Work

Six systems, one idea: autonomy that is verifiable, replayable, and fail-closed — from robots in simulation to agents under review.

Sim-to-Real Control Systems

A preregistered 40-run campaign in NVIDIA Isaac Sim: closed-loop DSP filtering on a Franka feedback path. All three headline paired metrics improved in 20 of 20 seeds — and still 0 of 40 runs passed the full frozen gauntlet. The thresholds were frozen before the first run and were not relaxed afterward. Simulation only.

• Isaac Sim
• ROS 2
• DSP
• Preregistered

Agentic Systems Verifier

An agentic workflow that verifies systems-engineering specs: it parses the requirement, generates and executes tests, and returns RAG-grounded, traceable evidence for each check — so a reviewer sees why a spec passes, not just that it did.

• Agents
• RAG
• FastAPI
• Verification

Fieldheld Recorder

Turns autonomous coding-style runs into replayable, reviewable, rollbackable evidence bundles. Autonomy you can actually audit after the fact.

• Governance
• Evidence
• Autonomy

Articulated Robot Manipulation

Franka Panda 7-DOF pick-and-place in NVIDIA Isaac Sim, driven by Lula inverse kinematics on PhysX rigid-body physics. Physics-correct grasping before anything touches the real world.

• Isaac Sim
• Lula IK
• PhysX

Isaac Sim → ROS 2 Bridge

An OmniGraph ROS 2 bridge in Isaac Sim publishing joint state across Python and C++ ROS 2 nodes — the plumbing that lets a simulated controller talk to real-robot interfaces.

• Isaac Sim
• ROS 2
• OmniGraph

Franka Joint Control from Python

Commanding a Franka arm's joints directly from Python in NVIDIA Isaac Sim — the smallest honest starting point for everything above it.

• Isaac Sim
• Python
• Control