Autopilot History

The Comstock Autopilot


Long multi-day solo Roziere balloon flights are difficult without a way to keep the balloon at a constant altitude while the pilot sleeps.  Before the development of a balloon autopilot, no such means existed.


In the mid-1990s, FAI Hall of Fame balloonist Bruce Comstock created such a balloon autopilot for Steve Fossett to use on his attempts to make the first solo circumnavigation of the earth.  Comstock had been co-director for equipment preparation, inflation and launch of Fossett on his Pacific-crossing flight in early 1995.  On this flight, Fossett had used a rudimentary analog autopilot Comstock had originally designed in the late 1970s but had never fully developed.  That analog device was unreliable and tended to drift off altitude.


After Fossett’s successful Pacific flight, Comstock proposed that Fossett contract with him to design and build for him a reliable balloon autopilot based on a digital computer that would keep his balloon at any desired altitude.  Fossett agreed.


The autopilot Comstock created measures atmospheric pressure and rate-of-climb and uses these to determine when to turn the burner on and off to maintain a constant flight altitude.  It works on any hot air balloon or any temperature controlled gas balloon like those used on all round-the-world flight attempts.  It can typically maintain the altitude of a balloon within 30 feet for low elevations and less than 300 feet in the thin air at 30,000 feet altitude -- usually much better than this.


Fossett used the resulting autopilot on all his round-the-world flight attempts, including his successful nonstop circumnavigation of the earth by balloon in 2002.  Fossett named the device Comstock developed for him the “Comstock Autopilot.”  This autopilot has been used on all solo round-the-world flight attempts, including Fossett’s successful round-the-world flight and Fedor Konyukhov’s later duplication of Fossett’s flight.  Many other solo world records have been set using this device, and it has been used on other epic long-distance flights including a flight from New Zealand to Australia and one from Norway to the north pole.


A total of ten of these autopilots were built.  Of these, two are at the bottom of the Coral Sea (Steve Fossett), one was lost in an unsuccessful transPacific hot air balloon flight attempt (Michio Kanda), two are in the U.S. National Air and Space Museum (Steve Fossett’s second two), and two are in the possession of balloonist Kevin Uliassi.  One is in the Netherlands in the possession of balloonist Remko Wigger; this unit was used by Fedor Konyukhov on his solo rtw flight and has been used in a hot air balloon as recently as 2020.  The tenth unit is believed to have been destroyed during David Hempleman-Adams’ icy/watery landing on his flight to the north pole and back.


The Comstock Autopilot cannot not currently be easily duplicated because the type of electric variometer it incorporated is no longer available, having gone out of production in 1988.  In this autopilot a Ball Model 400 variometer is used to sense rate of climb with a sealed chamber and calibrated leak on one side of a stainless-steel-foil diaphragm, the deflection of which is measured by two transducer coils.  This produces a very accurate, very-low-noise d. c. signal for rate of climb.  This signal is mathematically differentiated by the autopilot program to calculate the rate of vertical acceleration, which, along with the difference from the set altitude and vertical velocity, is used by the program to determine how much to heat.  Virtually all electronic rate of climb indicators available now use a pressure transducer to measure atmospheric pressure, and from this they calculate rate of climb.  This requires signal conditioning over a period of time to damp most of the inaccuracy from the pressure data, which also damps out most of the acceleration data.


Given the lack of a source for noise-free rate of climb, the challenge now in designing a usable balloon autopilot is to create an algorithm that circumvents the need for good acceleration data.