feat: migrate world sync and add plan_to_joint_config from collide-align

[Papaercold]

Migrate and Refactor lwautosim Repository

[Papaercold]

Reach Skill: Extra EE Target Support — Debug & Verification Report

1. Changes

types.py — new field in EnvExtraInfo

object_extra_reach_target_poses: dict[str, dict[str, list[torch.Tensor]]] = field(default_factory=dict)

Structure: object_name → ee_link_name → list of poses [x, y, z, qw, qx, qy, qz] in object frame.
Added matching iterator machinery (_reset_extra_target_pose_iterators, get_next_extra_reach_target_pose).

reach.py — extra EE constraint now actually works

Before: object_extra_reach_target_poses was never read. _build_extra_target_poses only
handled cfg-driven passive modes (keep_current, keep_relative_offset, etc.).

After: Goal computation extracted into _compute_goal_from_offset. When extra_offsets
is populated from EnvExtraInfo.object_extra_reach_target_poses, each pose is transformed
object frame → world frame → robot root frame and passed to cuRobo as link_goals.
Takes priority over cfg-driven modes.

Other fixes in the same refactor:

• Removed duplicate target_pose assignment (dead variable)
• _compute_corrective_goal reuses _compute_goal_from_offset (no duplication)
• Added joint name guard in step() for robots with virtual base joints not in sim
• Removed verbose INFO logs

---

2. Usage

return EnvExtraInfo(
    object_reach_target_poses={
        "target_object": [torch.tensor([x, y, z, qw, qx, qy, qz])],  # primary EE, object frame
    },
    object_extra_reach_target_poses={
        "target_object": {
            "left_hand_link": [torch.tensor([x, y, z, qw, qx, qy, qz])],  # secondary EE, object frame
        }
    },
)

Poses in object frame are automatically transformed to robot root frame at planning time.
Secondary EE targets are soft constraints in cuRobo — satisfaction is not guaranteed if kinematically infeasible.

---

3. Verification

Tested on LW-BenchHub:

Pipeline	Result
CoffeeSetupMugPipeline-v0	✅
CheesyBreadPipeline-v0	✅
KettleBoilingPipeline-v0	✅
OpenFridgePipeline-v0	✅

python run_autosim.py --pipeline_id <PIPELINE_ID> --enable_cameras