Expand synthetic generator: parameterized assembly templates #7

New Issue

forbes · 2026-02-02T19:33:08Z

forbes commented

2026-02-02 19:33:08 +00:00

Summary

Expand SyntheticAssemblyGenerator with parameterized assembly templates beyond the current chain/rigid/overconstrained generators. Per Phase 1.2 of the repository plan.

New templates to implement

Tree topology

Random spanning tree with configurable branching factor
Each branch can use a different joint type
Produces underconstrained assemblies (no closed loops)

Loop topology

Single closed loop of bodies (ring)
Closing the loop typically introduces redundancy depending on joint types
Good source of overconstrained training data

Star topology

Central hub body connected to N satellite bodies
All joints radiate from center
Useful for testing many constraints on a single body

Mixed topology

Combines tree + loop elements
Random graph with configurable edge density
Most realistic approximation of real CAD assemblies

Requirements

Each template takes n_bodies and optional joint_types parameter
Configurable body count ranges per complexity tier:
- Simple: 2-5 bodies
- Medium: 6-15 bodies
- Complex: 16-50 bodies
Each template returns (bodies, joints, analysis) like existing generators
Update generate_training_batch() to include new templates in the random selection
Complexity tier parameter on batch generation

Depends on

#5 (generator port)

## Summary Expand `SyntheticAssemblyGenerator` with parameterized assembly templates beyond the current chain/rigid/overconstrained generators. Per Phase 1.2 of the repository plan. ## New templates to implement ### Tree topology - Random spanning tree with configurable branching factor - Each branch can use a different joint type - Produces underconstrained assemblies (no closed loops) ### Loop topology - Single closed loop of bodies (ring) - Closing the loop typically introduces redundancy depending on joint types - Good source of overconstrained training data ### Star topology - Central hub body connected to N satellite bodies - All joints radiate from center - Useful for testing many constraints on a single body ### Mixed topology - Combines tree + loop elements - Random graph with configurable edge density - Most realistic approximation of real CAD assemblies ## Requirements - [ ] Each template takes `n_bodies` and optional `joint_types` parameter - [ ] Configurable body count ranges per complexity tier: - Simple: 2-5 bodies - Medium: 6-15 bodies - Complex: 16-50 bodies - [ ] Each template returns `(bodies, joints, analysis)` like existing generators - [ ] Update `generate_training_batch()` to include new templates in the random selection - [ ] Complexity tier parameter on batch generation ## Depends on - #5 (generator port)

forbes added the phase:1 feature labels 2026-02-02 19:33:08 +00:00

forbes referenced this issue

2026-02-02 19:33:21 +00:00

Expand synthetic generator: geometric diversity #8

forbes referenced this issue

2026-02-02 19:33:53 +00:00

Dataset generation CLI #10

forbes referenced this issue from a commit

2026-02-02 20:38:20 +00:00

feat: parameterized assembly templates and complexity tiers

forbes closed this issue

2026-02-02 20:38:24 +00:00