Skip to content

Carry the seventh shape wire element and refuse physics explicitly - #40

Merged
Jepson2k merged 5 commits into
mainfrom
feat/mujoco-world-model
Sep 8, 2026
Merged

Carry the seventh shape wire element and refuse physics explicitly#40
Jepson2k merged 5 commits into
mainfrom
feat/mujoco-world-model

Conversation

@Jepson2k

@Jepson2k Jepson2k commented Sep 5, 2026

Copy link
Copy Markdown
Collaborator

Part of the coordinated feat/mujoco-world-model change. Branch-matched PRs:
waldoctl (types), par6 (backend), Waldo-Commander (frontend).

waldoctl's Shape wire form grows a seventh element, physics, which marks a
shape as a body in a backend's contact simulation. parol6 has no contact
simulation, so it accepts the wider wire and refuses physics explicitly
rather than ignoring it — a silently dropped body would be a shape the caller
believes is simulated and the arm treats as scenery.

Testing

parol6's suite plus Waldo-Commander's, which exercises this backend end to end
(360 passed).

🤖 Generated with Claude Code

Jepson2k and others added 5 commits September 4, 2026 21:31
waldoctl's Shape.to_wire() now emits [kind, params, pose, collision,
margin, name, physics]. ShapeWire carries the seventh element (defaulted,
so pre-physics peers still decode) and hands it to shape_from_wire, and
the set-level validation refuses any shape that declares physics: this
backend has no contact simulation, and silently flattening a body to its
geometry would let a program believe it was being simulated.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
# Conflicts:
#	tests/integration/test_shapes_e2e.py
`--examples` is opt-in and CI ran a bare `pytest`, so nothing had executed
these scripts in a long time — and they are the code a user copies. Half of
them failed on their own happy path against a stock simulator.

Three never referenced the robot before commanding planned motion, which is
refused outright. `precision.py` looked like it did: it homed only when the
arm was not already NEAR the home angles, but a freshly started robot reports
positions it has not homed to, so the guard skipped the very step the moves
below needed. Proximity is not referencing; all three now home first, as the
refusal's own remedy says.

`demo_showcase.py` swept the wrist +/-20 degrees about each tool axis, which
at that pose leaves the reachable set part-way through and comes back as
"IK: partial path failure". Ten degrees clears it and still shows what the
sweep is there to show — the pencil tip holding still while the wrist turns.

CI now runs them after the suite, so the next one to rot says so.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The example and the program loaded in Waldo Commander had drifted apart, and
the WC one is the copy that gets run and watched. It sweeps 40 degrees with
the pencil offset on tool X; the example had been changed to offset on tool Z
by b1fe085 and nobody re-checked the sweep against it, so at 20 degrees the
rotation path left the workspace and the move was refused with "IK: partial
path failure" (190/200 poses reachable).

Copied across as-is rather than reconciled by hand, so the script that ships
as the example is the script that actually gets demonstrated. It runs
standalone too — `pytest tests/test_examples.py --examples` drives it as a
subprocess against the simulator.

Note b1fe085's reasoning still stands on its own terms ("SSG-48 tool
transform is along Z"), and precision.py still carries the Z offset; if that
one is wrong too it wants the same treatment.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Same treatment as demo_showcase, and the same divergence: b1fe085 moved the
pencil offset to tool Z in the example while the program actually run and
watched in Waldo Commander kept it on tool X and sweeps 40 degrees.

Copied across as-is so the script that ships as the example is the script
that gets demonstrated. Verified standalone — `pytest
tests/test_examples.py::test_example_runs[precision.py] --examples` drives it
as a subprocess against the simulator.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@Jepson2k
Jepson2k merged commit 4272b98 into main Sep 8, 2026
13 checks passed
@Jepson2k
Jepson2k deleted the feat/mujoco-world-model branch September 8, 2026 11:33
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