"The complexity of life does not require complex hardware. It requires efficient flow."
HFPS is a high-performance, data-oriented AI engine designed to simulate complex emergent behaviors through fluidic pressure propagation. Unlike traditional AI approaches, HFPS models agents as physiological entities that respond to environmental pulses, allowing for the simulation of massive crowds and organic ecosystems with
HFPS operates on a fundamental belief: Behavior is a sub-product of anatomy and physics.
Instead of treating AI as a black box of probabilistic weights, HFPS treats it as a biological system. We don't "train" a model to decide what to do; we build an anatomy with pressure channels and let the laws of physics dictate the emergent response. This creates a type of intelligence that is fundamentally deterministic, auditable, and extremely fast.
- Determinism: If you know the pressure levels and the stimulus, you know the result. No "hallucinations".
- Authenticity: Behavior feels organic because it's governed by homeostasis (recovery and decay).
- Efficiency: Arithmetic over matrices. It runs on everything from microcontrollers to high-end servers.
Every agent possesses 4 floating-point channels representing their internal state. Events in the world apply pressure to these channels:
- Vitality: Physical health or energy.
- Security: Fear/Safety level. (Negative = Panic).
- Dominance: Aggression or social hierarchy.
- Engagement: Social interest and curiosity.
HFPS utilizes a Static Linked-List Grid for spatial partitioning:
- Fixed Memory: No heap allocations during the simulation loop.
- Cache Locality: Optimized for L1/L2 cache hits (Structure of Arrays).
- Performance: Check thousands of neighbors in microseconds.
Agents exhibit real-time biological adaptation:
- Habituation: Agents develop "calluses" to repeated stimuli. Sustained pressure reduces sensitivity.
- Recovery: When stimulus ceases, homeostasis returns sensitivity to baseline.
- DNA Drift: Prolonged exposure to stress permanently mutates the agent's genetic modifiers (Epigenetics), creating unique personalities over time.
- Rust Toolchain (latest stable)
git clone https://github.com/creator16/hfps.git
cd hfpsThe current build (main.rs) is a Visual Kernel Visualization. It demonstrates the internal state of a single agent responding to sensor inputs.
cargo run --releaseControls (Sensor Simulation):
[1]Simulate Danger: Triggers a "Predator/Attack" signal (Panic).[2]Simulate Food: Triggers a "Resource/Grass" signal (Feeding).[3]Simulate Social: Triggers a "Interaction" signal (Engagement).
Observe how the Brain State changes and the bars fluctuate based on pressure and decay (homeostasis).
You can define the "species" of your agents using simple TOML files in data/.
Example: data/ovelha.toml
name = "Sheep"
[sensitivity]
Security = 2.0 # Highly paranoid (2x multiplier)
Vitality = 1.0
[decay_rates]
Security = 0.5 # Calms down slowly
[thresholds]
channel = "Security"
value = -50.0
flag = 4 # FLAG_FLEEINGHFPS is an evolving research project. We strongly encourage forks, modifications, and pull requests.
- Serialization: The
.dnaformat is simple JSON but likely needs a more robust binary format for large swarms. - Multithreading: While
SoAis ready for SIMD, we haven't implementedRayonfor parallel agent updates yet. - Bindings: Currently only runs in Rust. We need C-bindings to export
hfps.dllfor Unity/Unreal/Godot.
- Fork the repo.
- Create a branch for your feature (
git checkout -b feature/swarm-logic). - Implement your changes (e.g., adding a
Reproductionsystem). - Submit a Pull Request.
"We are building the anatomy, not the script."
This project is open-source under the MIT License.
Created with ❤️ for the Open Source Community.