A Godot 4.x-based physics simulation environment inspired by tools like Endorphin, focused on interactive runtime editing, ragdoll experimentation, and extensible simulation "world states" (e.g., vacuum vs atmosphere). The long-term goal is to integrate AI-driven optimization strategies to reduce CPU workload while preserving believable motion.
This project is licensed under the GNU General Public License v3.0. See the License section for details.
This project is under active development. Core interaction, selection, UI tooling, ragdoll workflows, fluid dynamics, and ML inference are in place, with larger simulation systems (soft bodies, procedural animation) planned.
-
Blender-like editor workflow Outliner + object properties panel, runtime editing, scene-style organization
-
3D viewport interaction Select objects via mouse click directly in 3D space, with visual highlight/outline on selected objects
-
Drag & drop object manipulation Grab and move rigid bodies directly in 3D space at runtime using raycasting; supports flick/throw velocity and floor collision awareness
-
Runtime physics authoring for basic rigid bodies Edit common properties at runtime for primitives (cube, sphere, etc.), including:
- Mass
- Gravity scale
- Friction
- Absorbent
- Rough
- Bounce
- Angular damping
- Linear damping
-
Ragdolls with per-bone control Ragdolls are implemented and support runtime editing per bone, including:
- Mass (per bone)
- Gravity scale (per bone)
- Angular damping (per bone)
- Linear damping (per bone)
-
Switchable environment states Multiple simulation modes managed by a global
EnvironmentManagerautoload that applies external forces consistently across the scene, including:- Vacuum (no external drag/resistance)
- Atmosphere (air drag/resistance applied as per-tick forces)
-
Fluid dynamics (FluidVolume) Spawnable
Area3D-based fluid volumes with configurable density, buoyancy, and drag. Objects and ragdoll bones interact with fluid volumes based on their own density, with visual color indication for placed volumes. -
ONNX-based ML inference (via GDExtension) Integrates the Godot ONNX AI Models Loader GDExtension to load and run pre-trained
.onnxmodels directly inside Godot. Exposes anONNXLoadernode withload_model(path)andpredict(input_data)methods, enabling in-engine ML inference as a foundation for the planned AI optimization layer.
Most game engines provide solid physics, but experimentation-heavy simulation workflows often need:
- Fast iteration with live parameter tuning
- Clear per-part control (especially for ragdolls)
- Simple toggles for world "conditions" (vacuum/air, etc.)
- A path toward smarter performance strategies beyond brute-force CPU stepping
This project aims to become a focused sandbox for that style of work — eventually with AI-assisted optimization.
PhySim uses the Godot ONNX AI Models Loader by mat490 to enable ML model inference inside Godot.
This GDExtension is built on top of ONNX Runtime and adds the ONNXLoader node to Godot, which is used as the inference backend for PhySim's planned AI-driven optimization layer.
To set it up:
- Download or clone the Godot-ONNX-AI-Models-Loaders repository
- Copy the
bin/folder from that repository into the root of your PhySim project directory - Godot will automatically detect the
.gdextensionfile and register theONNXLoadernode
⚠️ TheONNXLoadernode requires models that accept integers, floats, or strings as inputs. Ensure any.onnxmodel used with PhySim is compatible with this format.
-
AI-driven performance optimization Techniques under consideration include adaptive stepping, sleeping/activation heuristics, LOD-like simulation fidelity, constraint simplification, and scenario-specific approximations. The
ONNXLoaderGDExtension is already integrated as the inference layer for this work. -
Soft body physics Deformable bodies and constraints (approach TBD)
-
Procedural animation Layered procedural controllers that can complement physics-driven motion
Requirements:
- Godot Engine 4.x
- The
bin/folder from Godot-ONNX-AI-Models-Loaders placed in the project root (see Dependencies)
Run locally
Clone the repository:
git clone https://github.com/DbrittoRicky/PhySim.gitSet up the ONNX GDExtension (see Dependencies).
Open the project in Godot:
Import the folder containing project.godot
Press Play to run the simulation environment.
- Physics parameters are intended to be hot-editable for rapid iteration
- World "states" act like presets or profiles that modify external forces and resistance consistently across the scene, managed by the
EnvironmentManagerautoload - The
ONNXLoadernode is the inference entry point for future AI-driven features — models are loaded at startup and queried during simulation
Contributions are welcome, especially in these areas:
- UI/UX polish for editor-like workflows
- Stability improvements for ragdolls and constraints
- Performance profiling and benchmarking tooling
- Environment state system (more modes, better parameterization)
- Research prototypes for soft bodies and procedural animation
- ML model training and integration via the ONNX inference layer
By contributing to this project, you agree that your contributions will be licensed under the same GNU General Public License v3.0 that covers the project. Please ensure you have read and understood the license before submitting a pull request.
Milestone 1: Simulation scalability
- Stress-test scenes with many objects/ragdolls
- Add scalable simulation controls (quality tiers, selective activation)
Milestone 2: AI optimization layer
- Collect telemetry from simulation runs
- Train or tune
.onnxheuristics to reduce compute while maintaining motion quality - Add "optimize scene" suggestions (sleep thresholds, solver settings, etc.) powered by the integrated
ONNXLoader
Milestone 3: New physics domains
- Soft bodies prototype
- Procedural animation layer
This project uses the following third-party component:
- Godot ONNX AI Models Loader by mat490 — a GDExtension for loading and running ONNX models in Godot 4. Used here as the ML inference backend. (License: see upstream repository)
PhySim is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with this program. If not, see https://www.gnu.org/licenses/.
The full license text is available in the COPYING file included in this repository.
For a list of all copyright holders, see the AUTHORS file.
Note on third-party components: The
bin/folder from the Godot ONNX AI Models Loader GDExtension is a binary dependency and is governed by its own upstream license. PhySim's GPL license applies to PhySim's own source code only.

