Course Syllabus
6.4420[J] / 6.8420: Computational Design and Fabrication
Fall 2026 · Tuesday and Thursday · 75 minutes
Each lab is started in class as a guided follow-along. Students complete it and submit. Submissions are due one week after the lecture at 11:59 PM.
| # | Date | Lecture | Covered | In-class lab | Due |
|---|---|---|---|---|---|
| 1 | Thu Sep 10 | Introduction; Working with AI Agents | Course scope and the computational design pipeline; project requirements and timeline. AI agents, tool use, and the structure of AI-assisted development. | ||
| 2 | Tue Sep 15 | Software Design with AI | Design decisions, formal specification, phased implementation, and test-driven verification as a development method. | Develop a small software tool from design decisions through formal specification, actor–critic review, and a tested first implementation phase. | Tue Sep 22 |
| 3 | Thu Sep 17 | Design representations I: data structures | Geometric representations as data structures: parametric curves and surfaces, boundary representations, triangle and tetrahedral meshes, voxels and octrees, signed distance fields, and neural implicit representations. | Export a solid model in several representations and compare what each encoding preserves and discards. | Thu Sep 24 |
| 4 | Tue Sep 22 | Design representations II: programs, CAD and assemblies | Geometric representations as programs: constructive solid geometry, parametric modeling, CAD feature trees and their reference semantics, multi-part assemblies, and LLM-generated CAD. | Build a parametric CAD model in code, generate an interactive parameter viewer, locate the ranges where the geometry kernel fails, and add constraint handling. | Tue Sep 29 |
| 5 | Thu Sep 24 | Design spaces | Design spaces as representation, parameter ranges and constraints; graph-based design representations; formal grammars and L-systems as generative design spaces; feasibility and sampling strategies. | Sample a parametric model across its parameter ranges, record infeasible configurations, and map the feasible region. | Thu Oct 1 |
| 6 | Tue Sep 29 | Simulation: design to performance | Forward simulation from geometry to performance: discretization, boundary conditions and solver selection. Definition of performance metrics and formulation of design objectives. | Run a finite element analysis under load and perform a mesh convergence study on displacement and peak stress. | Tue Oct 6 |
| 7 | Thu Oct 1 | Building the dataset | Dataset construction for design: generation by sampling and simulation, acquisition from repositories and literature, cleaning, deduplication, unit consistency, and representation choice. | Profile, clean, split and baseline a design dataset, and compare it against a provenance-tracked pipeline assembled from literature. | Thu Oct 8 |
| 8 | Tue Oct 6 | Surrogate models | Surrogate models for design evaluation: training data requirements, model selection, accuracy against simulation cost, and validity outside the training distribution. | Train a surrogate model on a simulated design sweep and evaluate its accuracy against the solver and outside the training distribution. | Tue Oct 13 |
| 9 | Thu Oct 8 | Inverse design | Inverse design: objective and constraint formulation, gradient-based and gradient-free search, and verification of optimized designs against the original forward model. | Optimize a design against a constrained objective using a surrogate, then verify the result with the original solver. | Thu Oct 15 |
| Fri Oct 9 | ADD DATE · Tue Oct 13: no class, Monday schedule | ||||
| 10 | Thu Oct 15 | Manufacturing instructions | Translation from geometry to machine instructions: process abstractions, slicing, toolpath generation, additive and subtractive process constraints, and design for manufacturability. | Generate machine instructions for an optimized design and identify features that violate process constraints. | Thu Oct 22 |
| 11 | Tue Oct 20 | PROJECT PITCH | Team project proposals presented in class. | Submit a one-page written project proposal incorporating feedback from the pitch session. | Tue Oct 27 |
| 12 | Thu Oct 22 | Simulation II: simulation domains | Survey of forward simulation domains: CFD, rigid-body dynamics, molecular dynamics, electromagnetics, thermal, acoustics and contact, compared on evaluation cost, differentiability, discretization sensitivity, stability and tooling availability. | Set up and run a simulation in a second physical domain, assess convergence and discretization adequacy, and record solver cost. | Thu Oct 29 |
| 13 | Tue Oct 27 | Data visualization with AI | Exploratory and explanatory visualization with AI assistance: plot generation from specification, high-dimensional and uncertainty display, and visualization as a diagnostic for data and models. Includes team standups. | Construct an interactive visualization of project data from a written specification. | Tue Nov 3 |
| 14 | Thu Oct 29 | Design representations III: DSLs and metamaterials | Domain-specific languages for design: language design, L-system variants, inverse procedural modeling, and graph grammars. Architected materials as a worked example, including periodic unit cells and homogenization. | Implement a domain-specific language for a periodic unit cell and compute effective properties across its parameter range. | Thu Nov 5 |
| 15 | Tue Nov 3 | Design spaces II: learned and generative | Learned design spaces: autoencoders and latent representations, interpolation, and diffusion and language models as design generators with conditioning and controllability. | Train an autoencoder on a design dataset, interpolate between designs in latent space, and evaluate the validity of decoded samples. | Tue Nov 10 |
| 16 | Thu Nov 5 | Foundation models and fine-tuning | Pretrained models and adaptation: transfer learning, full and parameter-efficient fine-tuning, prompting, and evaluation of scientific and engineering foundation models. Includes team standups. | Fine-tune a pretrained model with a parameter-efficient method and compare against zero-shot performance on the same task. | Thu Nov 12 |
| 17 | Tue Nov 10 | Simulation III: learned simulators and model trust | Learned and hybrid simulation: graph and convolutional network simulators, physics-informed networks, learned preconditioners, and generalization behavior. Model validation, uncertainty quantification and design under uncertainty. Opens with a recap of the Lecture 12 domain comparison. | Replace a numerical solver with a learned surrogate and quantify the resulting accuracy and speed trade-off. | Tue Nov 17 |
| 18 | Thu Nov 12 | Simulation-to-reality gap | Discrepancy between simulation and physical outcome: model error, unmodeled physics, dimensional tolerance and material variability. Domain randomization, system identification and calibration. | Re-evaluate an optimized design under dimensional tolerance and material property variation. | Thu Nov 19 |
| 19 | Tue Nov 17 | PROJECT CHECKPOINT | Mid-project review. | presented in class | |
| Wed Nov 18 | DROP DATE | ||||
| 20 | Thu Nov 19 | Inverse design II: optimization and differentiable simulation | Numerical optimization for design: problem formulation, constraint handling, conditioning, and differentiable simulation with adjoint gradients. | Compare finite-difference and adjoint gradients on a design objective and measure solver calls against parameter count. | Mon Nov 30 |
| 21 | Tue Nov 24 | Inverse design III: topology optimization | Topology optimization: density-based methods, sensitivity analysis, filtering, mesh dependence and checkerboarding. Generative design pipelines. | Run density-based topology optimization with and without filtering across two mesh resolutions. | Tue Dec 1 |
| Thu Nov 26 | NO CLASS: Thanksgiving | ||||
| 22 | Tue Dec 1 | Inverse design IV: multi-objective and design-space exploration | Multi-objective design: Pareto optimality, scalarization and evolutionary methods, dominance and hypervolume. Visualization and navigation of high-dimensional design spaces. | Compute a two-objective Pareto front over a design space and explore it interactively. | in class only |
| 23 | Thu Dec 3 | Self-driving laboratories | Sequential experiment selection: design of experiments, Gaussian process surrogates and acquisition functions, active learning, and autonomous experimentation systems. | Run a Bayesian optimization loop over a design space and compare its sample efficiency against random search. | in class only |
| Fri Dec 4 | ASSIGNMENT CUTOFF: at most one item may fall due after this date | ||||
| 24 | Tue Dec 8 | Sustainability and lifecycle | Environmental cost of design decisions: lifecycle assessment, embodied energy, and material selection as design objectives and constraints. Designated cut if the term runs long, or presentation overflow. | Add an environmental objective to an existing multi-objective problem and recompute the trade-off front. | in class only |
| 25 | Thu Dec 10 | FINAL PRESENTATIONS | Project presentations. | Final report due Thu Dec 10 |
Course Summary:
| Date | Details | Due |
|---|---|---|