Aerospace and Space Exploration
DART (2005 autonomous rendezvous demonstration)
NASA's Demonstration of Autonomous Rendezvous Technology, DART, launched in April 2005 to prove a small spacecraft could find and maneuver near another satellite entirely on its own, without ground control. The plan called for DART to approach the target satellite MUBLCOM, hold at close range, and retreat. A biased GPS velocity reading triggered repeated navigation resets that burned through DART's maneuvering fuel, and a separate targeting error meant its collision-avoidance system never had an accurate read on distance. DART struck MUBLCOM at low speed, then ran out of fuel and began its planned retirement sequence without ever registering that a collision had occurred. This is not the 2022 Double Asteroid Redirection Test, which deliberately hit an asteroid and succeeded.
- Company
- NASA
- Started
- 2005-04-15
- Ended
- 2005-04-15
- Collapse speed
- Sudden
- Preventability
- High
- Lesson transfer
- Industry-wide
- Last reviewed
- 2026-08-17
Narrative
The story
The ambition
DART, the Demonstration of Autonomous Rendezvous Technology, launched on April 15, 2005, aboard a Pegasus XL rocket from Vandenberg Air Force Base. Built by Orbital Sciences Corporation under a NASA Research Announcement contract, the roughly 360-kilogram spacecraft was a technology demonstrator with a narrow, ambitious goal, prove that a spacecraft could locate, approach, and maneuver around another satellite entirely by itself, using onboard GPS and an optical sensor, with no commands from the ground. NASA framed the capability as a building block for future automated docking, servicing, and space-station resupply missions, and after the 2001 loss of the Orbital Space Plane program that had originally sponsored it, DART became NASA's first flight demonstration under the newly announced Vision for Space Exploration.
The rise
The target was MUBLCOM, a small Orbital Sciences communications satellite launched in 1999 for the Defense Advanced Research Projects Agency and left in orbit after its original mission ended. DART's flight plan had four phases: launch and early orbit, a rendezvous phase using GPS to close the distance, a proximity-operations phase in which DART would fly a scripted sequence of close-range maneuvers past MUBLCOM using its Advanced Video Guidance Sensor, and finally a departure and retirement burn. The mission covered its first eight hours almost exactly as planned, DART reached its second staging orbit roughly 40 kilometers behind and 7.5 kilometers below MUBLCOM, even though ground controllers had already started noticing something off in the navigation telemetry.
The cracks
DART's onboard navigation software periodically compared its estimated position and speed against measurements from its primary GPS receiver, and when the two diverged too far it executed a computational reset that discarded the estimate and restarted from the raw GPS reading. That primary GPS receiver had a factory defect that made its velocity measurement consistently off by about 0.6 meters per second, a bias the navigation software's design was supposed to tolerate but, due to an unfixed known bug and a late, inadequately tested change to the software's confidence weighting, could not. The result was a reset roughly every three minutes for the whole mission, each one triggering a burst of corrective thruster firings that drained DART's pressurized-nitrogen fuel supply far faster than planned. Separately, as DART moved toward MUBLCOM its guidance logic was supposed to fly through a small imaginary waypoint, a sphere just over six meters across positioned 200 meters behind the target, in order to switch the optical sensor into full range-measuring mode. DART missed that waypoint by less than two meters, close enough that the sensor kept supplying only bearing data, never an accurate distance reading.
The collapse
Because it never received the range data it needed, DART's collision-avoidance logic, which depended on the same flawed navigation picture as its main guidance, could not tell how close it actually was to MUBLCOM. At the moment of impact DART's own system believed it was 130 meters from MUBLCOM and retreating; in reality it was closing at 1.5 meters per second. Fewer than 11 hours into what was meant to be a 24-hour mission, DART struck MUBLCOM at low speed. Neither spacecraft was destroyed, the impact nudged MUBLCOM into a slightly higher orbit and MUBLCOM automatically reset and recovered, but DART had no way to detect that a collision had happened. Minutes later, DART's fuel-remaining logic, which the investigation later found had itself overestimated fuel consumption and prematurely declared the tank empty with roughly 30 percent of the propellant still aboard, triggered the spacecraft's pre-programmed departure and retirement sequence, and DART carried it out as designed, unaware it had already hit its target.
The aftermath
Because DART failed its main objectives, NASA declared a Type A mishap, its most serious classification, and convened a Mishap Investigation Board chaired by Scott Croomes of Marshall Space Flight Center. With no physical wreckage recoverable, since both spacecraft remained in orbit, the board relied on telemetry, hardware testing, and simulation to reconstruct two separate timelines: one for the fuel depletion and premature retirement, another for the collision itself. Of 27 defined mission objectives, DART fully or partially met 11; none of the 14 objectives tied to the proximity-operations phase, the actual technology the mission existed to prove, were met. The board's full 70-page report was never released publicly because it contained information restricted under arms-export regulations, but NASA issued a public summary of its findings and recommendations in 2006. Beyond the immediate navigation bug, the board cited a chain of process failures: an undocumented late change to the software's confidence weighting, inadequate testing forced by schedule pressure ahead of launch, a systems-engineering process that failed to specify collision-avoidance requirements precisely enough, and a contractor design and review process that did not catch the undersized targeting waypoint.
The lessons
DART's failure was not a single dramatic malfunction but a chain of small, individually survivable errors, a biased sensor, an unfixed known software bug, a late and undertested weighting change, a waypoint sized too small by a matter of meters, that combined into a spacecraft whose collision avoidance system depended entirely on the same faulty picture of reality as the guidance system it was supposed to be checking. The Mishap Investigation Board's central recommendation reflects that: future spacecraft performing autonomous rendezvous should have collision-avoidance sensing that is genuinely independent of the primary navigation system, not a second consumer of the same potentially wrong data. The board also flagged a structural tension common to high-risk, low-budget technology demonstrations, DART was procured with broad requirements and wide contractor discretion because it started as a cheap experimental flight, but as it grew into a higher-profile milestone for NASA's exploration program, the tolerance for failure shrank faster than the rigor of its review process did.
Causal timeline
Failure Anatomy
- 2005-04-15
DART launches
DART launched on a Pegasus XL rocket from Vandenberg Air Force Base on April 15, 2005, beginning a planned 24-hour autonomous rendezvous demonstration with the target satellite MUBLCOM. [1]
- 2005-04-15
Early phases succeed despite navigation anomalies
DART completed its launch, early-orbit, and rendezvous phases largely as planned, reaching a staging orbit near MUBLCOM, even as ground controllers noticed irregularities in the navigation telemetry that could not be corrected from the ground. [2]
Information failure - 2005-04-15
GPS bias drives a fuel-draining reset loop
A biased GPS velocity reading, reintroduced into the navigation system on every reset due to an unfixed software bug and an under-tested late weighting change, caused DART's navigation estimate to diverge and reset roughly every three minutes, each reset triggering thruster firings that consumed fuel far faster than planned. [3] [4]
Technical failureInformation failure - 2005-04-15
DART collides with MUBLCOM
Having missed the targeting waypoint needed for accurate range data, DART's collision-avoidance system could not perceive its true closing distance and struck MUBLCOM at about 1.5 meters per second, nudging it into a slightly higher orbit; the impact caused no lasting damage to either spacecraft. [5] [7]
Poor executionTechnical failure - 2005-04-15
DART retires without knowing it had collided
Minutes after the collision, DART's onboard logic, which had itself overestimated fuel usage and prematurely declared the tank near empty, triggered the spacecraft's pre-programmed departure and retirement sequence, ending the mission having met only 11 of 27 objectives and none of the 14 tied to proximity operations. [9] [10]
Technical failure - 2006-05
Mishap board publishes findings
NASA's Mishap Investigation Board, unable to examine physical wreckage since both spacecraft stayed in orbit, reconstructed the failure from telemetry and simulation and issued a public summary of its findings and recommendations in 2006, after the full report was withheld as export-controlled. [11]
Information failure
Structured analysis
What Went Wrong
Root causes
Biased GPS velocity reading triggered runaway navigation resets. DART's primary GPS receiver had a factory defect that made its velocity measurement consistently biased by about 0.6 meters per second. A known but unfixed software bug reintroduced that biased reading into the navigation system every time it reset, and a late, inadequately tested change to the software's confidence weighting meant the calculations could never converge, producing a reset roughly every three minutes and a burst of corrective thruster firings each time. [3] [4]
Missed targeting waypoint left collision avoidance blind to distance. DART's guidance logic needed to fly through a roughly 6.3-meter-radius waypoint 200 meters behind MUBLCOM to switch its optical sensor into full range-measuring mode. It missed that waypoint by less than two meters, so the sensor never supplied accurate distance data, and the collision-avoidance system, which relied on the same navigation picture, could not judge how close DART actually was to MUBLCOM. [5] [6]
Contributing factors
Undocumented late software change. A change to the navigation reset logic, made late in development to fix a units-conversion error in a simulation model, introduced the use of the GPS receiver's raw velocity output in a new way that most of the DART team was unaware of, so the receiver's known velocity bias was never corrected in the flight software or represented in its preflight simulation model. [12]
Schedule pressure cut short testing of the change. The Mishap Investigation Board found that because the late change to the navigation software's confidence weighting happened close to the planned launch date, it was never adequately tested on flight hardware and software, and pressure to hold the launch schedule was the root cause of that decision. [13]
High-risk, low-budget procurement with limited government oversight. DART was procured as a high-risk, low-budget technology demonstration under a NASA Research Announcement, which set broad requirements and left most detailed design decisions to contractor Orbital Sciences Corporation, including reuse of Pegasus launch-vehicle software architecture that the board later found unsuited to adaptive, autonomous in-space operations. [14]
Immediate trigger
Collision avoidance could not see the closing distance to MUBLCOM. At the moment of impact, DART's navigation system calculated that it was 130 meters from MUBLCOM and retreating at 0.3 meters per second, when it was actually closing at 1.5 meters per second, so its collision-avoidance logic took no effective action and DART struck the target satellite. [7]
Visible symptoms
Excessive fuel consumption during proximity operations. Ground controllers observed DART using significantly more maneuvering fuel than expected as it transitioned into proximity operations, and it became clear the mission would likely end early from exhausted fuel reserves well before the collision was understood to have occurred. [8]
Premature retirement without registering the collision. Roughly 11 hours into a planned 24-hour mission, DART's onboard logic declared its fuel depleted and executed its pre-programmed departure and retirement sequence, unaware that it had collided with MUBLCOM three minutes and 49 seconds earlier. [9]
Warning signs
Ground controllers noticed navigation anomalies early. During the mission's launch, early-orbit, and rendezvous phases, ground operations personnel already noticed irregularities with DART's navigation system, though the spacecraft had no ability to receive uplinked commands and no action could be taken to correct the situation in flight. [2]
Affected groups
Keep reading
Related failures
Evidence
Claims & sources
Every numbered marker in the analysis links to the claim it rests on, and each claim to its sources.
- [1]
DART launched on April 15, 2005, aboard a Pegasus XL rocket from Vandenberg Air Force Base, beginning a planned roughly 24-hour mission to autonomously rendezvous with the satellite MUBLCOM.
- [2]
DART completed its launch, early-orbit, and rendezvous phases largely as planned, even though ground operations personnel noticed irregularities with the navigation system during those early phases, and DART had no ability to receive uplinked commands so the ground crew could not intervene.
- [3]
DART's primary GPS receiver had a factory defect that made its velocity measurement consistently biased by about 0.6 meters per second from what it should have been.
- [4]
A known but unfixed software bug reintroduced the biased GPS velocity reading into DART's position and speed estimate every time the navigation software executed a reset, and an inadequately tested late change to the software's confidence weighting meant the estimate could not converge, producing a diverging-calculation and reset cycle roughly every three minutes throughout the mission, each one triggering excessive corrective thruster firings.
- [5]
DART's guidance logic required the spacecraft to fly into a roughly 6.3-meter-radius spherical waypoint positioned 200 meters behind MUBLCOM in order to switch its optical Advanced Video Guidance Sensor into full range-measuring capability; telemetry analysis showed DART missed this waypoint by less than 2 meters.
- [6]
Because DART missed the targeting waypoint, its sensor never supplied accurate range data, and its collision-avoidance system, which depended on the same navigation data source as the main guidance system, could not accurately determine the distance to MUBLCOM.
- [7]
At the moment of collision, DART's navigation system calculated that it was 130 meters from MUBLCOM and retreating at 0.3 meters per second, when it was actually closing on MUBLCOM at 1.5 meters per second, and DART struck MUBLCOM at that low relative speed.
- [8]
Ground controllers observed DART consuming significantly more maneuvering fuel than expected as it moved into proximity operations, and it became apparent the mission would likely end early due to exhausted fuel reserves.
- [9]
DART's onboard logic declared its fuel supply depleted, overestimating actual fuel usage by roughly 30 percent, and executed its pre-programmed departure and retirement sequence about 11 hours into the mission, roughly 3 minutes and 49 seconds after DART had already collided with MUBLCOM without registering that the collision had occurred.
- [10]
Of 27 defined mission objectives, DART fully or partially met 11, and none of the 14 objectives tied to the proximity-operations phase were met.
- [11]
Because DART failed to achieve its main objectives, NASA declared a Type A mishap and convened a Mishap Investigation Board, which reconstructed the failure from telemetry, hardware testing, and simulation since no physical wreckage was recoverable, and NASA published a summary of the board's findings and recommendations in 2006 after withholding the full report as export-controlled under ITAR.
- [12]
A late change to the navigation system's reset logic introduced use of the primary GPS receiver's raw velocity output in a way most of the DART team was unaware of, so the receiver's known velocity bias was never corrected in the flight software or incorporated into its preflight simulation model, and the change was not adequately documented.
Moderate Reported explanation Overview of the DART Mishap Investigation Results (For Public Release) - [13]
The Mishap Investigation Board found that a late change to the navigation software's confidence weighting was never adequately tested on flight hardware and software because it occurred close to the planned launch date, and that pressure to maintain the launch schedule was the root cause of the decision to forgo that testing.
Moderate Reported explanation Overview of the DART Mishap Investigation Results (For Public Release) - [14]
DART was procured by NASA as a high-risk, low-budget technology demonstration under a NASA Research Announcement that set broad requirements and left detailed design decisions to contractor Orbital Sciences Corporation, which carried over software architecture from its Pegasus launch vehicle that the board later found unsuited to adaptive autonomous in-space operations.
Moderate Reported explanation Overview of the DART Mishap Investigation Results (For Public Release)
Sources
NASA Releases Report on Loss of Contact with DART/MUBLCOM (2006 mishap-report release)
NASA Headquarters · 2006-05
DART: Rendezvous
eoPortal
DART (satellite)
Wikipedia