Research on guidance, navigation, and fail-safe decisions when flow, weather, or lost signals change the outcome.
These are research examples. The first three are cases of Counterfactual Fluid Dynamics, which couples flow, ionization chemistry, navigation, and guidance in one program; its project site carries the tutorial, validation records, and stated limits. The fourth concerns unmanned aircraft. All four run in simulation, and each states its model fidelity.
Re-entry guidance through radio blackout
Problem
Above roughly Mach 12 the shock layer ionizes; once electron density passes a critical value, the plasma sheath reflects every satellite navigation frequency. Guidance must commit to a bank angle while the vehicle dead-reckons without a fix.
Approach
One program couples the flow layer, finite-rate ionization chemistry, a 17-state error-state Kalman filter, and guidance. It detects blackout from the computed electron density, pauses at onset, and flies candidate bank angles from that shared state before committing to one.
Result
The run flies 17 branches from one onset. Scored 10 s after the fork, the committed 11.5° bank ends 2.07 m from a shared aim point, against 20.00 m at zero bank. Peak electron density falls within a ±0.7-decade band around NASA’s RAM-C II flight measurement. Navigation error peaks at 42.4 m before the link returns and settles after reacquisition. Because the shared past is computed once, the 17 branches cost 1,831 solver steps against 3,927 for rerunning each from the start, and the whole case runs within minutes on a regular laptop.
Limits
A point-mass three-degree-of-freedom trajectory in a fixed atmosphere, a two-dimensional flow layer, and uncalibrated chemistry with rates valid to about 8 km/s. Branch misses are scored against truth states a flight system cannot observe.
Navigation drift through blackout depends on the atmosphere actually flown: it sets the blackout window and moves the accelerometer bias away from its calibration point. A standard-day assumption misses both effects.
Approach
The run alternates the baseline descent into six weather conditions, flies each with eight receiver-noise draws concurrently, and reduces the 48 descents to a dispersion table.
Result
Polar-winter blackout drift averages 58.71 m against 41.60 m on a standard day, a 1.41-fold difference resolved at 5.7 standard deviations. Every descent reacquires with terminal navigation error under 1 m, and the full table runs within minutes on a regular laptop.
Limits
Temperature and density only, not the full DO-160 environmental list of humidity, winds, and statistical coverage. One flight anchor, so conditions away from it are extrapolated, and a labeled thermal coefficient for the accelerometer bias.
After blackout, a powered vehicle must decide when to light its landing engine. The plume interacts with the shock layer and changes the drag that guidance relies on, and the ignition margin depends on the navigation drift in the weather actually flown.
Approach
The program forks the plume-coupled state at five throttle settings mid-burn, sizes the ignition margin from the weather-dispersion table, and flies informed and uninformed guidance from the same baseline.
Result
At 20 % throttle the vehicle decelerates at 7.472 m/s2; coasting, it decelerates at 10.595 m/s2, because the plume collapses aerodynamic drag. Sizing the margin from the measured weather, 62.30 m against 47.52 m under standard-day belief, moves the burn start by 14.06 m at a cost of 7.28 kg of propellant; touchdown arrives at 1.81 m/s.
Limits
Inherits the re-entry case’s point-mass trajectory, two-dimensional flow layer, and fixed atmosphere. Plume drag rests on a cited empirical correlation whose fidelity the example flags as provisional.
A drone loses its satellite fix, then its command link, then a battery cell, over ground with water, trees, and a work crew. None of the documented fail-safe defaults weighs who stands near the landing site.
Approach
The fail-safe maps the ground from thermal and LiDAR readings, proposes landing sites, and submits each landing or ditching proposal to norms adapted from published emergency procedures for drone pilots, encoded in the Effect Ethos.
Result
In 1,000 randomized scenarios flown by three controllers, touchdowns within 10 m of a person fell from 35 with the textbook fail-safe to 1 (0.1 %; one-sided 95 % upper bound 0.47 %). The norms rank people above the aircraft, so the protocol loses more drones than the same controller without them: 36.3 % against 26.8 %. The campaign also caught six design flaws in earlier versions of the controller before deployment, one of them in the norm layer itself, each traced to the rule or stage that caused it.
Limits
A synthetic world with simulated sensors registered in the world frame. The norm-based controller was refined on the same scenario set that scores it.
The source code and verification records for every example are public.
The Center undertakes consulting, contract research, and training for
industry partners.