Hey guys,
I’m building Exstrike. It’s a real-time safety filters (using Control Barrier Functions + computer vision) designed to prevent physical crashes when autonomous robots or AI navigation models mess up.
The goal is to build a drop in SDK for robotics,drones,and tactical ground vehicles (dual use/defense) that forces the machine to stop before hitting anything, even if the primary AI panics or loses GPS
The problem in physical AI right now is massive: navigation models fail, sensors blind out, and multi-thousand dollar hardware gets destroyed in a split second. Most systems rely on soft software limits that fail under pressure. Exstrike solves this at the lowest level by decoupling perception safety math, and hardware override into a dedicated 3-model architecture
Hercule (LLM to Spatial Perception Engine):
Hercule bridges high-level AI models and Vision-LLMs (Claude, Gemini, GPT-4o) with real-world physical precision and hallucinate spatial boundaries. Hercule ingests raw camera feeds at high frame rates to perform continuous spatial mapping, velocity estimation, and obstalce LLM agents precise physical grounding in real time without cloud latency or heavy compute overhead.
Soteria (Deterministic Control Barrier Engine):
Soteria is the mathematical brain that enforces physical safety boundaries. it takes the spatial vectors from Hercule and continuoulsy solves real-time Quadratic Programs (QP) using Control Barrier Functions (CBFs). Rather than guessing what to do, Soteria calculates the exact physical envelope of the vehicle. As long as the primary navigation AI stays set, Soteria remains passive; the moment a trajectory breaches the safe set, Soteria dynamically modifies the control commands.
Armageddon (Hard Stop & Failsafe Execution Unit):
Armageddon is the zero-latency emergency hardware override. Sitting directly above the motor controlllers and actuators, it operates as a fail-closed circuit breaker. If Hercule loses visual tracking or Soteria calculates an unavoidable impact trajectory within a critical time to collision threshold, Armageddon executes a total override cutting drive signals or engaging maximum regenerative braking instantly. It completly bypasses the main AI and flight controller, ensuring zero delay and zero guesswork when a crash is imminent.
This is going to be huge because every single autonomous unit deployed in high-risk, real-world environments needs a bulletproof safety fallback. We are building the industry standard for physical crash prevention from the ground up.
I’ve arleady validated the core logic and real-time obstacle tracking on a physical test bench.
Now, I need a technical co-founder to help scale this raw prototype into a production grade, ultra low latency c++/ros 2 framework
Who i’m looking for
Deep hands on skills in c++, ros 2, or computer vision (OpenCV,CUDA).
low-level systems builders who care about hard real-time execution and raw perfomance.
An ambitious technical partner / co founder to join the initial build phase
We’re building the foundational layer fore the next decade of physical AI. If you can write low latency c++ and want to build high-stakes defense systems, DM me. Let’s make history.