Sync rSkill files from OpenRAL/openral@466e89dae (1 file(s))

README.md

@@ -46,13 +46,10 @@ openral:
 > **OpenRAL rSkill** — wraps the upstream `moveit_msgs/action/MoveGroup`
 > action server so the Reasoner can dispatch a collision-free Cartesian
 > end-effector pose goal (position + orientation) through the same
-> `ExecuteRskill` path used by VLA skills. No model weights — the manifest
-> is the entire artefact. New under
-> the
-> `rskill-moveit-*` family.
+> `ExecuteRskill` path used by VLA skills. No model weights: the manifest
+> is the entire artefact. Part of the `rskill-moveit-*` family.
 
-This package uses `kind: ros_action` — a discriminator
-on `RSkillManifest.kind` that selects the
+`kind: ros_action` (a discriminator on `RSkillManifest.kind`) selects the
 [`ROSActionRskill`](../../python/rskill/src/openral_rskill/ros_action_rskill.py)
 engine at resolve time, with `ros_integration.goal_builder: pose` selecting
 the [`PoseGoalRskill`](../../python/rskill/src/openral_rskill/pose_goal_rskill.py)
@@ -62,9 +59,8 @@ specialisation) and of [`rskill-moveit-multi-joints`](../rskill-moveit-joints/)
 (joint-space). The engine constructs an `rclpy.action.ActionClient` on the
 host `RskillRunnerNode`, sends one goal built from the lowered `pose` block,
 awaits the result, and replays the returned `trajectory_msgs/JointTrajectory`
-one waypoint per `step()` call onto `/openral/candidate_action` so the safety
-supervisor still applies its per-joint envelope check to every commanded
-position.
+one waypoint per `step()` call onto `/openral/candidate_action` so the C++
+safety kernel still checks every commanded position.
 
 ## What this skill does
 
@@ -135,12 +131,9 @@ trajectory replayed one waypoint per `step()` as a 1-row `JOINT_POSITION`
 | in  | `pose.orientation` | unit quaternion in `pose.quaternion_order` (default `xyzw`) | LLM-overridable |
 | out | per-waypoint `Action` | `joint_targets=[[n_dof floats]]`, `horizon=1`, `is_terminal=False` | One chunk per `step()` until completion is signalled by exception |
 
-Why one row per chunk (`chunk_size: 1` is schema-enforced for
-`kind: ros_action`): the OpenRAL safety supervisor only validates row 0
-of every `ActionChunk` today
-([`supervisor_node.py`](../../packages/openral_safety/openral_safety/supervisor_node.py)).
-Packing the full trajectory as one chunk with `horizon=N` would let
-waypoints 1..N actuate unchecked.
+`chunk_size: 1` is schema-enforced for `kind: ros_action`: each waypoint is
+its own `ActionChunk`, so the C++ safety kernel validates (and can reject) it
+before the next one is sent.
 
 ### GPU-accelerated planning (cuMotion)
 
@@ -212,7 +205,7 @@ reads the `link_name ← tool_frame` offset from TF:
 | `ros_integration.interface_name` | `/move_action` |
 | `ros_integration.goal_builder` | `pose` (selects `PoseGoalRskill`) |
 | `ros_integration.result_trajectory_field` | `planned_trajectory.joint_trajectory` |
-| `commercial_use_allowed` | `true` (apache-2.0) |
+| `is_commercial_use_allowed` | `true` (apache-2.0) |
 
 Full schema:
 [`openral_core.schemas.RSkillManifest`](../../python/core/src/openral_core/schemas.py).
@@ -231,7 +224,10 @@ End-to-end, with a real MoveIt launch up:
 # 1. Bring up MoveIt for your robot (example: Panda)
 ros2 launch moveit_resources_panda_moveit_config demo.launch.py
 
-# 2. Dispatch a Cartesian end-effector pose goal:
+# 2. Boot the OpenRAL deploy graph against the same robot
+openral deploy sim --config scenes/deploy/<deploy-scene>.yaml --robot franka_panda
+
+# 3. Dispatch a Cartesian end-effector pose goal:
 ros2 action send_goal /openral/execute_rskill openral_msgs/action/ExecuteRskill \
     "{rskill_id: 'OpenRAL/rskill-moveit-multi-eef_pose', deadline_s: 30.0, prompt: 'move to pre-grasp',
       goal_params_json: '{\"pose\": {\"position\": [0.4, 0.0, 0.5], \"orientation\": [0.0, 0.0, 0.0, 1.0]}}'}"
@@ -242,24 +238,24 @@ ros2 action send_goal /openral/execute_rskill openral_msgs/action/ExecuteRskill
 - **Reachability is the planner's call.** A pose outside the arm's dexterous
   workspace simply fails to plan — there is no base-repositioning fallback
   here (that is the navigate rung of the ladder).
-- **OpenRAL safety supervisor does not do collision checking.** We trust
-  MoveIt's internal FCL pass; the per-joint envelope check still runs per
-  waypoint. Kernel-side collision checking is a separate ADR + multi-PR effort.
-- **No velocity / jerk bound at the supervisor.** Same posture as
-  `rskill-moveit-multi-joints`: the per-joint position envelope runs per
-  waypoint; richer bounds are tracked separately.
+- **Kernel collision checks are deployment-configured.** MoveIt plans
+  against its own FCL `PlanningScene`; the C++ safety kernel separately checks
+  each waypoint against the per-joint position envelope and, when the deploy
+  enables them, self-collision and world-voxel collision.
+- **No velocity / jerk bound on `JOINT_POSITION` chunks.** Same posture as
+  `rskill-moveit-multi-joints`: position limits per waypoint, not the rate
+  between waypoints.
 
 ## License
 
 The rSkill package itself (this manifest + README) is **Apache-2.0**. The
 wrapped MoveIt code (`moveit_msgs` IDL, `moveit2` planners) is **BSD-3-Clause**
-and lives outside this repository — installed via `ros-${ROS_DISTRO}-moveit`.
-Both postures are commercial-use-permissive.
+and lives outside this repository, installed via `ros-${ROS_DISTRO}-moveit`.
+Both are commercial-use-permissive.
 
 ## See also
 
-- MoveIt goal-builder library + rskill-moveit-* rename
-- ROS-wrapped rSkills
+- [Wrapped-ROS rSkills](../../docs/reference/rskills.md#wrapped-ros-rskills-kind-ros_action)
 - [`openral_rskill.ros_action_rskill`](../../python/rskill/src/openral_rskill/ros_action_rskill.py) — engine source
 - [`openral_rskill.pose_goal_rskill`](../../python/rskill/src/openral_rskill/pose_goal_rskill.py) — goal-lowering adapter
 - [`rskills/rskill-moveit-joints/`](../rskill-moveit-joints/) — sibling joint-space MoveIt wrapper