Advancing Actuation and Control in Flapping-Wing Robots

Abstract

Interest in flying insect robots (FIR) can be attributed to the potential benefits of such platforms. From search and rescue operations, farming automation, and military applications to space exploration these robots have ample use cases. These robots draw inspiration from their natural counterparts. The main goal of this thesis is to develop FIRs with improved agility, control, and maneuverability. Key contributions include automated trimming, precise voltage-to-torque mapping, novel drag-based longitudinal actuation, first principle model identification, model-based control, and discrepancy modeling of system dynamics. These advancements aim to revolutionize micro-aerial robotics and eventually enable applications in search and rescue, environmental exploration, and disaster response. FIRs are hand-assembled, leading to manufacturing inconsistencies. This results in two major issues, (1) initial rapid rotation and tumbling instead of straight takeoff due to inherent manufacturing torques, and (2) unknown mapping between input voltage and generated torque, causing inconsistent control responses. Correcting for inherent torques, typically around 1$\mu$Nm, is challenging due to the lack of precise sensors. Current methods are error-prone and time-consuming. This research introduces a torque measurement device using feedback from motion capture cameras, automating the correction process without user intervention or damaging airborne flights. Validation on two robots showed improved takeoff stability in open-loop flights. Manufacturing inconsistencies also contributes to varied voltage-to-torque mapping between different robots. This complicates building a generalized controller for tasks like hovering. Using an updated torque measurement device, this research measured torque via static angular deflections. The obtained mapping was validated and led to successful vertical takeoffs in open-loop flights. This dissertation also addresses the problem of insufficient pitch torques generated with flapping-wing robots. Flying robots, including quadrotors, tilt their thrust vector forward or backward to achieve longitudinal motion. Flying insect robots relies on piezo-actuated flapping wings because of unfavorable downward size scaling in motor-driven propellers. The pitch torque for these robots is generated by moving the mid-stroke position of the wings, essentially by flapping the wings in the front or back of the body. This tilts their thrust vector to achieve longitudinal motion. Piezoelectric actuators, however have a limited bending capacity on the piezo layer limiting the maximum forward and backward movement of flapping wings. This, in turn, limits the extent to which the robot can flap its wing to the front or back of the body. This research proposes to use a paddling-type motion of the wings in combination with pitch torque to achieve higher agility in the longitudinal motion. In the paddling motion, the wings are flapped faster in the forward stroke than in the backward stroke to generate net backward drag force or vice versa. For the first time, the longitudinal movement from a stable hovering position is shown in free flight solely using paddling-type motion. This dissertation also addresses modeling and optimal control of flapping-wing robots. Flying insect robots require a fast, responsive control system for sophisticated maneuvers. Traditional PID-type feedback control, with manually tuned gains, is insufficient. This research implemented the first optimal control demonstration on an FIR using a first principle model, enabling computationally efficient onboard control. Using an LQR, stable hovering and trajectory tracking were demonstrated on a 150 mg FIR, achieving translational velocities up to 25 cm/s. This research was further continued to enable high-speed maneuvers on these robots. As high speed maneuvering is critical for efficient exploration but are often not possible on these robots due to complicated unsteady flapping wing aerodynamics, manufacturing uncertainties, and rapid wear. This research used discrepancy modeling via sparse identification of non linear dynamics (SINDy) to learn stroke-averaged translational models from free-flight data. Besides drag coefficients, we found that 20\% of the discrepancy in the longitudinal direction is due to input pitch torque, and 30\% of the discrepancy in the lateral direction is due to input roll torque. After compensating these discrepancies a model was developed, leading to a 49\% increase in hovering precision and a 21\% increase in trajectory tracking precision, with speeds up to 26.5 cm/s. This work can enable high-speed, time-optimal maneuvers for insect-sized flapping-wing robots. The research presented in this thesis has the potential to make flying insect robots (FIRs) more practical for real-world use cases. Key advancements include enhanced maneuverability through a new actuation capability and improved model identification enabling fast trajectory tracking potentially leading to more efficient flights. Dynamic modeling and implementation of optimal control opens avenues for optimization under actuator constraints in different environmental conditions. Additionally, the development of a drag model and the establishment of a torque-to-translation speed relationship facilitate high-speed, precise control, making these robots suitable for navigating confined spaces in future.