Requisitos de la documentación

This experiment demonstrates the performance of the proposed system in outdoor en- vironments. The experiment is conducted in a typical winter day at Philadelphia, PA, where the wind speed goes up to 20km/hr. The total travel distance is approximately

170m with a total duration of166s (Fig. 4.11). Representative images captured by the onboard camera are shown in Fig. 4.12. Note that the outdoor environment is largely un- structured, consisting of trees and vegetation, demonstrating the ability of the system to also operate in unstructured environments. However, we do note that this particular out- door environment is very righ in texture, which is favorable for vision-based approaches. In practice, we may run into featureless environments that lead to failure the proposed algorithm. In addition, GPS signal with varying quality may be available in outdoor en- vironmentswhich gives additional information for state estimation. This motivates the Ch. 5, which focuses on the development of fusing multiple heterogeneous sensors in a consistent way to improve system robustness in a wide variety of environments.

Figure 4.11: 3D map generated in outdoor experiment after loop closure

(a) (b)

Figure 4.12: Images from the onboard camera during autonomous navigation of complex outdoor environments.

4.6

Discussion

In this chapter, we propose a loosely-coupled, combined monocular-stereo approach that is able to accurately track the state of the MAV in real-time with 20Hz visual update rate, and 100Hz update rate after fusion with IMU. We decouple different components in the system to reduce computation load, and show that such decoupling leads to a fast algorithm that is able to run on mobile processors with limited computation. Our approach is sufficient for feedback control with speed up to4m/s.

However, we note that our vision-based approach fails in featureless environments such as rooms with only white walls. However, these kind of environments can be ideal for laser-based approaches as they perfectly satisfy the 2.5D assumption. This motivates and the development of multi-sensor fusion methodologies (Ch. 5) that are able to opti- mally fuse vision and laser, as well as other measurements, in a consistent manner.

Chapter 5

Multi-Sensor Fusion for Indoor and

Outdoor Operations

This chapter describes a methodology for fusing information from multiple sensors, in- cluding those presented in previous chapters (Ch. 3 and 4) to improve system robostness in a large variety of environments. The main goal of this work is to develop a modular and extensible approach to integrate noisy measurements from multiple heterogeneous sensors that yield either absolute or relative observations at different and varying time intervals, and to provide smooth and globally consistent state estimates in real time for autonomous flight. The first key contribution, that is central to our work, is a principled approach, building on [84], to fusing relative measurements by augmenting the vehi- cle state with copies of previous states to create an augmented state vector for which consistent estimates are obtained and maintained using a filtering framework. A sec- ond significant contribution is our UKF formulation in which the propagation and update steps circumvent the difficulties that result from the semi-definiteness of the covariance matrix for the augmented state. Finally, we demonstrate results with our experimental platform (Sect. 1.3 and Fig. 1.2(c)) to illustrate the robustness of our framework in large- scale, indoor-outdoor autonomous aerial navigation experiments involving traversals of over 440meters at average speeds of 1.5m/s with winds around10mph while entering

and exiting two buildings.

We aim to develop a modular framework that allows easy addition and removal of sensors with minimum coding and mathematical derivation. We note that in the popular EKF-based formulation [89, 114], the computation of Jacobians can be problematic for complex systems like MAVs. As such, we employ a loosely coupled, derivative-free Unscented Kalman Filter (UKF) framework [38]. Switching from EKF to UKF poses several challenges, which will be detailed and addressed in Sect. 5.2.1.

5.1

Multi-Sensor System Model

We define vectors in the world and body frames as(·)w_and(·)b _{respectively. For the sake}

of brevity, we assume that all onboard sensors are calibrated and are attached to the body frame. Themainstates of the MAV is defined as:

xt =

pw_t, Φw_t, p˙b_t, abb_t, ωbb_t, zbw_t T

wherepw

t = [xwt , ytw, ztw]Tis the 3D position in the world frame at timet. Note thatpwt

can also be interpreted as the position of the body frame at timetwith respect to the world frame(·)w_{, other parameters also follow similar interpretation. Φ}w

t = [ψtw, θtw, φwt]T is

the yaw, pitch, and roll Euler angles that represent the 3-D orientation of the body in the world frame 1_{, from which a matrixR}w

t that represent the rotation of a vector from the

body frame at timetto the world frame can be obtained.p˙b_tis the 3D velocity in the body frame. a_bb

t andωbbt are the bias of the accelerometer and gyroscope, both expressed in

the body frame. zbw_t models the bias of the pressure altimeter in the world frame. We consider an IMU-based state propagation model:

ut= ab_t, ω_tbT vt= _a vt, ωvt,bvt _T xt+1 =f(xt, ut,vt) (5.1)

whereutis the measurement of linear accelerations and angular velocities from the IMU

in the body frame. vt ∼ N(0,Dt) ∈ R13 is the process noise. avt and ωvt represent

additive noise associated with the gyroscope and the accelerometer. bvtmodel the Gaus-

sian random walk of the gyroscope, accelerometer and altimeter bias. The functionf(·)

is a discretized version of the continuous time dynamical equation [46].

Exteroceptive sensors are usually used to correct the errors in the state propagation. Following [84], we consider measurements as either beingabsoluteorrelative, depend- ing on the nature of the underlying sensor. We allow an arbitrary number of either abso- lute or relative measurement models.