LiDAR pts
0
AMDVS is a self-funded, research-grade autonomous platform. This page is its live console: the 3D field view and the tactical map run the same simulation — dispatch it, stress it, stop it.
Every control below drives the live simulation — the same core mission loop AMDVS is built around: GPS dispatch, hazard reporting en route, dynamic rerouting, and emergency stop with alert.
Reflex → perception → navigation → strategic reasoning → human interface. Deterministic safety at the bottom; the LLM advises and never touches the real-time control loop.
Verified 30 FPS steady-state on Jetson Orin Nano — CSI capture via jetson-utils, IMX219 160° on CAM0.
Independent hardware watchdog drives a fail-safe relay — silence means motors off. Neither safety layer depends on the Jetson.
Command-to-relay-open latency, measured on every trigger from the command center above.
MQ-series readings streamed with hazard thresholds — spikes when a hazard is reported en route.
3S LiPo → motor rail (SA8339 VM, bulk cap) and compute rail (5V/5A BEC → Jetson barrel jack). One pack, isolated noise.
Ground autonomy first, LLM integration second, drone cooperation third.
ROS 2, OpenCV, YOLOv8, TensorRT, camera perception, lane following, Nav2, sensor fusion, and safety watchdogs on a tracked test mule.
High-level reasoning layer that reads system state, summarizes events, explains failures, and helps choose safe next actions — strictly advisory.
Overhead reconnaissance, AprilTag / ArUco precision landing on the rover dock, and recharge on deck.
ROS 2, Nav2, sensor fusion, EKF, watchdog design, LiDAR, GNSS, IMU, Jetson hardware, and embedded reliability.
YOLOv8, TensorRT FP16, OpenCV, PyTorch on Jetson, fiducials, real-time inference, and perception pipelines.
Python, C/C++ fundamentals, asyncio, Linux, Git, GitHub, Flask, WebSockets, algorithms, and data structures.
Builder before anything else. I learn by making, breaking, measuring, and rebuilding.
Safety is part of intelligence. A robot is not smart if it cannot fail safely.
Every project is a brick. Reusable code, documented experiments, and systems I can explain end to end.
Self-funded means disciplined. No waste, no fake complexity, no black boxes I cannot debug.
If you are building autonomous systems, perception pipelines, robotics infrastructure, or embedded AI, this is the work I want to be around.