【導(dǎo)讀】索和救援，甚至也能用于娛樂和游戲方面。每條腿的端部是由十二個魚鉤組成的夾持裝置，這些鉤子以一種特殊的方式對。齊使得每個鉤子可以在墻壁的表面獨(dú)立運(yùn)動。這樣的設(shè)計具有不需要用于緊壓避免來平衡。該算法考慮了機(jī)器人的運(yùn)動學(xué)以及施加在足墊的接觸力。計還提供給機(jī)器人審查其夾持力的能力以達(dá)到和保持在其附連到墻壁的高可靠性。最終，實(shí)驗(yàn)證明了特殊的夾持裝置的高可靠。性和工作區(qū)域并且具有重要的軍事和民用的優(yōu)勢。為能夠以自行和可靠的方式移動。此外，機(jī)器人還應(yīng)該要小巧、結(jié)構(gòu)緊湊以及便于單人攜?；谶@種設(shè)計的機(jī)器人被稱為CLIBO。我們用CLIBO實(shí)驗(yàn)表明了可靠的爬壁是可行的。以及協(xié)助搜尋和救援行動。器及爆炸物的平臺。一種可替代的腿構(gòu)造已經(jīng)進(jìn)行了檢驗(yàn)。一系列鉤子抓握實(shí)驗(yàn)同時表明鉤子能夠約束彼此抓取物體的干。這種旋轉(zhuǎn)方式可以防止夾緊裝置主體和墻壁發(fā)生碰撞。因此，需要一個基于從伺服馬達(dá)獲得的位

　　

【正文】 d is used in real time for determining what the joint angles are in order to position the leg EEs at a desired the distance Z has been determined by the user’s interface according to environment, the 4 angles can subsequently be calculated by (4)–(8). . Equilibrium analysis The legs are posed of smart servo motors able to measure torque operating on the leg joints. Using this feedback we can calculate the force acting on the EE based on the torque of the joints. The force calculation contains the gravitation force acting at the links’ center of mass. The data from determining the reaction forces acting on the leg EEs indicates one of two states. Large 21 forces indicate that the legs are overloaded. This is dangerous for the robot’s stability and needs to be dealt with immediately. Small forces can indicate that a leg has been detached from the wall. The configuration parameters vector vp, which consists of the four joint angles of the actuators θ1, . . . , θ4, the orientation angle of the central body θ0 and its global position θw, dW can be defined as follows: where θw and dW (Fig. 4) are the position parameters of the robot related to the global frame and are given by let rF be the vector from the origin of the global frame to the EE force Jacobian will then be: and the gravitational forces Jacobian: where the vector from the origin of the global frame to the ith link center of mass. 22 For each leg, the torques, acting on joints θ1, . . . , θ4 and on the central body θw, dw, θ0 due to the reaction force f and the links’mass mi, are: where (fx, fy, fz) are the force vectors acting on the leg’s have received a vector of torques (and one force Fw): whereMw, Fw,M0 are torques and force acting on the central , these parameters have no significance in our case. The other four parameters M1, . . . ,M4 are the torques acting on the joints of the legs. These parameters are measured by the servo motors. Therefore, we obtained four equations: 23 These four equations are the torques of the joints which features four unknown parameters fx, fy, fz , θ0. These equations are solved numerically to obtain the contact force. As expected, the solution shows that the expressions for M1, . . . ,M4 are independent of θw, dw. This means that the joints’ torques do not depend on the position of the robot on the wall. The control program can now solve these four equations in real time at any given position of the robot, providing us with information about the forces operating on the robot. The reaction force analysis is performed for one leg at a time. The analyses are pared one to another in order to analyze the weight distribution on the robot’s legs. 3. Motion planning In this section we describe CLIBO’s motion planning algorithm which allows it to climb vertical, rough textured walls. The motion planning is based on the ability of the motors to measure the applied torque and therefore to estimate the contact force at the gripper. CLIBO’s control is based on active position control and not on active force control. This way, torques and force equilibrium are obtained passively. Active force control is not feasible with our hardware, due to torque error readings from the actuators internal torque sensor and active force control under such errors may cause loss of stability. Therefore, equilibrium is not checked. Instead,applied torques and contact forces are calculated constantly for each leg separately during the motion. The main assumption of the robots’ motion is that a leg will keep trying to get a grab on the wall and eventually succeed. If succeeding to grab only after multiple tries, there can be some central body configurations which will not be feasible. There is no a priori knowledge of the surface texture,hence this assumption is inevitable. . Lootion principle The principle of CLIBO’s lootion is based on the motion of the central body along a given path. The path for the central body is predefined by the user prior to climb. There is no prior knowledge of the surface to be climbed other than its perpendicularity,therefore, the footholds position are decided online while climbing. The path given by the user prior to climb is discretized into small segments. The robot moves its central body toward a temporary position in a segment of its path while searching for opportunities to move its legs. Fig. 4 shows the flow chart of the lootion algorithm. The robot receives from a higher level planner a path on the wall. We wish to move the robot’s 24 central body along the given parametric path S(ρ): R ?→ R2, where the parameter ρ ∈ [0, Γ ] is such that the Γ is maximal at the end of the path. Let Δρ be a path increment in the robot’s path is a step of the body center. We discretize the path into elements. Hence, the kth discrete point along the path is sk = Sk(kΔρ). DenoteΔsk = sk+1?sk as a discrete path element where sk+1 = S((k + 1)Δρ). Every increment is then subdivided into smaller segments with length δ to be executed by body movements. Hence, every increment in the robot’s path is a Δsk step divided into smaller, δ, substeps conducted by body movements. For every increment Δsk, the central body of the robot moves in δ steps along the linear line created by the start and the end of the increment. The motion of the central body is done by leaving the contact points in their current location and moving the central body in a coordinated fashion using the closed chain every substep δ, torque and angle are measured at the actuators. Using inverse kinematics and static analysis (Section 2) we obtain the robot’s EE positions and the forces that act on them.. The motion planning algorithm is a reactive algorithm which continuously checks for the following four states. In each state the robot reacts different