Skip to content

The H1-2 sim runs the real robot's estop and operator gates - #12

Merged
maxconwa merged 2 commits into
mainfrom
sim_estop
Nov 15, 2026
Merged

maxconwa merged 2 commits into
mainfrom
sim_estop

Conversation

@maxconwa

Copy link
Copy Markdown
Contributor

The H1-2 sim runs the real robot's estop and operator gates

Branch sim_estop. The H1-2 sim bringups now start up as the real robot does.

The estop

  • Every H1-2 bringup starts estop_node, and safety_node always reads it. h12_control.launch.py no longer has a use_estop argument.
  • The RoboCasa and Isaac sims serve the real box's serial lines (ESTOP,BOOT, then ESTOP,0|1 at 50 Hz) on socket://127.0.0.1:<15700 + ROS_DOMAIN_ID>.
  • ENTER on the sim's terminal, or in the RoboCasa viewer, toggles the button. A stopped sim reads as unplugged.

Configs and delays

  • sim_safety_split is deleted from golem_safety and golem_upper_ik.
  • The sims run the real pairs, relax_safety_split and safety_split, with the real 3 s / 5 s stagger.

The legs

  • golem_lowerbody_rl/config/h12/sim.yaml waits for /lowerbody/confirm_engage and uses the real 0.25 s max_state_gap.
  • It now differs from real.yaml only in the band release and the IMU trim.

Other changes

  • h12_sim always runs navigation; the use_nav argument is gone.
  • h12_models.launch.py is renamed golem_vision_models.launch.py.
  • The ros image installs python3-serial.
  • CLAUDE.md and docs/RUNNING.md describe the new start-up.

Submodules (merge these first)

Operator flow in the sim

docker/scripts/docker_run.sh robocasa --restart
docker/scripts/docker_run.sh ros --restart && ros2 launch golem_bringup h12_sim.launch.py
ros2 service call /lowerbody/confirm_engage std_srvs/srv/Trigger   # once the legs hold the pre-pose

Tests

  • tests/sim/test_sim_stack.sh --h12: 14 of 14 checks pass, with the confirm and the estop box.
  • tests/robocasa: 182 pass.
  • colcon tests for golem_safety, golem_bringup, golem_lowerbody_rl and estop pass.
  • golem_upper_ik: 7 timing tests fail under the build machine's load, and the same 7 fail on unmodified main in the same container.

Conflict with grasp_benchmark
Both branches add an argument to open_viewer (viewer_sync.py, h12_mujoco.py). Whichever merges second keeps both sim_rate and key_callbacks.

🤖 Generated with Claude Code

maxconwa and others added 2 commits November 15, 2026 15:47
…gates

The H1-2 sim bringups now start up as the real robot does:

- The estop. Every H1-2 bringup starts estop_node, and safety_node always
  reads it (h12_control.launch.py has no use_estop argument). The RoboCasa
  and Isaac sims serve the real box's lines (ESTOP,BOOT, then ESTOP,0|1 at
  50 Hz) on socket://127.0.0.1:<15700 + ROS_DOMAIN_ID>
  (golem_robocasa/estop_server.py, CL_isaaclab_sim's copy), which
  estop_node opens as a pyserial URL (estop submodule). ENTER on the sim's
  terminal or in the RoboCasa viewer toggles the button; a stopped sim
  reads as unplugged.
- The configs and delays. sim_safety_split is deleted from golem_safety and
  golem_upper_ik; the sims run the real pairs (relax_safety_split,
  safety_split) with the real 3 s / 5 s stagger.
- The legs. config/h12/sim.yaml waits for /lowerbody/confirm_engage and
  keeps real.yaml's 0.25 s max_state_gap; it differs from real.yaml only in
  the band release and the IMU trim.
- h12_sim always runs navigation (no use_nav), and h12_models.launch.py is
  renamed golem_vision_models.launch.py.

The ros image installs python3-serial for estop_node's socket:// box. The
tests (golem_bringup, golem_safety, golem_upper_ik, golem_lowerbody_rl,
tests/stack, tests/sim, tests/g1) follow; tests/sim/test_sim_stack.sh --h12
passed 14 of 14 checks with the confirm and the estop box.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MezUeauFS9tKMgZKJxo1HC
…lbacks

open_viewer keeps main's sim_rate (the real-time factor overlay) and this
branch's key_callbacks (the estop box's ENTER toggle); h12_mujoco passes both.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MezUeauFS9tKMgZKJxo1HC
@maxconwa
maxconwa merged commit 03f0c5c into main Nov 15, 2026
4 checks passed
@maxconwa
maxconwa deleted the sim_estop branch November 15, 2026 23:35
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant