A light software architecture for a Humanoid Soccer Robot

In this paper, we present a software architecture that can be implemented on a humanoid platform with low- computational power to make it to autonomous. The platform we used is the Robovie-V of V-Stone. the application we chose is the RoboCup soccer competitions. We will present the simple behaviour selection architecture implemented with a finite state machine (FSM) in our robot. We will present the highly optimized algorithms used for image processing and we will give some hints on how it is possibile to extend the flexibility of a low computational power humanoid with a customized operating system. These solutions are quite general and can be applied to any humanoid platform with low-computational power. I. I NTRODUCTION Our robots are based on a modified Robovie-M platform of VStone. These robots are fully autonomous: CPU, power supply, sensor and obviously actuators are on-board. Two of the main activities of an autonomous robot are: processing the information gathered from the environment and select the most appropriate behavior to act in the environment. In the history of the RoboCup Humanoid League, there are many successful examples of autonomous humanoid robots performing complex tasks in a challenging environment, as the soccer game. Most of these robots needs relatively large computational resources to process the information gathered from the environment and to appropriately plan and execute their actions. For instance, the robots of Darmstadt Dribbl ers (12) mount an Intel PXA 255 at 512 MHz, the robots of NimbRo (11) mount a FSC Pocket Loox 720, which features a 520MHz XScale processor PXA-272 with 128MB RAM. In the end, the robots of the three-times World Champion TeamOsaka (13) use two CPUs: a GeodeLX 800 MHz with 256 MB of RAM and a Renesas SH2-7054 with 384 + 64 KB of RAM. The main CPU, i.e. the Geode LX 800, is used for image acquisition, image processing and robot behaviour control, while the second CPU, i.e. the SH2-7054, is used to generate the commands to be sent to the motors, to send the commands to the single joint motors and to close the loop on them. We present an autonomous platform that uses only this second one CPU (i.e. the Renesas SH2-7054) to completely control the robot from the image acquisition to the motor control. With our platform, image processing has been highly optimized in order to allow fast features extraction and information gathering. The vision module collects information that will be the input for the reasoning module that involves the development of behavior control. Complexity of soccer