Design and Construction of a Humanoid Robot Platform for Walking Gait Research

The desire to one day see a robot assistant cleaning our houses or walking our dogs has led to an increased interest into humanoid robotics research, and more specifically humanoid locomotion. At present, due to the high cost of robot platforms, there are only a small number of labs in the world conducting meaningful research on full sized humanoid robots. The existence of a low-cost humanoid platform would pave the way for greater involvement and development in the field of humanoid locomotion. This thesis describes the complete design and construction of an affordable humanoid robot platform for walking gait research, from the mechanical structure and actuator selection, right through to the required electronics and power storage implementation. The software required to operate the robot is discussed, from low-level feedback control through to high-level motion planning. The distributed nature of computational resources employed on this robot is outlined, along with the interaction with the robots sensors and actuators. A position based control methodology is proposed and implemented using traditional feedback loops on the robot. Control parameters were initially hand-tuned but subsequently improved via the implementation of an off-line evolutionary algorithm. Shortcomings in the mechanical design limited the success of this control scheme, with significant positional error observed in all joints whilst executing a walking gait. Actuator non-linearities as well as significant flexion in the underlying structure contributed to this positional error. To compensate, a series of adaptive control techniques were in turn amalgamated with the initial control loop in an attempt to learn the system dynamics of the robot and provide adequate compensatory signals. These additional control schemes realised a slightly improved level of accuracy in simulation in the joint control space but not enough to compensate for the robots significant mechanical flexion. Extensive hand tuning of algorithm parameters and excessive memory requirements prevented their implementation on the real robot. This robot competed at two international robot competitions with acceptable results. In 2002, the robot competed in the Humanoid League of RoboCup02, in Fukuoka, Japan. The robot was the largest competing humanoid by a considerable margin and it achieved a ranking of 7th in both the freestyle and walking distance category, out of the ten humanoids competing. In 2005 the robot competed in the RoboCup competition, this time in Osaka, reaching the semi-finals of the penalty shootout for robots over 650 mm in height. The final robot was capable of rudimentary walking and other simple movements such as penalty shootout soccer skills, validating the structures ability to withstand the forces required to execute a walking gait