EFFICIENT THRUST GENERATOIN IN FISH CAUDUAL FINS WITH POLICY SEARCH
Open Access
- Author:
- Shan, Yixi
- Graduate Program:
- Mechanical Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- June 22, 2018
- Committee Members:
- Dr. Bo Cheng, Thesis Advisor/Co-Advisor
Dr. Anne Martin, Committee Member - Keywords:
- Fish locomotion
Policy Search
Reinforcement Learning - Abstract:
- Fishes are remarkable swimmers that represent diverse forms and functions of aquatic locomotion. Fish locomotion has been under extensive investigations in terms of swimming patterns, morphological features, hydrodynamic performance, vortex dynamics, etc. Thrust generation is a crucial aspect of fish locomotion that depends on a variety of morphological and kinematic parameters. It is known that the shapes and stiffness of caudal fins affect the efficiency of thrust generation; however, it is still unclear how their effects depend on the kinematic patterns of the caudal fins. In this thesis, the kinematic parameter space of caudal fin motion is searched using reinforcement learning (RL) to optimize the efficiency of thrust generation with caudal fins of multiple shapes and stiffness. Specifically, a policy search RL algorithm, i.e., Parameter Exploring Policy Gradient (PEPG), is applied on a robotic fish model operating in a mineral-oil tank. The robotic fish actuates its caudal fin with two degrees of freedom, i.e., flapping and rotation. The thrust generated by the caudal fin and the actuation torque are measured by a six-component force/torque sensor while the robot is fixed rigidly in the tank. Efficiency of thrust generation is investigated in two stages. First, I test whether fin spanwise-rotation motion is beneficial by optimizing the rotation amplitude and phase delay (relative to flapping) for nine designs of caudal fins with three different shapes and three stiffness. The result shows that the rotation motion does not contribute to efficiency of thrust generation, as efficiency is maximized at zero-amplitude rotation for all caudal fins tested. Informed by this result, next I optimize the flapping amplitude and trajectory profile without fin rotation also for the nine designs of caudal fins. Result shows that smaller flapping amplitude results in higher efficiency and linear flapping trajectories are preferred over sinusoidal trajectories. In addition, flexible fins are found to be more efficient although they generate less thrust. Therefore, the results indicate that caudal-fin flapping amplitude is mainly determined by the thrust-generation requirement and there is no local optimal identified in the range of amplitude tested.