氏 名 | シュ ジョンシャ 朱 忠祥 |
本籍(国籍) | 中国 |
学位の種類 | 博士 (農学) | 学位記番号 | 連研 第346号 |
学位授与年月日 | 平成18年3月23日 | 学位授与の要件 | 学位規則第4条第1項該当 課程博士 |
研究科及び専攻 | 連合農学研究科 生物資源科学専攻 | ||
学位論文題目 | Modeling and Autonomous Navigation of Farm Mobile Robots on Sloping Terrain ( 傾斜地における農用移動ロボットのモデリングと自律走行 ) |
||
論文の内容の要旨 | |||
In order to increase the efficiency of farming operations and lighten the labour intensity, agricultural robots and techniques for autonomous navigation in the field were proposed by many researchers. A farmer can employ agricultural robots to take over more of the repectitive and time consuming tasks such as seeding and tilling the land, weeding and harvesting the crops, so that he can spend his time more usefully. This study seeks to develop an autonomous navigation system for tractor operations on sloping
terrain for lightening the labour intensity and increasing the productivity.
The detailed objectives are In Chapter 2, an overview of the experimental control system was presented. All hardware and their specifications were listed and their applications in this research were also discussed. Some sensors were calibrated before doing the experiments. The chapter then discussed the object-oriented and multi-thread programming technologies. Chapter 3 presented the formulation of a neural network(NN) vehicle model for estimating vehicle behaviour on sloping terrain. In this research, a 7-6-5-3 NN vehicle model was structured. For ensuring that the NN model was accurate over a wider range, the training pairs were acquired along the lemniscate of Bernoulli courses. A supervised training method, called a back propagation(BP) algorithm, together with a genetic alogorithm(GA) was used to train the NN model. Especially, the GA explored a preferable global minimum area in the global space before the BP beginning to be executed. The NN vehicle model was validated by the lemniscate of Bernoulli courses and the sinusoidal course. Comparisons between the experiments and the simulations were conducted and discussed. In this chapter, the open-loop control characteristics of vehicle motion on sloping terrain were also investigated by the simulations and experiments. A two-hierarchy fuzzy logic controller(FLC) and a new path-tracking method were developed to guide a tractor along arbitary paths on sloping terrain in Chapter 4. The FLC organized the rule sets with a multi-layer architecture. The upper level FLC, according to the high-layer rule set, used the terrain slope angle and vehicle posture to select the lower level FLCs. Then the crisp output of every lower level FLC was inferred based on each low-layer rule set. Finally, these outputs were coordinated to produce the resultant amount of steering angle change desired for vehicle motion. The chapter detailed every stage in the design of the FLC, namely linguistic varriables and fuzzy sets, rule sets, inference engine and defuzzification. In the designed path tracking method, both the location and orientation of the navigation point were employed to determine the current steering angle for vehicle motion. The path-tracking controller was composed of feedforward and feedback component elements. The feedforward component was based on the curvature of the current navigation point and could be calculated from geometrical relationship of the vehicle bicycle model. The feedback component was determined by the designed two-hierarchy FLC. The theoretical work was experimentally tested using the autonomous tractor on a meadow with a slope angle ranging from 10° to 20°. Chaper 5 presented the development of a platooning control system of tractors on sloping terrain. Overview of the system incorporating the positioning systems was introduced. In this system, the absolute positions of the vehicles were measured and this made the system able to deal with follow-up motions for not only straight but also curvous courses. The development of the tractor platooning system mainly included two steps. One was to create a reference course for the following vehicle from the position points of the leading vehicle online and in real time. The other was to guide the following vehicle along the reference course. Both steps were carried out parallel. At the first step, for smoothing the trajectory of the leading vehicle, the least squares fitting processing was executed to filter some noise caused by vehicle vibration and uneven ground surface. According to a given lateral offset, the reference course for the following vehicle to track was determined based on the geometrical relationship between the leading and following vehicles' trajectories. At the second step, the path-tracking controller presented in Chaptor 4 was used for the following vehicle to follow up the leading one along the reference course. Feild tests were conducted with three types of courses on a meadow with a mean slope angle of about 7°. The conclusions drawn from the results of this research are as follows: (和訳) 農作業効率を改善すると同時に労働強度を軽減するために、数多くの研究者が圃場における ナビゲーション技術と農業ロボット技術を提案してきた。 このような技術を応用することにより農家は耕うん・整地・播種・除草や収穫などの農作業を、 効率的に実施することができる。 本論文は、労働強度の軽減化と労働生産力を高めるために、トラクタの自律走行システムの開発を行う。
詳細な目的は次の通りである: 第2章では、本実験の制御システムの概要を述べた。 全てのハードウェアとその仕様を記載し、本研究でのそれらの使い方を説明した。 実験に取り掛かる前に、一部のセンサ類の較正を行った。 また、本研究で使用したプログラミング言語はビジュアルC++である。 特に、その中のオブジェクト指向プログラミング方法とマルチスレッドのプログラミングテクノロジについても説明した。 第3章では、傾斜地での車両の挙動を予測するNN車両モデルの定式化を提案した。
本研究では、7-6-5-3の内部構造を持つNN車両モデルを構築した。
より広い範囲で、このNNモデルが成立するように、ベルヌーイのレムニスケート曲線のコースを設定し、
そのコースに沿った走行結果を教師信号とした。 第4章では、傾斜牧草地で予め与えられた任意の走行軌道に沿ってトラクタを誘導させるため、
上下2層で構成されるファジィ論理コントローラ(FLC)と新しい軌道追従手法を開発した。
FLCは多層構造でルール集合を組織した。
上層のFLCは上層のルール集合により、斜面傾斜度と車両の姿勢角を利用して、下層FLCの全制御過程を決定した。
さらに、下層のルール集合によって下層FLCの出力を推定した。
最終的に、全ての下層FLCの出力を総合化し、車両運動に必要な操舵角のレート(微分値)を算定した。
本章ではFLC、すなわち、言語の変数とファジィ集合・規則セット・推論処理機関と非ファジィ化の
全ての設計段階を詳細に検討した。 第5章では、傾斜地におけるトラクタ系のプラトーニング制御システムの開発について述べた。
位置を測定するシステムを含め、制御システムの概要を説明した。
本研究で採用した追従システムは絶対座標系を利用した新しいシステムで、かつ直線軌道追従だけでなく、
曲線軌道にも適用できるところに新規性がある。 |