MIT Robot Locomotion Group - Agile Flight Project



click autonomous for prop-hang video

click for autonomous hover transition


  Rick Cory - Ph.D. Candidate, EECS, email: recory [at] mit [dot] edu
  Zack Jackowski - Ornithopter Design, UROP MIT MechE
  Russ Tedrake - Assistant Professor, EECS, email: russt [at] mit [dot] edu

Former Staff/Students

  Stephen Proulx - Ornithopter Design Staff, B.S. Northeastern Univ.
  Guilherme Cavalcanti - Ornithopter Design, Undergraduate Intern, Olin College



The dynamics of flight are well understood for flight envelopes that constrain themselves to portions of the state space that lend themselves to piece-wise linear models. Through use of wind tunnel measurements and gain scheduling techniques, we can develop autopilots that perform reasonably well around trimmed flight conditions. However, an experienced human pilot by far outperforms conventional autopilots in terms of maneuverability and agility, primarily because of the ability of the human pilot to compensate for non-linear dynamic effects that dictate the response of the aircraft. Similarly, the highly unsteady flows that dictate the dynamics of flapping-wing flight make it difficult for our robots to match the agility of biological flappers. We believe that proper sensing, intelligent mechanical design, and exploitation of natural dynamics will play a key role in the solution to the control problem for our flying robots. Furthermore, we believe that the control problem can be approached independently of the dynamic modeling problem through established methods in model-free machine learning and optimal control.


  We're currently working on a two-meter wingspan autonomous robotic bird named Phoenix (it actually does rise from the ashes...and no crash has proven to be fatal yet!). Phoenix is designed to provide an ideal platform for investigating motor control of maneuverable flapping-wing flight in an outdoor setting, with the focus of the design being the optimization of load carrying capacity for onboard computation and sensing, robustness to crashes, and ease of parts replacement. As the video below shows, we have accomplished steady-level autonomous flight carrying a 400g instrumentation payload.

The Phoenix Robot


  We've also been developing an ornithopter (robotic bird) with a differential flapping amplitude transmission. This mechanism will allow for quick turning and agile maneuvering. This design has two degrees of freedom in the tail and flapping frequency is provided through a geared down brushless dc motor. In parallel we are working on an embedded computer solution for this desigin, which will allow all sensing, learning, and control to be done on board.


Flight In Killian Court
Differential Amplitude Transmission
Training Wheels



  We are developing adaptive autopilots that will be capable of performing aggressive aerial maneuvers such as loops, prop-hangs, and knife edges. Our hope is to close the computational loop on board our fixed wing aircraft before moving on to the more complex task of applying closed loop control to our flapping-wing robot birds.