TODDLER AND PEGASUS, THE UNH WALKERS

by Andrew Kun

The UNH Robotics Laboratory has been involved in walking robot research for numerous years. Over these years the Laboratory has produced increasingly sophisticated versions of its two legged walker, and recently a quadruped robot. The goal of this article is to give the reader a feel for what these projects are all about - where we are now, and where we are headed.

The first question on most people's mind when thinking about biped or quadruped walking robots is what the purpose of such robots may be. The answer is that, for many types of terrain, walkers are much better suited than wheeled vehicles, therefore if we want robots to traverse these types of terrain it makes sense to build biped robots, or robots with more than two legs. One example is walking on rocky terrain, where wheeled vehicles break down easily. Humans, and four legged animals on the other hand can walk over rocky terrain with relatively little difficulty. Certain types of terrain, like stairs, are created with human bipedal walking in mind. Therefore, robots that would work in such an environment would have to be legged robots.

TODDLER, the UNH biped robot, is approximately 1 m tall and it weighs 11 kg. It has ten joints actuated by DC motors. Twenty sensors provide information about the joint angles, pressure distribution on the feet, and the acceleration of the biped body. The robot uses CMAC neural networks in its control architecture. This allows us to avoid using complicated mathematical equations in the control algorithm - instead TODDLER "learns" how to walk, just like a toddler would. And, just in case you did not know why we named our robot TODDLER, this should have been the moment of revelation for you . As for using CMACs instead of an accurate mathematical model, you can just imagine the level of complexity of the math describing a system with ten DC motors, and twenty sensors, and also the cost of the computer that could deal with this math in real time!

PEGASUS is the UNH quadruped robot. Its hardware is based on the UNH biped. PEGASUS is approximately 50 cm tall, and weighs about 22 kg. It has twenty joints, and thirty-eight sensors. This robot also uses CMAC neural nets to "learn" how to walk. Pictures of both robots and mpeg movies of TODDLER walking can be seen on the Robotics Lab home page at
http://www.ece.unh.edu/robots/rbt_home.htm.

One of the facts we are very proud of at the Robotics Lab is the high level of student involvement in research projects, and both TODDLER and PEGASUS are great examples of this. As a graduate student I designed and built much of the current on-board electronics of TODDLER as part of my Master's Thesis research. Two students helped me in this project - Steve Scalera, who received his MSEE from UNH in 1995, and Kim Roy, who earned her BSEE in May 1996. More recently Brant Buchika did much of the work that enabled us to increase the data throughput between the biped and off-board equipment.

PEGASUS was built by Steve, with the help of Kim, and myself. Brant continued the project investigating the trotting gait for the quadruped - this work earned him a Master's degree in June 1996. Currently there are two students working on control schemes for the robot: Jeff Woodward (Woody), who is working on his Master's Thesis, and sophomore Jon Scalera (we just wanted to make sure that there is at least one Scalera working in the lab at all times).

Andrew Kun ,Jeff Woodward, Jonathan Scalera, and Micheal Shannon with the biped and quadruped walking robots.

TODDLER can walk for extended periods of time without human support at a medium speed. However it has problems walking slowly. Walking slowly means keeping the stepping foot in the air for long periods of time, and this in turn means that the biped has to balance on one foot. Our main goal at this point is to "teach" TODDLER to walk slowly.

Brant used CMAC neural nets to control the trotting gait of PEGASUS. The main thrust of the current research on PEGASUS is using CMAC neural nets to control the stepping gait - this work is going to be the crux of Woody's Master's Thesis.