Tom Colvin

tcolvin at stanford
Department of Aeronautics and Astronautics
Durand Building
Stanford, CA 94305

Background

Career Interests

• Risk analysis and probabilistic methods.
• All sorts of POMDP / PSR problems. Autonomous anything!
• The creation and analysis of national public policy.
• Any technology that makes space flight cheaper / publicly accessible.
• Agriculture is fascinating; I'd kind of like to run a farm.

Research Foci

Aircraft Safety In The Presence Of A Rocket Launch

Traditional methods for safely integrating space launch and reentry traffic into the National Air Space use hazard areas (i.e. no-fly zones) that protect aircraft from larger areas, for longer times, than are necessary. I am proposing a new class of hazard area called compact envelopes, which are dynamic in time, contoured in space as a function of altitude, and whose boundary represents a quantifiable level of safety. The generation of these compact envelopes incorporates a probabilistic risk analysis of off-nominal vehicle operations and leverages expected improvements in air traffic management procedures from NextGen. One key assumption is that airborne aircraft can safely react to any off-nominal event if given a few minutes advance warning. The compact envelope method has the potential to offer hazard areas that significantly decrease disruption to the National Airspace System (NAS), as compared to traditional methods, while maintaining or even improving the level of safety to aircraft. Using NASA’s FACET software, we simulate the effects of compact envelopes on the NAS, versus traditional methods of creating hazard areas. The tool that we have developed for the probabilistic risk assessment and creation of compact envelopes is called the Stanford University Framework for Aircraft Risk Management (SU-FARM). SU-FARM is released as open source.

Decision Making Under Uncertainty

Rarely do launches occur at their nominal lift-off time. Current solutions simply block off the hazarded airspace for the entire launch window. The Compact Envelope concept also does not presently address this problem adequately because the no-fly zones are turned on exactly five minutes prior to the stated nominal launch time, but there is no mechanism for handling a delay that takes place during that five-minute window before launch.

I am exploring the usage of Partially Observable Markov Decision Processes (POMDP) and Predictive State Representations (PSR) for incorporating uncertain times of launch. It can be shown that some decision-making problems cannot be accurately controlled with a kth-order Markov method because limited-length histories do not provide a sufficient statistic for the uncertainty of the problem. Instead of keeping track of state-action histories, a PSR uses predictions of future state-action combinations to control the decision problem and can be shown to provide a sufficient statistic for the system dynamics.

I am also getting into Aerial Unmanned Vehicles as a platform for implementing and testing POMDP and PSR algorithms. Stay tuned!

Publications