5655-5658 are free, then restart.
• Command streams mix when Walker S2 is also running: both use ports 5655-5658; run only one robot at a time.
• Controller reports no /mc/sdk/robot_state: the bridge is not running (usually a port conflict). Check with ros2 topic hz /mc/sdk/robot_state (about 48 Hz).
• Apple is missed or flies away: keep hand gains at 10/1 and effort_limit_sim=2; do not increase them to old debug values.
• X11 unavailable / no simulation window: run xhost + on the host and verify DISPLAY is forwarded; on a headless machine use --headless with save_data.sh --batch.

After collection, continue with Imitation Learning Platform.