Port JacobianVerifier to solver/datagen/jacobian.py #3

New Issue

forbes · 2026-02-02T19:32:18Z

forbes commented

2026-02-02 19:32:18 +00:00

Summary

Move JacobianVerifier class from data/synthetic/pebble-game.py (L375-902) into solver/datagen/jacobian.py.

Class overview

Builds the constraint Jacobian matrix for numerical rank verification. The pebble game gives a combinatorial necessary condition for rigidity; the Jacobian provides the sufficient condition by detecting geometric degeneracies (e.g., parallel revolute axes creating hidden dependencies).

The generalized velocity vector for n bodies is v = [v1_x, v1_y, v1_z, w1_x, w1_y, w1_z, ..., vn_x, ..., wn_z]. Each scalar constraint contributes one row to J such that dC_i/dt = J_i @ v = 0.

Methods to port

__init__(bodies) — sets up body index mapping, empty Jacobian
_body_cols(body_id) — column range for a body in J
add_joint_constraints(joint) — dispatches to per-type builder
_make_row() — zero row of width 6*n_bodies
_skew(v) — skew-symmetric matrix for cross product
Joint builders (one per JointType):
- _build_ball (3 rows), _build_fixed (6), _build_revolute (5)
- _build_cylindrical (4), _build_slider (5), _build_planar (3)
- _build_distance (1), _build_parallel (3), _build_perpendicular (1)
- _build_universal (4), _build_screw (5)
_perpendicular_pair(axis) — generates orthogonal basis
get_jacobian() — returns full J matrix
numerical_rank(tol) — SVD-based rank
find_dependencies(tol) — incremental rank check to identify redundant rows

Requirements

Import types from solver.datagen.types (RigidBody, Joint, JointType)
Full type annotations
Preserve Jacobian derivation docstrings (Chappuis reference)
No behavioral changes
__all__ = ["JacobianVerifier"]

Depends on

#1 (shared types)

## Summary Move `JacobianVerifier` class from `data/synthetic/pebble-game.py` (L375-902) into `solver/datagen/jacobian.py`. ## Class overview Builds the constraint Jacobian matrix for numerical rank verification. The pebble game gives a combinatorial necessary condition for rigidity; the Jacobian provides the sufficient condition by detecting geometric degeneracies (e.g., parallel revolute axes creating hidden dependencies). The generalized velocity vector for n bodies is `v = [v1_x, v1_y, v1_z, w1_x, w1_y, w1_z, ..., vn_x, ..., wn_z]`. Each scalar constraint contributes one row to J such that `dC_i/dt = J_i @ v = 0`. ## Methods to port - `__init__(bodies)` — sets up body index mapping, empty Jacobian - `_body_cols(body_id)` — column range for a body in J - `add_joint_constraints(joint)` — dispatches to per-type builder - `_make_row()` — zero row of width 6*n_bodies - `_skew(v)` — skew-symmetric matrix for cross product - Joint builders (one per JointType): - `_build_ball` (3 rows), `_build_fixed` (6), `_build_revolute` (5) - `_build_cylindrical` (4), `_build_slider` (5), `_build_planar` (3) - `_build_distance` (1), `_build_parallel` (3), `_build_perpendicular` (1) - `_build_universal` (4), `_build_screw` (5) - `_perpendicular_pair(axis)` — generates orthogonal basis - `get_jacobian()` — returns full J matrix - `numerical_rank(tol)` — SVD-based rank - `find_dependencies(tol)` — incremental rank check to identify redundant rows ## Requirements - [ ] Import types from `solver.datagen.types` (RigidBody, Joint, JointType) - [ ] Full type annotations - [ ] Preserve Jacobian derivation docstrings (Chappuis reference) - [ ] No behavioral changes - [ ] `__all__ = ["JacobianVerifier"]` ## Depends on - #1 (shared types)

forbes added the phase:1 port labels 2026-02-02 19:32:18 +00:00

forbes referenced this issue

2026-02-02 19:32:27 +00:00

Port analyze_assembly to solver/datagen/analysis.py #4

forbes referenced this issue

2026-02-02 19:33:37 +00:00

Ground truth labeling pipeline #9

forbes referenced this issue from a commit

2026-02-02 19:50:18 +00:00

feat: port JacobianVerifier to solver/datagen/jacobian.py

forbes closed this issue

2026-02-02 19:50:19 +00:00