MIR-Vehicle: Cost-Effective Research Platform for Autonomous Vehicle Applications
This paper illustrates the MIR (Mobile Intelligent Robotics) Vehicle: a feasible option of transforming an electric ride-on-car into a modular Graphics Processing Unit (GPU) powered autonomous platform equipped with the capability that supports test and deployment of various intelligent autonomous vehicles algorithms. To use a platform for research, two components must be provided: perception and control. The sensors such as incremental encoders, an Inertial Measurement Unit (IMU), a camera, and a LIght Detection And Ranging (LIDAR) must be able to be installed on the platform to add the capability of environmental perception. A microcontroller-powered control box is designed to properly respond to the environmental changes by regulating drive and steering motors. This drive-by-wire capability is controlled by a GPU powered laptop computer where high-level perception algorithms are processed and complex actions are generated by various methods including behavior cloning using deep neural networks. The main goal of this paper is to provide an adequate and comprehensive approach for fabricating a cost-effective platform that would contribute to the research quality from the wider community. The proposed platform is to use a modular and hierarchical software architecture where the lower and simpler motor controls are taken care of by microcontroller programs, and the higher and complex algorithms are processed by a GPU powered laptop computer. The platform uses the Robot Operating System (ROS) as middleware to maintain the modularity of the perceptions and decision-making modules. It is expected that the level three and above autonomous vehicle systems and Advanced Driver Assistance Systems (ADAS) can be tested on and deployed to the platform with a decent real-time system behavior due to the capabilities and affordability of the proposed platform.