Shared command foundation
Create, inspect, undo, serialize, and replay project changes through deterministic operations shared by every interface.
Manifold
Build robots, worlds, and experiments in one AI-native workbench
Manifold is a browser-oriented robotics workspace being built for a continuous path from robot and arena design to motion, simulation, and learning. People, scripts, and the AI agent work through the same validated commands, so every change remains inspectable and reproducible.
Product architectureOne command path, every interface
ProjectRobotKinematicsArenaSimulationLearning
Product direction
Robotics workflows should be visual, deterministic, and AI-ready
Manifold is designed around a shared product model rather than a collection of disconnected tools. The graphical interface, public API, scripts, and AI planner all emit the same typed commands, pass through the same validation, and update the same versioned project state.
First-class robot modeling
Build articulated robots from links, joints, actuators, sensors, frames, materials, and reusable assets, with URDF and MJCF interchange.
Kinematics and control
Pose robots, solve targets, inspect limits, define trajectories, and connect position, velocity, and effort controllers.
Simulation to learning
Construct arenas, run reproducible physics, record results, and grow toward datasets, policies, evaluation, and robot-learning experiments.
Five-phase roadmap
A deliberate path from scene state to robot intelligence
The roadmap builds the contracts that later capabilities depend on. AI integration begins with the command foundation and becomes more autonomous as the robotics system matures.
The phases describe the product direction. Capabilities are released only after their behavior, validation, and integration boundaries are tested.
Project, scene, and commands
Establish versioned project data, transforms, scene entities, undo and redo, serialization, the 3D viewport, and the first shared AI capabilities.
Robot modeling
Represent articulated robots as native project data and support validated editing, reusable assets, and common robot-description formats.
Kinematics and control
Add robot state, forward and inverse kinematics, Jacobians, target poses, trajectories, joint-limit handling, and controller definitions.
Simulation and arenas
Introduce engine adapters, deterministic time stepping, dynamics, contacts, sensors, arena construction, recordings, metrics, and playback.
Advanced AI and robot learning
Enable complete experiment plans, result-aware revisions, reusable tasks, datasets, policies, batch simulation, randomization, and evaluation.