Robotics systems
ROS 2 integration, navigation, manipulation, control scaffolds, and sim-first validation.
Physical AI · Robotics · Edge Systems
I lead Physical AIfrom the benchto the field.
I’m Obinna Edeh. I’ve spent my career leading teams that get new hardware to perform at scale. At EmbodiedEdge, my Physical AI lab, I apply the same discipline to robots and Jetson hardware, and I do the bench work myself.
BACKGROUNDLeading deployments at scale
BUILDINGPhysical AI & robotics systems
OPERATINGEdge deployment & observability
ACTIVE SYSTEMS
⌁
// ABOUT ME
Operational AI for the real world.
I lead the work of getting Physical AI and edge systems to hold up outside the lab.
I’ve led engineering teams, field crews and vendors through nationwide deployments of hundreds of sites a year. One lesson held on every program: a spec sheet describes ideal conditions; the site shows what the hardware actually does. Physical AI has the same gap.
// SELECTED WORK
Systems, not slides.
Each project is built around reproducible artifacts, operational evidence, and a clear path to deployment.
01
Bench2Field
The benchmark said GO; the robot would drop frames. The detector alone meets its 33.3 ms deadline, but the full frame takes 47.8 ms on the rover’s Jetson Orin NX.
02
Physical AI on Jetson
A learned policy running on the robot’s own compute: GR00T through TensorRT at 9.8 Hz on Jetson AGX Thor, with a failed evaluation corrected and kept on the record.
03
Physical AI Safety Observability
A safety layer priced before anyone depends on it: a vision-language model on the device costs 4.35 s per frame at 66.6 W.
04
Urban Edge Vision Analytics
One edge box watching several street cameras and turning what it sees into events an operator can review. Runs end to end on live cameras; measurement is next.
05
Jetson Edge AI Security
Found a power problem in a runtime setting, not the model: cut a 54.1 W draw on Jetson AGX Thor to idle, with zero missed deadlines.
06
Field Commissioning Copilot
Built so a technician never acts on an unsourced number: answers commissioning questions from manufacturer documentation and says when no source covers the question.
// WHAT I BUILD
Edge AI
Jetson inference and edge runtime workflows measured with latency, telemetry, reliability, power, and thermal evidence.
Digital twins
Isaac Sim and OpenUSD workcells that connect robot assets to repeatable validation.
Agentic operations
Human-bounded copilots that reason over telemetry, logs, manuals, and runtime state.
// CURRENT PROOF POINTS
Built to be inspected.
AGX Thor2026-08-20 · 101.6 ms TRT · 9.8 Hz
Bench2FieldOrin NX 2026-10-02 · 26.9 ms model · 47.8 ms full frame
Isaac SimRTX 5090 · 2026-08-20 · 35/36 sim turns · 19.2 mm p50
Edge IDSAGX Thor 2026-09-08 · 0.0237 ms p95 · 54.1 W to 24.3 W