—Executing safe and precise flight maneuvers in dynamic high-speed winds is important for the ongoing commodi-tization of uninhabited aerial vehicles (UAVs). However, since the relationship between various wind conditions and its effect on aircraft maneuverability is not well understood, it is challenging to design effective robot controllers using traditional control de-sign methods. We present Neural-Fly, a learning-based approach that allows rapid online adaptation by incorporating pre-trained representations through deep learning. Neural-Fly builds on two key observations that aerodynamics in different wind conditions share a common representation and that the wind-specific part lies in a low-dimensional space. To that end, Neural-Fly uses a proposed learning algorithm, Domain Adversarially Invariant Meta-Learning (DAIML), to learn the shared representation, only using 12 minutes of flight data. This pretraining phase enables rapid online learning through a composite adaptation law, which only needs to update a set of linear coefficients for mixing the basis elements to effectively correct for the wind effects. When evaluated under challenging wind conditions generated with the Caltech Real Weather Wind Tunnel with wind speeds up to 43 . 6km / h ( 12 . 1m / s ), Neural-Fly achieves precise flight control with substantially smaller tracking error than state-of-the-art nonlinear and adaptive controllers. In addition to strong empirical performance, the exponential stability of Neural-Fly results in robustness guarantees. Finally, our control design extrapolates to unseen wind conditions, is shown to be effective for outdoor flights with only on-board sensors, and can transfer across drones with minimal performance degradation.

—Neural-Fly is a learning-based control method that allows rapid online learning by incorporating a pre-trained representations of unmodeled aerodynamics with adaptive control. Not is Neural-Fly robust to different tasks, environments, and drones, but also it builds off a standard control architecture, creating a straightforward way to begin integrating learning-based control into safety critical applications. In this workshop, we will present the latest results for Neural-Fly on field robots, discuss implementation considerations, and show case other applications of Neural-Fly, such as fault tolerant control.

The quadrotor is a $6$ degrees-of-freedom (DoF) system with underactuation. Adding a spherical pendulum on top of a quadrotor further complicates the task of achieving any output tracking while stabilizing the rest. In this report, we present different types of controllers for the nonlinear dynamical system of quadrotor and pendulum combination, utilizing feedback-linearization and control Lyapunov function with quadratic programming (CLF-QP) approaches. We demonstrated trajectory tracking for quadrotor-only case as well as quadrotor-pendulum-combined case.

Executing safe and precise flight maneuvers in dynamic high-speed winds is important for the ongoing commoditization of uninhabited aerial vehicles (UAVs). However, because the relationship between various wind conditions and its effect on aircraft maneuverability is not well understood, it is challenging to design effective robot controllers using traditional control design methods. We present Neural-Fly, a learning-based approach that allows rapid online adaptation by incorporating pretrained representations through deep learning. Neural-Fly builds on two key observations that aerodynamics in different wind conditions share a common representation and that the wind-specific part lies in a low-dimensional space. To that end, Neural-Fly uses a proposed learning algorithm, domain adversarially invariant meta-learning (DAIML), to learn the shared representation, only using 12 minutes of flight data. With the learned representation as a basis, Neural-Fly then uses a composite adaptation law to update a set of linear coefficients for mixing the basis elements. When evaluated under challenging wind conditions generated with the Caltech Real Weather Wind Tunnel, with wind speeds up to 43.6 kilometers/hour (12.1 meters/second), Neural-Fly achieves precise flight control with substantially smaller tracking error than stateof-the-art nonlinear and adaptive controllers. In addition to strong empirical performance, the exponential stability of Neural-Fly results in robustness guarantees. Last, our control design extrapolates to unseen wind conditions, is shown to be effective for outdoor flights with only onboard sensors, and can transfer across drones with minimal performance degradation.

We present Neural-Swarm2, a learning-based method for motion planning and control that allows heterogeneous multirotors in a swarm to safely fly in close proximity. Such operation for drones is challenging due to complex aerodynamic interaction forces, such as downwash generated by nearby drones and ground effect. Conventional planning and control methods neglect capturing these interaction forces, resulting in sparse swarm configuration during flight. Our approach combines a physics-based nominal dynamics model with learned deep neural networks with strong Lipschitz properties. We make use of two techniques to accurately predict the aerodynamic interactions between heterogeneous multirotors: 1) Spectral normalization for stability and generalization guarantees of unseen data and 2) heterogeneous deep sets for supporting any number of heterogeneous neighbors in a permutation-invariant manner without reducing expressiveness. The learned residual dynamics benefit both the proposed interaction-aware multirobot motion planning and the nonlinear tracking control design because the learned interaction forces reduce the modelling errors. Experimental results demonstrate that Neural-Swarm2 is able to generalize to larger swarms beyond training cases and significantly outperforms a baseline nonlinear tracking controller with up to three times reduction in worst-case tracking errors.

Realtime model learning proves challenging for complex dynamical systems, such as drones flying in variable wind conditions. Machine learning technique such as deep neural networks have high representation power but is often too slow to update onboard. On the other hand, adaptive control relies on simple linear parameter models can update as fast as the feedback control loop. We propose an online composite adaptation method that treats outputs from a deep neural network as a set of basis functions capable of representing different wind conditions. To help with training, meta-learning techniques are used to optimize the network output useful for adaptation. We validate our approach by flying a drone in an open air wind tunnel under varying wind conditions and along challenging trajectories. We compare the result with other adaptive controller with different basis function sets and show improvement over tracking and prediction errors.

We present a delay-compensating control method that transforms exponentially stabilizing controllers for an undelayed system into a sample-based predictive controller with numerical integration. Our method handles both first-order and transport delays in actuators and trades-off numerical accuracy with computation delay to guaranteed stability under hardware limitations. Through hybrid stability analysis and numerical simulation, we demonstrate the efficacy of our method from both theoretical and simulation perspectives.

Fixed-wing vertical take-off and landing (VTOL) aircraft pose a unique control challenge that stems from complex aerodynamic interactions between wings and rotors. Thus, accurate estimation of external forces is indispensable for achieving high performance flight. In this paper, we present a composite adaptive nonlinear tracking controller for a fixed-wing VTOL. The method employs online adaptation of linear force models, and generates accurate estimation for wing and rotor forces in real-time based on information from a three-dimensional airflow sensor. The controller is implemented on a custom-built fixed-wing VTOL, which shows improved velocity tracking and force prediction during the transition stage from hover to forward flight, compared to baseline flight controllers.

Precise near-ground trajectory control is difficult for multi-rotor drones, due to the complex aerodynamic effects caused by interactions between multi-rotor airflow and the environment. Conventional control methods often fail to properly account for these complex effects and fall short in accomplishing smooth landing. In this paper, we present a novel deeplearning-based robust nonlinear controller (Neural-Lander) that improves control performance of a quadrotor during landing. Our approach combines a nominal dynamics model with a Deep Neural Network (DNN) that learns high-order interactions. We apply spectral normalization (SN) to constrain the Lipschitz constant of the DNN. Leveraging this Lipschitz property, we design a nonlinear feedback linearization controller using the learned model and prove system stability with disturbance rejection. To the best of our knowledge, this is the first DNNbased nonlinear feedback controller with stability guarantees that can utilize arbitrarily large neural nets. Experimental results demonstrate that the proposed controller significantly outperforms a Baseline Nonlinear Tracking Controller in both landing and cross-table trajectory tracking cases. We also empirically show that the DNN generalizes well to unseen data outside the training domain.

Precise near-ground trajectory control is difficult for multi-rotor drones, due to the complex aerodynamic effects caused by interactions between multi-rotor airflow and the environment. Conventional control methods often fail to properly account for these complex effects and fall short in accomplishing smooth landing. In this paper, we present a novel deep-learning-based robust nonlinear controller (Neural-Lander) that improves control performance of a quadrotor during landing. Our approach combines a nominal dynamics model with a Deep Neural Network (DNN) that learns high-order interactions. We apply spectral normalization (SN) to constrain the Lipschitz constant of the DNN. Leveraging this Lipschitz property, we design a nonlinear feedback linearization controller using the learned model and prove system stability with disturbance rejection. To the best of our knowledge, this is the first DNN-based nonlinear feedback controller with stability guarantees that can utilize arbitrarily large neural nets. Experimental results demonstrate that the proposed controller significantly outperforms a Baseline Nonlinear Tracking Controller in both landing and cross-table trajectory tracking cases. We also empirically show that the DNN generalizes well to unseen data outside the training domain.

An automated flying transport vehicle that capitalizes on the strengths and complexities of a fixed and rotary winged aircraft. The air transport vehicle comprises a body aerodynamically designed to generate lift and a plurality of rotors that can generate lift as well as forward thrust from which a fixed wing portion of the air transport vehicle will begin to generate additional lift allowing for a sustained flight.

A new design method for multi-rotor aircraft with distributed electric propulsion is presented to ensure a property of robustness against rotor failure from the control perspective. Based on the concept of null controllability, a quality measure is derived to evaluate and quantify the performance of a given design with the consideration of rotor failure. An optimization problem whose cost function is based on the quality measure is formulated and its optimal solution identifies a set of optimal design parameters that maximizes an aircraft’s ability to control its attitude and hence its position. The effectiveness of the proposed design procedure is validated through the results of experimentation with the Autonomous Flying Ambulance model being developed at Caltech’s Center for Autonomous Systems and Technologies.

Hybrid vertical take-off and landing vehicles (VTOL) with lift production from wings and distributed propulsive system present unique control challenges. Existing methods tend to stitch and switch different controllers specially designed for fixed-wing aircraft or multicopters. In this paper, we present a unified framework for designing controllers for such winged VTOL vehicles that are commonly found in recent flying car models. The proposed method is broken down into nonlinear control of both position and attitude with forces and moments as inputs, and real-time control allocation that integrates distributed propulsive actuation with conventional control surface deflection. We also present a strategy that avoids saturation of distributed propulsion control inputs. The effectiveness of the proposed framework is demonstrated through simulation and closed-loop flight experiment with our winged VTOL flying car prototype.

This work examines the control of nonlinear dynamic systems to synthesize a flight controller for Bat Bot (B2). B2 is a bat-inspired Micro Aerial Vehicle (MAV) which has articulated arm wings with several actuated and passive joints. B2 is designed to mimic the flight apparatus of actual biological bats, which is distinguished from other animals such as birds because bats employ numerous joints–more than 20 joint angles per each wing–and exhibit sophisticated motion patterns. B2 has significantly fewer degrees of freedom (DoF) than biological bats and the existing movements in the mechanism of B2 are: flapping motion of the wings, folding and unfolding of the wings and dorsoventral movements of the legs. The current work contributes to recent attempts to produce autonomous flapping flight motions in a bio-inspired robot with nontrivial morphology. This work uses the Lagrange framework to develop a nonlinear dynamic model for B2, and it proposes a nonlinear controller based on the theory of singular perturbation in order to track desired attitude angles.

It is challenging to analyze the aerial locomotion of bats because of the complicated and intricate relationship between their morphology and flight capabilities. Developing a biologically inspired bat robot would yield insight into how bats control their body attitude and position through the complex interaction of nonlinear forces (e.g., aerodynamic) and their intricate musculoskeletal mechanism. The current work introduces a biologically inspired soft robot called Bat Bot (B2). The overall system is a flapping machine with 5 Degrees of Actuation (DoA). This work reports on some of the preliminary untethered flights of B2. B2 has a nontrivial morphology and it has been designed after examining several biological bats. Key DoAs, which contribute significantly to bat flight, are picked and incorporated in B2's flight mechanism design. These DoAs are: 1) forelimb flapping motion, 2) forelimb mediolateral motion (folding and unfolding) and 3) hindlimb dorsoventral motion (upward and downward movement).

This paper addresses the problem of motion planning for fast, agile flight through a dense obstacle field. A key contribution is the design of two families of motion primitives for aerial robots flying in dense obstacle fields, along with rules to stitch them together. The primitives are obtained by solving for the flight dynamics of the aerial robot, and explicitly account for limited agility using time delays. The first family of primitives consists of turning maneuvers to link any two points in space. The locations of the terminal points are used to obtain closed-form expressions for the control inputs required to fly between them, while accounting for the finite time required to switch between consecutive sets of control inputs. The second family consists of aggressive turn-around maneuvers wherein the time delay between the angle of attack and roll angle commands is used to optimize the maneuver for the spatial constraints. A 3-D motion planning algorithm based on these primitives is presented for aircraft flying through a dense forest.

In recent years, few works [1] have applied rigorous mathematics to design sophisticated flapping motions similar to bats. However, the morphological and kinematic features of flying vertebrates are still the underappreciated aspects of locomotor control designs for Unmanned Aerial Vehicles (UAVs). In this work we introduce Bat Bot, in short B2, shown in Fig. 1-(a), which is a biologically inspired articulatedwinged micro aerial vehicle that is capable of mimicking a variety of motions and maneuvers, including but not limited to steady level flight without a conventional tail, rapid turning, inverse perching and hovering. To achieve such goals, biologic bats were examined to determine certain key morphological features of B2. Because of the inefficiency of electric actuators compared to the biologic muscles of these mammals, certain compromises have to be made to enable B2 for self-sustained flight.(a) B2(b)(c)

This paper presents a systematic flight controller design based on the mathematics of parametrized manifolds and calculus of variations for the Bat Bot (B2), which possesses many articulated wings. Wing kinematics and morphological properties are crucial in the powered flight of flying vertebrates. The articulated skeleton of these mammals, which contains many degrees of actuation and underactuation, has made it difficult to understand the connection between the bat's flight dynamics and its intricate array of physiological and morphological specializations. B2 is a biomimetic micro aerial vehicle (MAV) that possesses similar morphological properties to a bat in order to duplicate bats powered ballistic motion. In an effort to design the advanced flight control algorithm for B2, this paper reports two major contributions. First, a systematic mathematical framework is introduced that evaluates the holonomically-constrained Lagrangian model of a flapping robot with specified active and passive degrees of freedom (DoF) in order to locate physically feasible and biologically meaningful periodic solutions using optimization. These are parametrized constraint manifolds; the flapping wing dynamics are governed by these manifolds. Second, calculus of variations and the well-recognized method of inverse dynamics are applied in order to synthesize the flight control algorithm for the flapping wings.