Design, Estimation and Control of an Aerial Skeleton System with Distributed Rotor Actuation

Abstract

In this thesis, we present key theoretical components for realizing flying aerial skeleton system called LASDRA (large-size aerial skeleton with distributed rotor actuation). Aerial skeletons are articulated aerial robots actuated by distributed rotors including both ground connected type and flying type. These systems have recently attracted interest and are being actively researched in several research groups, with the expectation of applying those for aerial manipulation in distant/narrow places, or for the performance with entertaining purpose such as drone shows. Among the aerial skeleton systems, LASDRA system, proposed by our group has some significant advantages over the other skeleton systems that it is capable of free SE(3) motion by omni-directional wrench generation of each link, and also the system can be operated with wide range of configuration because of the 3DOF (degrees of freedom) inter-link rotation enabled by cable connection among the link modules.

To realize this LASDRA system, following three components are crucial: 1) a link module that can produce omni-directional force and torque and enough feasible wrench space; 2) pose and posture estimation algorithm for an articulated system with high degrees of freedom; and 3) a motion generation framework that can provide seemingly natural motion while being able to generate desired motion (e.g., linear and angular velocity) for the entire body. The main contributions of this thesis is theoretically developing these three components, and verifying these through outdoor flight experiment with a real LASDRA system. First of all, a link module for the LASDRA system is designed with proposed constrained optimization problem, maximizing the guaranteed feasible force and torque for any direction while also incorporating some constraints (e.g., avoiding inter-rotor air-flow interference) to directly obtain feasible solution. Also, an issue of ESC-induced (electronic speed control) singularity is first introduced in the literature which is inevitably caused by bi-directional thrust generation with sensorless actuators, and handled with a novel control allocation called selective mapping. Then for the state estimation of the entire LASDRA system, constrained Kalman filter based estimation algorithm is proposed that can provide estimation result satisfying kinematic constraint of the system, also along with a semi-distributed version of the algorithm to endow with system scalability. Lastly, CPG-based motion generation framework is presented that can generate natural biomimetic motion, and by exploiting the inverse CPG model obtained with machine learning method, it becomes possible to generate certain desired motion while still making CPG generated natural motion.

본 논문에서는 비행 스켈레톤 시스템 LASDRA (large-size aerial skeleton with distributed rotor actuation) 의 구현을 위해 요구되는 핵심 기법들을 제안하며, 이를 실제 LASDRA 시스템의 실외 비행을 통해 검증한다. 제안된 기법은 1) 전방향으로 힘과 토크를 낼 수 있고 충분한 가용 렌치공간을 가진 링크 모듈, 2) 높은 자유도의 다관절구조 시스템을 위한 위치 및 자세 추정 알고리즘, 3) 자연스러운 움직임을 내는 동시에 전체 시스템이 속도, 각속도 등 원하는 움직임을 내도록 할 수 있는 모션 생성 프레임워크로 구성된다.

본 논문에서는 우선 링크 모듈의 디자인을 위해 전방향으로 보장되는 힘과 토크의 크기를 최대화하는 구속 최적화를 사용하고, 실제 적용가능한 해를 얻기 위해 몇가지 구속조건(로터 간 공기 흐름 간섭의 회피 등)을 고려한다. 또한 센서가 없는 액츄에이터로 양방향 추력을 내는 것에서 야기되는 ESC 유발 특이점 (ESC-induced singularity) 이라는 문제를 처음으로 소개하고, 이를 해결하기 위해 선택적 맵핑 (selective mapping) 이라는 기법을 제시한다. 전체 LASDRA 시스템의 상태추정을 위해 시스템의 기구학적 구속조건을 만족하는 결과를 얻을 수 있도록 구속 칼만 필터 기반의 상태추정 기법을 제시하고, 시스템 확장성을 고려하여 반 분산 (semi-distributed) 개념의 알고리즘을 함께 제시한다. 마지막으로 본 논문에서는 자연스러운 움직임의 생성을 위하여 CPG 기반의 모션 생성 프레임워크를 제안하며, 기계 학습 방법을 통해 CPG 역연산 모델을 얻음으로써 전체 시스템이 원하는 움직임을 낼 수 있도록 한다.