The loads seen at each joint of a humanoid robot during various points in a gait can fluctuate greatly, often by an order of magnitude. For example the load seen at the ankle during the swing phase is inconsequential compared to the load seen when the same ankle is in the single support phase of a typical walking gait. As a result, feedback control techniques perform poorly in minimising trajectory errors and cope poorly with unforeseen changes in system dynamics. This research explores the addition of an adaptive control system inspired by the architecture of the cerebellum to improve system response. The cerebellar architecture learns to compensate the changes in load that occur during a cycle of motion. These control systems are heavily based on the Cerebellar Modeled Articulated Controller (CMAC) inspired by Albus in the mid 70's.
An electronic Inertial Measurement Unit (eIMU) has been mounted on the robot, in between the eyes. This is to simulate the human inner ear, largely responsible for balance in a human. This eIMU can provide a variety of measurements including 3 axis linear and angular acceleration. A closed loop control loop has been implemented to maintain torso inclination in both the pitch and roll axis while under external disturbances. Research is continuing to use this feedback to increase the stability of the robot while walking.
The current control boards are now two years old and are beginning to show their age. The new controllers being implemented have shifted away from the Texas Instruments DSP to the Motorola 68xxx Processor line. The MC68HC376 processor chosen includes a 16 channel TPU, ideal for multiple motor control. The old integrated H-bridges, which were very inefficient and heavy have been replaced with a semi-discrete solution, drastically improving efficiency and removing the need for heavy heatsinks.
The foot of the GuRoo is currently being redesigned. Good foot mechanics are crucial in humanoid robots, as the feet serve as the interaction point with the surface of the ground. This version will improve the performance of the robot with the addition of vibration damping in the soles, as well at the sensors to measure the centre of pressure, vital in the calculation of the Zero Moment Point.
The Vision system is currently capable of streaming a single image to a custom built vision processing board. This is being upgraded to stereo vision, with color segmentation, object recognition and localisation to be implemented on the Hitachi SH4 based vision board. A new mechanical design for the head is also underway, which will be shortly implemented on the GuRoo
This Page Last Updated:
15th Jan 2004
Copyright Damien Kee 2004
damien@itee.uq.edu.au