ROS for Autonomous Robotics
The Southwest Research Institute (SwRI) is making use of the ROS framework on several ground vehicle programs including basic research, commercial, and military applications.
MARTI
In 2006, SwRI initiated a large, internally funded development program for unmanned ground vehicles. The Mobile Autonomous Robotic Technology Initiative (MARTI) program resulted in a wide range of unmanned vehicle technologies that have been applied to both commercial and military applications. The initial software architecture was roughly modeled on NIST's 4D/RCS architecture and used a proprietary framework for inter-process communication, simulation, and data analysis. Soon after the initial public release of ROS, SwRI began to transition the entire vehicle architecture to the new open-source framework. By the summer of 2011, the port of the MARTI vehicle platform to ROS was complete. Features of the platform include:
- Vehicle interfaces including LIDAR, cameras, IMU, GPS, vehicle CAN bus, and actuation system
- Real-time vehicle controller
- Multiple path planners
- Perception technologies including terrain classification, lane following, obstacle detection and recognition
- Persistent world model capable of fusing a priori map and terrain data with sensed information
- Various autonomy modes including tele-operation, dismount following, convoying, and fully autonomous navigation
- Simulation and data visualization tools
These capabilities leverage many ROS tools including ROS Core, Rviz, OpenCV, and PCL.
The MARTI platform is a stock 2006 Ford Explorer with additional actuators, computing and sensors. ROS is used at all architecture levels above the real-time actuation control.
On-road and Commercial Vehicles
Due to the inherent modularity of ROS, the MARTI technologies are being readily ported to other platforms and applications. Commercial vehicle manufacturers are leveraging the technologies for active safety system development. Camera-based perception technologies are especially exciting in this area due to their rugged design, low-cost, and high-capability.
SwRI has leveraged ROS to developed robust object detection and tracking capabilities such as this pedestrian detector
Off-road Navigation
In addition to on-road applications, SwRI is actively working on off-road navigation challenges. ROSs’ tight coupling with openCV has enabled rapid development of vision-based perception capabilities. Real-time modeling of the complex off-road environment is a key capability. SwRI is advancing the state of the art by fusing vision-based localization, mapping, terrain classification, and obstacle detection, tracking and recognition.
Off-road Navigation - Example of off-road terrain classification using visual texture methods
Off-road environment
Confidence image
Segmented image
Classification
Related Terminology
automation engineering • manufacturing systems • robotics • Robot Operating System • ROS • off-road commercial vehicles • on-road commercial vehicles• MARTI
