Loading...

Modeling,Design and Simulation of Falling and Landing Process in a Robotic Cat

Sadati, Mohammad Hadi | 2013

1085 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 44126 (08)
  4. University: Sharif University of Technology
  5. Department: Mechanical Engineering
  6. Advisor(s): Meghdari, Ali
  7. Abstract:
  8. Motion dynamics of cat species has always been attractive to be studied. Flexibility in motion due to specific skeleton and complex muscle-skeleton mechanism, control concepts, special way of running, high speed direction change while moving, ability of twisting the body during free fall, and landing on four limbs were widely investigated in literature and the results have been used in different branches such as control, robotics and aerospace.In this project, kinematic and dynamic equations of cat species falling maneuver are derived using quaternions for a simple two-link robot, a three-link robot with tail, and a more complete eight-link model with the addition of legs which is designed utilizing the animal’s anatomy. Dynamic equations are derived based on the iterative TMT method which eliminates the highest order derivative in each step and results in a simplified matrix form, ideal for numerical calculations. A path planning algorithm, using semi-flat kinematic equations, similar to input-output linearization method in nonlinear control, is used in the absence of constraints for the simple two and three link models to calculate the preferable amounts of inputs for any desired output values. Then, to satisfy geometric, kinematic and dynamic constraints, a discrete direct single shooting optimization method is applied to estimate the desired input values, and its efficiency in comparison to cyclic input method is illustrated. This procedure is repeated for three different input series and the improvement in the maneuver performance for the model with tail is shown especially in presence of constraints. The maneuver is also analyzed in presence of legs’ motion and it is shown that their movements can only prepare the body configuration for impact without considerable improvement in the free fall features. To satisfy the control purpose, an extended Kalman filter is used to reject the noise generated by gyroscope which measures the spatial rotation angles and in combination an off-line model-based neural network identification method is used to compensate the uncertainties of the model. Two kinds of nonlinear controllers, a simple PI vs. a model-based neural network controller, using output-input linearization method are proposed and their performance has been investigated. Furthermore the free fall motion is improved by optimizing the dimensions and inertia properties of the links using Genetic algorithm which resulted in optimum parameters for the real model design.Then, a three-phase 2D landing is studied using force-control algorithms. The landing phases are: pre-landing in which the robot adopts the optimal configurations to absorb the in-joint impulses, single-contact in which the robot is settled on one pair of its limbs, and double-contact in which the second pair impacts the ground. In each step, desired input values are calculated so as to guarantee the stability of the body based on ZMP index, and reduce the impact force in contacting points using input-output linearization and single shooting optimization methods. The proposed steps are repeated using a model considering stiffness and damping for the contacting ground. Finally based on this model, a simple method neglecting impact equations is proposed and the results were investigated in comparison with the complete one
  9. Keywords:
  10. Optimal Motion ; Continuouse Impact ; Cat Falling and Landing ; Semi-Flat System ; Soft Landing ; ADAMS Software ; Input-Output Linearization ; Neural Network ; Model Based Control

 Digital Object List

 Bookmark

...see more