Implementing Central Pattern Generators for Bipedal Walkers using Cellular Neural Networks

Implementing Central Pattern Generators for Bipedal Walkers using Cellular Neural Networks by Bharathwaj Muthuswamy Master of Science, Plan II in Engineering—Electrical Engineering University of California, Berkeley Bipedal (two legged) robotics is a very interesting field in its own right. Besides the obvious joy of studying a complicated nonlinear (albiet very elegant) phenomenon, the main advantage of building bipedal walkers is to understand how humans walk. Nevertheless, designing and implementing a stable bipedal walker is a challenge because of the many degrees of freedom in the mechanisms, the intermittent nature of the contact conditions with the environment, and underactuation [12]. Our approach in solving this problem is biomimetic, that is, we want to mimic biology mechanical construction and neural control are the main ideas to build efficient bipedal walkers. To expand upon this, here is a summary of our goals: 1. Develop a simulation environment incorporating SolidWorks and SPICE for studying the dynamics of bipedal walking without sacrificing accuracy. SolidWorks [5] is an industry standard tool for solid modelling.Thus, SolidWorks provides us with a 3-d computer model of our robot complete with physical properties like centre of mass and moments of inertia. SPICE (S imulation Program W ith Integrated C ircuit Emphasis) [6] is the industry standard simulation tool for circuits. By interfacing the SolidWorks model of our robot to the control circuitry model implemented in SPICE, we hope to study the motion of the bipedal walker before we even construct the real robot! 2. Design a nonlinear analog circuit (a cellular neural network or CNN) that models a Central Pattern Generator (CPG) capable of producing the patterns required for bipedal walking. 3. Design a biomimetic mechanical body. 4. Implement a nonlinear control law for the CNN that guarantees dynamic stability. The goal of my Masters’ project is to complete (1) and (2) above; (3) and (4) are my Ph.D. topics at the University of California, Berkeley. For the latest information on all the goals above, please i visit our project homepage: http://robotics.eecs.berkeley.edu/~mbharat/raptor My Masters’ project report is organized as follows: In chapter 1, I talk about the theoretical ideas behind the CNN and the CPG. Chapter 2 deals with a very specific CPG model, the Chemical Synapse CPG that helps in generating patterns for walking motion. Chapter 3 shows an implementation of this CNN using op-amps and also gives some SPICE simulation results. Chapter 4 talks about the simulation environment we developed. Chapter 5 talks about our implemetation. I conclude with a look at what we learned from this project and what we hope to accomplish for the future.