Robotic lifting of objects is important because of a wide range application such as workpiece transportation in factory and arrangement in home. Lifting of objects with minimum number of mobile robots influences the efficiency of transportation and arrangement. Previously, various techniques were proposed to realize lifting of objects. However, lifting of objects with mobile robots in a minimum amount of time was not taken into account.

Mobile robots are expected to perform daily or routine domestic tasks for humans, many transportation tasks begin from object lifting; therefore, this research focuses on lifting of everyday household goods for transportation. The everyday household goods are divided into many categories by considering the shape primitive representation of the grasping part and the undulation of the bottom surface. Most of everyday household goods are the objects which (i) the shape primitive representation of the grasping part is cylinder or box, and (ii) the undulation of the bottom surface is small. Therefore, the objects which are belonging to these categories were considered in this research. The purpose is to realize lifting of objects as quickly as possible to transport them at home. Most of everyday household goods are not too heavy. They can be lifted at most two mobile robots. Therefore, an approach of lifting of objects with only two mobile robots was proposed in this research.

In order to realize lifting of everyday household goods, mobile robot mechanism, an object handling system consisting of a Gripper robot (a mobile robot equipped with a gripper) and a Lifter robot(a mobile robot equipped with a lifter), and sensor configuration are first designed by considering simple mechanism, weight distribution and enough constrained degrees of freedom. In order to realize lifting of objects with minimum number of mobile robots, a strategy for lifting of objects with the Gripper robot should be first proposed. If the object cannot be lifted by the Gripper robot, the Lifter robot should be used for cooperative grasp. It is difficult to obtain the accuracy model of an object with sensors. Therefore, in order to know what the parameters of the object is important for lifting an object with two mobile robots, a strategy for lifting of a object with two mobile robots should be designed next. Finally, lifting of objects with the Gripper robot and lifting of objects with two mobile robots can be combined to realize lifting of objects with one or two mobile robots in a minimum amount of time. Therefore, this study is carried out in three stages.

In the first stage, the study is focused on lifting of objects by one mobile robot with a parallel-jaw gripper based on feature extraction and grasping trials.

Human always grasp an object after judging where to grasp based on the visual features of the grasping part. Therefore, features of the grasping part are important for a mobile robot to grasp an unknown object. It costs more time to obtain and process all the scanned data of an unknown object to extract the features of the grasping part for object grasping. Little information acquisition and processing can decrease the time for grasping. The inspiration comes from object grasping based on features and little information processing discussed above.

Lifting of objects in a minimal amount of time can be realized by extracting features of the grasping part from the partial shape information of the object.

Partial information regarding the shape of object is acquired by a 2D range sensor installed on the mobile robot. Three features for finding maximal contact area to generate a stable lift are extracted directly from the partial shape information of the object to determine the candidate grasping points. Three features are designed based on the following three conditions: (1) There are flat parallel surfaces or parallel tangent planes on the objects. The tangent plane passes the grasping points and it is tangent to the object surface. (2)The distance between the parallel flat surfaces or parallel tangent planes is not larger than the maximum opening width of the gripper. (3) There is no obstacle near the grasping part when a robot is lifting objects.

The object is lifted after the grasping points are detected. However, a stable lift may not be generated only using partial shape information, therefore, whether an object can be lifted by a robot is judged with a 2D range sensor after performing a lifting trial. If a lifting trial is failed, the robot will find other candidate grasping points quickly to perform lifting trials until a stable lift is generated or the object cannot be lifted by one mobile robot is known.

The proposed approach is tested with experiments. A 93.7% overall grasp success rate is sufficient to enable a robot to lift objects based on the partial shape information. We compared the proposed algorithm with 3D model construction with respect to the lifting time. The lifting time of the proposed method is only about 52.5% of that of the 3D model construction; thus, confirming the statement that the proposed method can realize lifting of an object as quickly as possible.

In the second stage, the study is focused on lifting of objects with two mobile robots by considering transition between stable states. Lifting of objects is described as a stable initial state and a stable handling state. A strategy for fast transition from a stable initial state to a stable handling state by using two mobile robots is proposed. The shape of an object is assumed to known for two mobile robots. During object lifting, the Gripper robot grasps and lifts up the object from one side. This provides a space between the object and ground that can be used by the Lifter robot. The Lifter robot moves to the insertion position and inserts the lifter into the space under the object. Finally, two robots perform circular motion at the same time, thus, the stable handling state can be realized by using the two mobile robots. Circular motion performing can be formulated as a constraint optimization problem. The goal is to realize transition from a stable initial state to a stable handling state in a minimal amount of time. Four constraints are considered for generating a stable lifting: (1) Robot motion model: The motion model of robots must be satisfied during fast transition between stable states. (2) Mechanical constraints of two mobile robots: The Gripper robot and the Lifter robot cannot tilt or slide during fast transition between stable states. (3) Relative position of two mobile robots: The center of two mobile robots should be aligned in the stable handling state to guarantee the quality of object handling. (4) Loaded state of the object: The contact point between the object and the ground should be on the lifter in the handling state. (5) No collision during transition: There should be no collisions between the object and the Lifter robot during fast transition between stable states. The penalty method and multi-start local search method were chosen to acquire optimum velocities of two mobile robots for circular motion.

Simulations are conducted for testing the effectiveness of the proposed method. From the results, it can be concluded that the proposed method results in a shorter transition time than the other two methods (GCM: The Lifter robot rotates around its center and the Gripper robot performs circular motion. LCM: The Gripper robot rotates around its center and the Lifter robot performs circular motion). The transition time of the proposed method is only about 39% of LCM.

In the proposed approach, the grasping point and lifting point are known to the robots. If the robots were controlled directly by using these points without a relative position sensing feedback, a low success rate would result. Therefore, object grasping and lifting experiments with two real mobile robots are conducted to show the efficiency of the relative position sensing feedback. A 100% success rate shows that the relative position sensing feedback guarantees that two real robots go through fast transition from a stable initial state to a stable handling state.

In the third stage, lifting of objects with one or two mobile robots in a minimum amount of time is proposed by combining both first and second stage. Lifting of objects by the Gripper robot based on feature extraction and lifting trials is used for detection of grasping points. The object is transported if it can be lifted by a single robot. Otherwise, the Lifter robot is needed. It is necessary to maximize the use of payload of a robot for guaranteeing that minimum number of mobile robots are used; therefore, the detected grasping points which a largest force applied on the gripper is selected for the Gripper robot. During the Gripper robot is moving around the object and performing lifting trails, the grasping position of the robot, the lift velocity of the lifter and whether the object is tilted or not are obtained. In generally, the lift velocity of the lifter is slower if the force applied on the gripper is larger; otherwise, the object is lifted quickly. Therefore, the detected grasping points with slowest velocity are selected for the Gripper robot.

The length of width of the object is measured using the odometry and the scanned data by considering measurement error. The proposed approach can guarantee that the object can be grasped and lifted successfully by two mobile robots even if there exist measurement error. The contact point between the object and the ground is measured by the Lifter robot for calculating the insertion position. Finally, the strategy that proposed in the second stage is used for lifting of objects with two mobile robots.

Experiments are conducted. A 89.6% overall lift success rate is accurate enough to realize lifting of objects with one or two mobile robots as quickly as possible. From the view of the lifting time, the proposed approach is especially suited for the case when the objects that can be lifted by one robot or cannot be lifted by the two robots are more than those that can be lifted by two robots.

Lifting of objects with two mobile robots in a minimum amount of time was proposed in this research. Based on the experiments, it can be verified that the proposed method is highly applicable to the real environments for lifting and transporting everyday household goods.

Future research in the lifting of objects with mobile robots includes: (1) Extended to wide range of application. (2) Extended to multiple mobile robots. (3) Extended to multiple objects.

劉 兆甲提出の本論文は「Lifting of objects with two mobile robots(2台のロボットによる物体のリフティング)」と題し,全6章より構成される.

この論文は,2台の移動ロボットによる床にある日用品の認識とリフティングの高速化高効率化問題を扱っている.ロボットの利用台数と作業完了時間という二種類の評価関数を扱っている.ロボットの利用台数の最小化は,家庭における作業の効率利用の観点から重要である.作業完了時間を短縮することは,無駄な時間を減らし生産力向上の観点から重要である.

第1章において,ロボットが家庭と生産現場に広く受け入れられ,様々なアプリケーションに適用されていることを述べている.ロボットの利用台数の最小化と作業完了時間の最小化という二つの指標が実用的な観点から重要であることを主張している.それら二つの問題に対する提案手法の概要を示している.提案手法について,家庭においてのさまざまな片付け作業への適応可能性,二つの評価関数の考慮,という観点から議論している.

第2章において,ロボットの利用台数の最小化と作業完了時間の最小化の観点から問題の定式化をしている.この章では,まずは扱う物体について述べている.次に二つの問題に対する評価関数について述べている.定式化において,ロボットの利用台数の最小化が第一優先として,作業完了時間の最小化が第二優先として使われている.最後に動作時にロボット間に過大な内力がかからないことと,作業中に対象物が振動しないことを考慮しながら,ロボットの機構と戦略を設計した.1台のロボットが物体を把持し持ち上げるグリッパロボット,もう1台のロボットが物体を下から支え持ち上げるリフタロボット,という構成をとる.

第3章において,1台の移動ロボットによる物体のリフティングに関する提案手法について述べている.高速把持するために,3台のスキャナ式距離センサを搭載した移動ロボットが,形状モデルを持たない未知物体の局所的な形状情報を獲得して把持位置を認識することで物体把持を行う方法論を提案している.移動ロボットが前進しながら距離情報を蓄積することで物体の把持位置を抽出する.物体に関する距離センサ情報が以下の3つの条件を満たすときにグリッパによる把持位置が存在するとみなす.(a)物体に平行な表面または平面が存在すること.(b)当該平行表面/平面の間の距離がグリッパの最大間隔よりも短いこと.(c)当該平行表面/平面の外側にグリッパを差し込める空間が存在すること.もしある一連の計測により,上記のような把持位置を抽出した場合には,把持位置に到達し把持する.もし把持位置が存在しない場合には,物体周縁を一定量周回した地点で同様な計測を行い,把持位置が計測されるまで周回を続ける.物体の部分情報により物体を高速把持する方法で,把持安定性が保証できない. ここでは,物体の部分情報と把持試行が一緒に使われて高速把持を実現するアルゴリズムを提案している.提案手法は,3Dモデル構築による方法,すなわちロボットが物体の周縁を一周して形状情報を取得する方法と比較して作業完了時間最小化に有効である.

第4章において,2台の移動ロボットによる物体のリフティングに関する提案手法について述べている.複数の小型移動ロボットの協調により,床にある物体を持ち上げ,搬送,位置決めする物体ハンドリング作業が非常に重要である.ここでは2台の移動ロボットによる物体持ち上げ動作をできるだけ短時間かつ確実に遂行することを目指す.2台のロボットの協調動作生成規則を設計した.(a)グリッパロボットがセンサを用いて物体の把持位置を認識,アプローチし,物体を傾けながら持ち上げる.(b)その持ち上げた隙間にリフタロボットがリフトを差し込む.(c)最後にグリッパロボット,リフタロボットそれぞれが同期してある一定角度の円弧運動をすることにより,目標状態(ロボットと物体が一直線上に整列する状態)に達する.このうち(c)の円弧運動を,物体ハンドリング時の力学的制約,ロボットと物体の干渉回避,ロボットの動作制約を考慮した制約条件付き最適化問題として定式化し,ペナルティ法とランダム多スタート局所探索法を用いて解くアルゴリズムを提案した.シミュレーションおよび実機実験により提案手法の有効性を示した.

第5章において,1台か2台の移動ロボットによる物体のリフティングに関する提案手法について述べている.ここで第3章と第4章での物体のリフティングに関する提案手法を統合した.まずは1台のロボットで把持位置を抽出して把持試行をする.把持試行の結果によってロボットの利用台数を自動的に決める.2台のロボットを用いる場合,少ないセンサを使うため,物体の重さと持ち上げる速度の関係を利用して把持試行した時の持ち上げる速度から把持位置を選定する.そのため,物体を最小台数のロボットでリフティングすることが保証できる.物体情報によるリフタロボットの挿入位置を計算するために,計測誤差を考慮しながら,物体を計測する.物体の密度が未知で,4章の方法で2台のロボットの協調運動の速度を求める時,力学分析を考慮しない.把持の安定性を確保するために,最後にセンサで持ち上げに成功するかどうかを判断する.提案手法は従来方法と比較してロボットの利用台数最小化と作業完了時間最小化について有効であることを実機実験により示した.

第6章において,結論と今後の展望について述べている.ロボットの利用台数最小化と作業完了時間最小化という二種類の評価関数に対して,日用品のリフティングに対するハードウエアと戦略を設計した.物体の部分情報から把持位置を抽出すること,把持試行すること,把持位置を選定すること,2台のロボットの搬送制御することを統合した方法を提案し,他手法に対する有効性を示した.提案手法は,様々な物体に適用可能であり,ロボットの利用台数最小化と作業完了時間最小化について有効である.

以上を要するに,本論文では,2台の移動ロボットによる高速かつ高効に日用品のリフティングを,最適化問題として定式化した.物体のリフティング作業に対するハードウエア設計と戦略設計を統合した手法を定量的に評価した.これによって,本論文は家庭用ロボットの設計に寄与するところが大きく,部屋の片付けシステムの改善に対し有用であると考えられ,重要なものであると言える.

よって本論文は博士(工学)の学位請求論文として合格と認められる.