A Parametric Tool for Robot Design Exploration
1Department of Mechanical Engineering, MIT 2Department of Architecture, MIT
Draft enables high-level design decision-making for robots through medium-fidelity models and reinforcement learning. It does so by compiling robot kinematic trees from primitive links, which can be parametrized instantly to explore robot performance. Free parameters are fit to characterized trends for robot actuators and links to ground the models in what is credible to build. Below, explore how Draft allows for rapid design adjustment, and investigate how design changes influence performance.
Try it
Move the sliders and Draft regenerates the robot, here in your browser. You set the limbs, the trunk and the actuators. The fitted trends set everything else, and the feasibility gate refuses what no catalogued hardware could build.
This demo exposes a handful of parameters. Draft itself can change any value in a robot's parameter table or kinematic tree, and compiles robots of any tree, not only these two.
Dots are the surveyed catalogue; lines are the trends fitted to it. Hover a black point to see which actuators it is.
Same environment, different robots
Three quadrupeds generated from one description train in the same environment under the same curriculum. Ramp the difficulty and watch where each one gives out, and which of its joints runs closest to its motor's limits.
Robot performance is often limited by the cost of iterating on morphology and control together, since every computer-aided design (CAD) change has to be carried into a simulation-ready model before control work begins. Co-design methods attempt to close this gap, but each uses a model generator written for a single platform or lacks real-world data to suggest that its designs are plausible. We present Draft, a parametric generation tool whose generalized engine compiles any parametric tree of serial chains into a simulation-ready MJCF model, without CAD. It lets engineers explore design tradeoffs through easily adjustable models and evaluate how changes influence controller performance. Draft grounds the free parameters of each design using trends fitted to a survey of 114 actuators and 49 published robot descriptions, so that a generated robot is anchored to real-world hardware. We validate those trends holistically by building twins of four off-the-shelf robots, whose masses agree to a 1.10× geometric mean fold error. Finally, we demonstrate how Draft exposes design tradeoffs by evaluating three quadrupeds through a two-stage reinforcement learning curriculum.
One compiler builds any robot from four link primitives, with no CAD.
Trends fitted to 114 actuators and 49 published robots fill in everything a design leaves unstated.
Rebuilt from geometry and joint specs alone, four commercial robots land within 25% of their mass.
@misc{nguyen2026draft,
title = {Draft: A Parametric Tool for Robot Design Exploration},
author = {Nguyen, David and Coelho, Marcelo and Kim, Sangbae},
year = {2026},
eprint = {2609.38405},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2609.38405}
}