PAVE uses a variety of technology and software solutions to fulfill its goal of creating robust, adaptable and innovative autonomous platforms. When it comes to writing code that can function in real-time and provide accurate, reliable data, even the smallest task is important and special factors must be taken into account when dealing with robots that will be operating on surface streets and near humans. A large amount of information must be gathered– from the robot’s velocity, heading and trajectory, to the images it sees and the obstacles it detects, the algorithms designed for each project must be able to handle the vast amount of information coming in hundreds of times a minute, and be able to perform computationally intensive tasks on-the-fly.

PAVE‘s autonomous systems are based on the Sense-Plan-Act paradigm. Each robot must sense its environment, plan its movement through it, then act. The most powerful method available for sensing one’s environment is vision, and this plays a vital role in PAVE‘s autonomous platforms. Currently, the sensor preferred by the team is the Videre STOC color stereocamera. This device simultaneously captures two images that are slightly offset from one another, and by comparing these photographs can generate a 3D point cloud and depth map of the objects present in its field of view. This provides the robot with a three dimensional image of what is in front of it. From this, it can then discern the various objects in its sights.

Once the system is aware of its surroundings, it must analyze its environment and look for things such as obstacles and paths. While this may be an easy task for any human, it is difficult for even extremely powerful computers. To make a car drive itself, the vehicle must be capable of detecting lanes as quickly and as easily as a human, and should be able to recognize obstructions in its path. To do this, the PAVE software team has implemented various algorithms from cutting edge papers in the field.

Once a robot is aware of its surroundings, it can then begin to plan a path to move towards its goal point while avoiding obstacles, and in the case of Prospect 12, staying within the lanes of a road. The navigation approaches PAVE has take for Prospect 12 and Phobetor are quite different due to the nature of the environments that the robots are required to navigate. Whereas urban environments have highly structured and have predictable layouts, the open, almost random, terrain that is present at the Intelligent Ground Vehicle Competition requires an algorithm based on a completely different paradigm.

An extremely vital component of both Prospect 12‘s and Phobetor‘s software architecture is the IPC++ wrapper, which allows for the various processes that the robots run to communicate with one another through a central server. This greatly simplifies the process of adding new code into the software architecture by allowing any new programs to be added as individual modules into the IPC++ framework, so that the information they provide can be easily shared with other processes.

To read about a general overview of our obstacle detection software, visit its section on our website. Or, visit Phobetor‘s page to learn about how we overcame this problem for the IGVC competition.

If you are interested in learning more about lane detection, navigation, or IPC++, please visit their corresponding pages on our webpage.