A Distributed Control System for a Self-Contained Bipedal Robot

Abstract

Self-contained bipedal robots are emerging in our daily environment. As their tasks grow in complexity, controlling such robots will become a demanding endeavour. In the course of this dissertation a novel control infrastructure has been developed, im- plemented and successfully tested. The infrastructure distinguishes itself through a distributed architecture, using wireless communication and consistent deployment of many-core processors based on field-programmable gate arrays (FPGAs). The overall benefits comprise spared mass combined with vast computational power of five nodes of which each runs up to 40 truly parallel tasks in real-time with cus- tom communication structures. Reducing mass is crucial as in a human environment a lightweight robot is safer as it requires smaller forces to move its own mass and inflicts far less damage in case of an accident. The distributed approach avoids blind communication of large amounts of raw data throughout the robot. The sensors are located on the limbs in order to capture their angles and motions. And the valves and actuators of a pneumatic robot are ideally situated directly on the limbs, too. Mounting a controller on each of the shins, thighs and on the hips enables to colocate the controllers with the sensors and actuators there – in the course of this the wiring of the sensors effectively vanishes. Through the resulting degree of local processing the nodes become essentially autonomous, rendering robot- wide transmission of sensor data and valve control largely unnecessary. Using a wireless body area network (WBAN) to connect the distributed controllers on the robot’s limbs eases the robot’s construction and spares mass. Moreover, wireless technology enables each node to connect directly to each other node (N -to-N network), without quadratic growth of the hardware requirements. Despite the distributed architecture, the system remains straightforward and manage- able through the use of an integrated software/hardware co-design tool (ETHZ’s Active Cells) – and the high-level control becomes more concise, due to the abstraction inher- ently introduced by the distribution.