Skip to content

Sync feature/isaac-deploy from internal fork#2

Open
lgulich wants to merge 13 commits intoNVIDIA-ISAAC-ROS:impedence-control-supportfrom
lgulich:feature/isaac-deploy
Open

Sync feature/isaac-deploy from internal fork#2
lgulich wants to merge 13 commits intoNVIDIA-ISAAC-ROS:impedence-control-supportfrom
lgulich:feature/isaac-deploy

Conversation

@lgulich
Copy link
Copy Markdown
Collaborator

@lgulich lgulich commented Apr 9, 2026

Summary

Pushes the feature/isaac-deploy branch from the internal GitLab fork to GitHub. This branch includes:

  • All commits from impedence-control-support
  • 539f50e Remove Bazel build files for CMake-only branch
  • d87b6b0 Keep published /clock monotonic across simulation resets

This branch is needed so the isaac monorepo submodule can reference commit d87b6b0 from GitHub instead of from the internal GitLab fork.

Context: Part of migrating the mujoco_ros2_control submodule in the isaac monorepo from the internal GitLab fork to the public GitHub repo (requested in Isaac/isaac MR !5524).

lgulich and others added 13 commits January 26, 2026 10:21
- Add TF2 broadcaster for floating base (publishes odom_frame -> body transform)
- Add configurable odom_frame parameter (defaults to "odom")
- Add publish_floating_base_tf parameter to enable/disable TF publishing
- Add kp/kd command interfaces for impedance control mode
- Impedance control computes: effort = effort_ff + kp * (pos_cmd - pos) + kd * (vel_cmd - vel)
- Handle NaN values gracefully in impedance control
- Support combined position/velocity/effort with kp/kd for full impedance control
- Add /mujoco/reset service to reset simulation to initial state
- Use mj_resetData() to properly reset all qpos, qvel, and time
- Simplify getExecutableDir() using std::filesystem::path
- Use value_or() pattern in get_hardware_parameter_or
- Remove unused clamp() function and render members
- Simplify lambdas and use range-based for loops

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Write impedance effort through MuJoCo's actuator system (ctrl[]) instead
of directly to joint forces (qfrc_applied[]), so it goes through the
MJCF <motor> actuators defined in the model.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add build targets for MuJoCo hardware interface plugin, simulate module,
ros2_control_node binary, and ament_index package registration.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add tags = ["nolint"] to ros2_cc_library and ros2_cc_binary targets
since this is third-party code not subject to our linting rules.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace local mujoco/mujoco-3.4.0/ directory references with @Mujoco
http_archive from GitHub releases. This fixes CI which couldn't access
the gitignored mujoco directory. Also disables lint tests (nolint tag)
since this is third-party code.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The mujoco simulate module includes GLFW/glfw3.h which comes from
the system libglfw3-dev package. CI remote executors don't have this
package, so provide headers via an http_archive instead.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The GLFW header includes GL/gl.h by default which isn't available on
CI remote executors. MuJoCo simulate doesn't need OpenGL headers
directly, so define GLFW_INCLUDE_NONE to skip the include.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Build GLFW from its GitHub release source using the _GLFW_NULL headless
backend so no X11/Wayland/GL system headers are needed. Remove -lglfw
system link flag since GLFW is now a proper Bazel dependency.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…code

- Rename PKG to DEPLOY_ROS2_CONTROL for clarity
- Add _HARDWARE_INTERFACE_DEPS_SPDLOG_SAFE using :libfmt_hdrs_only to
  avoid spdlog's bundled fmt/ shadowing real libfmt-dev headers via
  -isystem ordering
- Add _LIBFMT_COPTS with explicit -I path for fmt headers
- Add tags = ["nolint"] to suppress cpplint/copyright/uncrustify on
  third-party code
- Add SPDLOG_FMT_EXTERNAL define

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
mj_resetData sets mj_data_->time back to 0, which caused the /clock
topic to jump backward.  With use_sim_time=true on the controller
manager this produced negative measured_period values that corrupted
PID controllers and destabilised learned policies.

Track a cumulative clock_offset_ that is incremented by the current
sim time before each reset, and add it in publish_clock() so the
published time never decreases.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@lgulich lgulich force-pushed the impedence-control-support branch 3 times, most recently from 4367f94 to cab7619 Compare April 14, 2026 12:44
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