Mobile Robot Homing Control Based on Odor Sensing.

摘要

As robotic systems transition fully out of the laboratory environment and into the real world, they will need to be more robust and autonomous in order to deal with different environmental conditions and increasing complex tasks. Many developers have attempted to solve this problem by using increasingly more sensors and computational resources to provide the control system with more information. Others have tried to more explicitly define the environment and task being performed. However, there are many relatively simple biological organisms that exhibit complex behaviors using limited resources. This dissertation reports on the use of a biologically-based solution to develop a robotic platform and control system that allow for emergent intelligence and robustness.;A new robotic platform, the EvBot III, was developed with ubiquitous modularity in software, hardware, and control systems as a goal. The EvBot III is comprised of (1) a differential drive base with an attached turret and sensor shield, (2) a StackableUSB(TM) single board PC-104 computer, (3) a general purpose data acquisition system (CRIM-Daq), (4) a modular control architecture, and (5) a 3D physics-enabled simulation environment. The driving portion of the base is designed such that different drive systems (legged system, ackerman drive, etc.) can be used without needing to change the main controller. Additionally, the sensor shield allows different configurations and sensing modalities to be switched out based upon the desired application. Therefore, the EvBot III is expected to decrease development time and accelerate the progress of robotic and computational intelligence research.;A flat, homogeneous sensorimotor control architecture was developed for the EvBot III. This network was developed in LabVIEW and run on a Windows 7 (64-bit) PC. Chemical sensing was selected as a test application because it is the most widespread sensing modality among living organisms. Here, six MQ-3 were used to sense the chemical signal that marked the EvBot III charging location. A simple alcohol homing algorithm was developed that normalized for sensor variation. This homing algorithm was used during training for the sensorimotor network, and produced a zigzag navigational strategy that is similar to ones found in nature. The sensorimotor network was developed to fully connect all sensorimotor elements, which included olfactory, power, and motor modalities. The sensorimotor experiments were conducted using various experimental methodologies to build the correct correlations between increased alcohol concentration and increased charge. Although the sensorimotor network developed correlations using this approach, it did not build sufficient correlation relating increased alcohol concentration with charging. Future work includes revisiting the experimental methodology employed, using more accurate alcohol sensors, and incorporating sensing modalities with a similar layout and response to the alcohol sensors.