Research map / Architectures and efficiency / Neural architecture search
MechaFormer: Sequence Learning for Kinematic Mechanism Design Automation
Diana Bolanos, Mohammadmehdi Ataei, Pradeep Kumar Jayaraman
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| sample more candidates per curve (Best@k, k=1 to 32) (ablation) | MechaFormer · Nurizada validation curves | median DTW: 3.09 -> 1.61 | Improved |
| add rotational sampling (8 input rotations of 45 degrees) (ablation) | MechaFormer · Nurizada validation curves | median DTW: 3.09 -> 1.76 | Improved |
| add topology sampling over all 24 mechanism types (ablation) | MechaFormer · Nurizada validation curves | success rate: median DTW 2.090, success rate 86.7% | Mixed |
| refine the model output with L-BFGS-B (hybrid, k=32) (ablation) | MechaFormer Best @ k=32 · Nurizada validation curves | median DTW: 1.605 -> 0.887 | Improved |
| initialize L-BFGS-B from MechaFormer samples instead of random starts (ablation) | L-BFGS-B local optimizer · Nurizada validation curves | median DTW: 14.258 -> 0.887 | Improved |
| initialize L-BFGS-B from model samples (10 samples, extended budget) (ablation) | L-BFGS-B local optimizer · 10 validation samples | final median DTW: 11.86 -> 0.71 (k=16), 0.51 (k=32) | Improved |
| increase optimization budget (maxfun 100 to 2000) (ablation) | L-BFGS-B from random init · 10 validation samples | median DTW | No clear effect |
| use MechaFormer instead of KNN retrieval on encoder embeddings | MechaFormer Best @ k=1 · Nurizada validation curves | median DTW: 3.799 -> 3.090 | Improved |
| raise sampling temperature from 0.1 to 1.0 (ablation) | MechaFormer · diversity evaluation | combined diversity: 13.4% -> 26.9% | Mixed |
| use MechaFormer with sampling instead of prior cVAE | MechaFormer · Nurizada dataset | mean Chamfer distance: 0.135 -> 0.119 (sampling), 0.077 (hybrid) | Improved |
| increase k for hybrid method from 16 to 32 (ablation) | Hybrid MechaFormer + L-BFGS-B · Nurizada validation curves | success rate: 94.4% -> 93.4% | Hurt |
Closest papers by meaning
- Aligning Optimization Trajectories with Diffusion Models for Constrained Design Generation (NeurIPS 2023 · similarity 0.85)
- Deep Generative Model for Mechanical System Configuration Design (AAAI 2025 · similarity 0.85)
- RobotSmith: Generative Robotic Tool Design for Acquisition of Complex Manipulation Skills (NeurIPS 2025 · similarity 0.83)
- BikeBench: A Bicycle Design Benchmark for Generative Models with Objectives and Constraints (NeurIPS 2025 · similarity 0.82)
- MorphVAE: Advancing Morphological Design of Voxel-Based Soft Robots with Variational Autoencoders (AAAI 2024 · similarity 0.82)
- Neural Contractive Dynamical Systems (ICLR 2024 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
