FLIGHT TEST AND SIMULATION RESULTS FOR AN AUTONOMOUS AEROBATIC HELICOPTER V. Gavrilets', I. Martino?, B. Mettler3and E. Feron', Laboratory for Information and Decision Systems, Department of Aeronautics and Astronautics, Mfl,Cambridge, u.4 quaternion attitude representation. A magnetic compass with reset circuit was added to provide small heading corrections. Introduction In this paper we shall present the flight test and simulation results for an X-Cell.60 helicopter, which performs aerobatic maneuvering under exclusive computer control. We have previously demonstrated in flight an autonomous axial roll maneuver [ 13. Our next goal is to demonstrate operationally useful, and more challenging maneuvers, namely split-S and Immelman. Both maneuvers provide a rapid way to change direction of flight by trading kinetic and potential energy. Previous flight test results showed that the engine performance was marginal at high collective settings. In order to provide more power to the helicopter with the avionics payload a .90-size engine (50% larger displacement than the old .60 size) was installed, along with 700 mm instead of 690 mm blades. During the flight tests we learned that a tight yaw rate control by the tail rotor is necessary to counteract large and fast changes in the collective during some of the maneuvers. We have stiffened the avionics box suspension system to avoid coupling of the box and helicopter yawing dynamics, and replaced an older Futaba S9402 analog servo on the tail rotor with a digital Futaba S9450 to increase actuator bandwidth. The X-Cell.60 is a model helicopter with a 5 ft rotor diameter, designed for competition aerobatics. Empty weight is 10 lbs, the avionics box with a custom landing gear and a suspension system weigh 7 lbs. The avionics system features an electronic governor, an inertial measurement unit, a single GPS receiver, a pressure altimeter, and a magnetic compass. The system described earlier [Z] has been modified to enable all-attitude flight. Previously a low-cost GPS receiver was used with one-second latency in position and velocity updates, which made it difficult to incorporate Kalman filter based estimator. A set of complementary filters was used to derive Euler attitude angles and the velocity vector. This state estimator proved adequate for aggressive trim trajectory tracking and for the autonomous axial roll [l]. However, the singularities in kinematic equations at zenith and nadir made split-S and Immelman maneuven impossible. We have integrated into the avionics package the G12 receiver from Ashtech, which features 10 Hz update rate and 50-millisecond latency. An extended Kalman filter with bias states for accelerometers and gyros was implemented to provide the all attitude navigation solution using the ~~ ~ An essential part of the project was the development of an adequate nonlinear mathematical model describing a miniature helicopter in aerobatic flight [3]. The X-Cell helicopter features a stiff hub, and due to its small size the dynamics are dominated by the main rotor forces and moments. Despite the stiff hub, the on-axes responses in the hub are an order of magnitude higher than the offaxis responses, which makes modeling task easier than for full scale helicopters and some small-scale helicopters, like Yamaha R-50 [7]. The originally developed X-Cell model [3] was refined. Instead of linear stability derivatives describing fuselage forces we have used effective drag areas [5]. An engine/governor model was added to account for the rotor speed variations during maneuvers, and for ~~ W.D. candidate,gavrick@rnit.edu M.S.candidate, martinos@mit.edu Post-doctoral associate, bmettler@mit.edu Associate Professor, feron@mit.edu . 0-7803-7367-7/02/$17.00 0 2002 IEEE 8.C.3-1 linearized lateral-directional dynamics of the helicopter is well described by a 5* order model, with the state vector containing side velocity, lateral flapping angle of the main rotor, roll rate, yaw rate and bank angle [ 1,3,9]. Similarly, linearized longitudinal-vertical dynamics is described by a 5* order model with the states being body axis forward and vertical velocities, longitudinal flapping angle, pitch rate and pitch angle. All states except the flapping angles can be accurately measured or estimated. The flapping angles participate in the lightly damped fuselage-rotor modes, which are characteristic of small-scale helicopters with stabilizer bars [lo]. We have applied notch filters on cyclic inputs to suppress the modes, and used the rigid body approximations to the notched plants for the LQ design [l, 91. The controller gains were scheduled with the forward speed. The lateral directional controller gains were switched discretely with the hysteresis logic; this prevented undesirable rolling and yawing during fast transitions between hover and forward flight. The bumpless transition between the gain sets is ensured by a proper initialization of the integrator states to keep surface commands continuous. the delay in the transmission of the main rotor torque to the fuselage through the engine. The engine was modeled as a first order lag, and the governor as a gain. The resulting transfer function from the rotor speed command to the rotor speed is well-damped second order; its damping ratio and natural frequency were derived from timefrequency decomposition of the engine sound recordings during commanded rotor speed transitions [2]. The model of the suspension system for the avionics box was added to account for its lightly damped modes and a possible effect on the gain margin. The next section summarizes our approach to implementing aerobatic maneuvers autonomously. Subsequent sections of the paper will describe implementation of the axial roll maneuver, and simulation studies for the Immelman and split-S maneuvers. An Approach To Aggressive Maneuvering Control laws for unmanned helicopters can be broken down into two categories: trim trajectory tracking controllers, and those for implementing extremely agile, or aerobatic maneuvers. To implement an autonomous aerobatic maneuver we used both kinds in different phases of the flight. A velocityheading rate/altitude tracking controller was used for the entry and recovery phases (trim trajectory controller), while an angular rate tracking controller with predefined commanded trajectories was used for the maneuver itself. Other approaches to tracking trim trajectories and non-aggressive transitions between trim trajectories have been suggested in the literature on small-scale helicopters. La Civita et al. have tested H-infinity loop shaping controller on a Yamaha R50 helicopter. In this approach, frequency weights are used to perform loop shaping and alleviate the lightly damped fiselage-rotor flapping mode. Johnson and Kannan [ 121 implemented neural network based adaptive controller on an improved version of R-50,the Yamaha RMAX helicopter. The Yamaha helicopters are an order of magnitude heavier than the X-Cell, feature a relatively flexible hub and were not designed for aerobatics. Trim Trajectoy Controller The trim trajectory controller has to feature both high bandwidth and good stability to guarantee fast recovery from an aerobatic maneuver. We have developed a state-feedback linear quadratic (LQ) controller [8] with integrators appended on tracking variables, which proved adequate for the task [l]. We used an analytical linearization of the full nonlinear model at different forward speed settings to obtain state transition and input matrices for the LQ design. For X-Cell the coupling between the lateral-directional and the longitudinal-vertical dynamics is negligible, except for the variations of the main rotor torque due to collective changes. The Human-CenteredApproach ToAutonomous Aerobatics In the beginning of the program we have performed a number of manual flights with an instrumented X-Cell60 helicopter [4, 131. The tests showed that the maneuvers performed by an expert pilot are repeatable; the pilot actions can be well approximated by piece-wise linear or piece-wise constant commands; and switching between the 8.C.3-2 commands occurs upon reaching specific attitude angles. Simulation studies [4] have also shown that some degree of continuous feedback is desirable for repeatable autonomous execution of maneuvers. We have designed decoupled body-axis angular rate tracking controllers. All controllers are constant gain proportional-integral with crossover frequencies above 5 radsec for all speeds from hover to 15 d s e c . We have applied notch filters on cyclics to suppress the lightly damped fbselagerotor flapping modes [9]. The pilot has flown a number of maneuvers (axial roll, loop, 180 and 540 hammerheads, split S and Immelman) with the rate tracking controllers and an open loop collective command. Recorded inputs and state trajectories were analyzed to create angular rate reference trajectories and execution strategies for various maneuvers. The attitude angles for the switching of the rate commands were determined. We have also observed that the collective command used by the pilot during the maneuvers is roughly proportional to cosine of the angle between the local vertical and the helicopter Z-axis. This allows the pilot to keep the thrust pointing up in all attitudes to reduce altitude loss, and close to zero when the rotor disk is close to being perpendicular to the ground (which reduces unwanted deviations from the flight path). setpoint for the altitude hold, and resumes tracking the speed setpoint. Experimental Axial Roll Maneuver During the axial roll maneuver the pitch and yaw rate commands are zero. The collective is varied proportionally to cosine of the bank angle. The roll rate trajectory, given in Figure 1, consists of four phases. - Roll rate command, deglsec 1.5 2 2.5 3 time, sec Figure 1. Roll Rate Reference Trajectory First, a linear ramp up to the maximum commanded rate is performed in a given time (0.3 seconds). Next, the roll rate command is kept at the maximum until an integrated roll rate (pseudoroll angle) reaches a specific value, determined from the pilot’s execution of the maneuvers with the rate tracking control augmentation system (3 10 degrees). At this point the roll rate command is linearly ramped down in a specific time (0.2 seconds). The last phase is a predefined settling period (0.2 seconds) with the zero roll rate command to arrest the roll rate and achieve a 360 degree revolution. The maneuver is exited upon reaching a 360 degree roll as determined by the integrated roll rate, or upon completion of the sequence described above. A safe set of states for the maneuver entry was determined with simulations. This imposes limits on the initial attitude angles, as well as the minimum entry speed and altitude. The exit condition from the maneuver is determined by either reaching a certain attitude or expiration of the time allocated for the maneuver. To guarantee a bumpless transition between the maneuver controller and the trim trajectory-tracking controller we initialize the integrator states such that the control surface commands are continuous. Prior to the maneuver the pilot brings the helicopter to the maneuver speed and altitude using the trim trajectory-tracking controllers, and chooses the direction of flight with a heading rate command. Before the maneuver, the helicopter is leveled by commanding a zero heading rate, whereas the forward speed and the altitude are maintained by the trim trajectory controller. The fully automatic maneuver sequence is engaged by a switch on the R/Ctransmitter. Upon the exit from the maneuver the helicopter accepts the exit altitude as a new More than 6 autonomous axial rolls have been executed in flight, and the state trajectories proved to be repeatable, and predicted well by the simulation [l]. The flight data for one of the autonomous rolls is given in Figure 2. 8.C.3-3 maneuver, which reduces the natural tendency of the helicopter to weathervane. At the end of the maneuver the pilot rapidly increased collective angle, which led to significant unwanted yawing immediately after the maneuver completion. This problem is commonly alleviated by WC pilots by using yaw rate control augmentation devices (hobby gyros) with PI controllers. Tight feedback gain on the integral of yaw rate error is necessary to alleviate collective-yaw coupling at low speed. Figure 4 shows cyclic and collective commands during the maneuver, the yaw rate command was not used by the pilot during the maneuver. We can see that commands are mainly piece-wise linear or piece-wise constant. 20 - lime, sec Figure 2. State Trajectories During An Autonomous Axial Roll. Simulated Immelman Maneuver An Immelman maneuver starts in a steady level forward flight (above 25 knots speed) and consists of half a loop (1 80 degrees in pitch) followed by half a roll. The maneuver provides a fast way of changing the direction of flight in a confined environment, e.g. between buildings in a city. 0 , -5 0 1 .J 2 3 4 5 6 1 8 - colieotb. deg Figure 3 shows an Immelman maneuver performed by an expert pilot on the instrumented XCell helicopter. -10 0 1 2 3 4 time. sec 5 6 7 8 Figure 4. Pilot Inputs During Manual Immelman Maneuver. ;P 1 0 -Fvd -2oa'; 2 Y 3 4 5 6 7 8 1 5 6 7 8 We have designed reference trajectories for the pitch rate during the pull-up half-loop, and for the roll rate during the following half roll. The trajectories contain the same four phases as the roll rate trajectory for axial roll given in Figure 1. The attitude values for initializing the ramp-down were determined in simulation. To increase maneuver exit velocity the pitch rate trajectory was selected such that the helicopter ends up in slightly nose down attitude (more than 180 degree loop). The collective was varied proportionally to cosine of the angle between local vertical and helicopter Z-axis. The time histories for the simulated Immelman are given in Figure 5. ,m1sec 3 ' llma, sec Figure 3. State Time Histories During Manual Immelman Maneuver. A proportional yaw rate feedback to the tail rotor pitch was used. As can be seen, the helicopter slows down to a zero forward speed during the 8.C.3-4 the helicopter spends longer time facing down; this will improve weathervaning. 20 -20 An 3 1 2 3 4 5 --- 50 3 4 6 7 6 7 - Altihde. rn I 5 I - Fwd Speed, mlm 0 -I, I 1 2 3 4 5 6 ? 8 'I 2 3 4 5 6 7 8 1 2 3 4 time. 88c 5 6 7 8 80' / 1 2 a 3 4 time, sec 5 6 7 t 2@ { - Altitude, rn 1 I 15 Figure 5. State Time Histories For Simulated Automatic Immelman Maneuver 10 0 The helicopter consistently exits the maneuver with the attitude very close to the desired: heading 180 degrees different from original, roll angle close to zero, pitch angle around 10 degrees nose down. However, after the exit from the maneuver and reengagement of the trim trajectory-tracking controllers the altitude hold increases collective, which leads to about 25 degree heading change. Much higher heading change was also observed in flight during manually executed Immelman maneuvers. This deficiency can be corrected by tighter turn rate tracking in the trim trajectory controller, or by making a wider loop to retain higher airspeed after the exit from the maneuver. An autonomous aerobatic maneuver was executed on a miniature model helicopter. Maneuver execution strategies were inspired by human pilots [ 1,4, 131. Operationally useful Immelman and split-S maneuvers were demonstrated in the simulation, and will be executed in flight tests to validate the suggested general approach to autonomous aggressive maneuvering. Simulated Split-S Maneuver Acknowledgements The split-S maneuver is an Immelman in reverse: starting from a steady level forward flight the helicopter performs half a roll, then half a loop. The maneuver provides a very quick way of reversing flight direction without losing speed by tradiig potential energy €or kinetic energy. It can be used in a tight environment given sufficient altitude. MIT students Kara Sprague, Rodin Lyasoff, Chris Sae-Hoo and Allen Wu have helped implement the avionics upgrade. Partial funding for this research was provided by the office of Naval Research under a Young Investigator Award and by the NASA grants NAG2- 144 1, NAG2-1522. Similarly to other maneuvers, the reference trajectories for roll and pitch rates were designed with fixed ramp up time, maximum command until preset pseudo-attitude is reached, fixed ramp down and settling times. Figure 6 shows the state vector time histories during the simulated split-S maneuver. A significant collective increase upon the exit from the maneuver led to unwanted yawing. Higher airspeed upon maneuver can be retained if References Figure 6. State Time Histories For Simulated Automatic Split S Maneuver Conclusion [ 11Gavrilets V., I. Martinos, B. Mettler and E. Feron, 2002, Control logicfor automated aerobatic flight of a miniature helicopter. Proceedings of the AIAA Guidance, Navigation and Control Conference. Monterey, CA. [2] Sprague K., V. Gavrilets, D. Dugail, B. Mettler, E. Feron and I. Martinos, 2001, Design and 8.C.3-5 applications of an avionics systemfor a miniature acrobatic helicopter. Proceedings of the 20&Digital Avionics Systems Conference, Orlando, FL. [8] Lublin L. and M. Athans, 1996, Linear quadratic regulator control. In: The Control Handbook, W. S . Levine ed., CRC Press. [3] Gavrilets V., B. Mettler and E. Feron, 2001, Nonlinear model for a small-size aerobatic helicopter. Proceedings of the AIAA Guidance, Navigation and Control Conference, Montreal, Canada. [9] Mettler, B., V. Gavrilets, E. Feron and T. Kanade. Dynamic compensationfor highbandwidth control of small scale helicopter. Presented at American Helicopter Society Technical Specialist Meeting, January 2002, San Francisco, CA. [4] Gavrilets V., E. Frazzoli, B. Mettler, M. Piedmonte and E. Feron, Aggressive maneuvering of small autonomous helicopters: a human-centered approach. Int'l Journal of Robotics Research, October 200 1, pp. 795-807. [lo] Mettler, B., M. Tischler, and T. Kanade. System identijication modeling of a small scale unmanned rotorcrajifor control design. Journal of the American Helicopter Society, Vol. 47, No 1, January 2002. [5] Bramwell, A.R.S., 1976, Helicopter Dynamics. Wiley and Sons, NY. [1 11 La Civita, M., T. Kanade, G. Papageorgiou, and W. Messner. Design andfright testing of a High-Bandwidth H-infinity loop shaping controller for a robotic helicopter. In proceedings of the 2002 AIAA Guidance, Navigation and Control Conference, Monterey, CA. [6] Feron, E., M. Brenner, J. Paduano, and A. Turevskiy,May-June 1998, Time-fiequency analysisfor the transferfunction estimation and application toflutter clearance, AIAA Journal of Guidance, Control and Dynamics, 21(3), pp. 3753 82. [12] Johnson, E. and S . Kannan. Adaptivefright controlfor an autonomous unmanned helicopter. In proceedings of the 2002 AIAA Guidance, Navigation and Control Conference, Monterey, CA. [7] La Civita M., W. Messner and T. Kanade, 2002, Modeling of small-scale helicopters with integrated fivst-principles and system identiJication techniques. Presented at the American Helicopter Society 58" Annual Forum, Montreal, Canada. [ 131 Piedmonte, M. and E. Feron, 1999. Aggressive maneuvering of autonomous helicopters: a humancentered approach. Proc. of 9* International Symposium on Robotics Research, pp. 413-419, Springer, US. 8.C.3-6