Unmanned Systems Technology 013 | AutonomouStuff Lincoln MKZ | AI systems | Unmanned Underwater Vehicles | Cosworth AG2 UAV twin | AceCore Neo | Maintenance | IDEX 2017 Show report

The publish/subscribe software architecture is based around sending messages rather than streams of data. The system consists of nodes performing different functions: some will be a publisher, sending out messages, while others will subscribe to these broadcasts. The publisher doesn’t need to know who is subscribing to its message stream. This allows a scalable system to be built where new functions – nodes – can be easily added without having to rewrite the whole system. The Data Distribution Service (DDS) is an end-to-end middleware that uses a publish/subscribe architecture for real-time control systems. This is an industry standard managed by the Object Management Group, who also developed other high-level design protocols. It is written in an ‘interface description language’ that is defined by the OMG for message definition and serialisation. A distributed discovery system allows any two DDS programs, also called nodes, to communicate without the need for a central tool. One node broadcasts the availability of data, and another then subscribes to that output. That makes the system more fault tolerant and flexible than a typical operating system, and DDS has been used in large military systems, flight controllers and space systems with commercial implementations developed by companies such as RTI, PrismTech and TwinOaks Software. Volkswagen’s research labs for example has an agreement with RTI to use DDS technology, and space agency NASA uses the Robot Operating System (ROS 2.0) on top of DDS. The benefits of end-to-end middleware such as DDS are that there is much less code to maintain, and the behaviour and exact specifications of the middleware have already been documented. DDS has recommended-use cases and a software API so that developers can work with it at different levels. A drawback of using end-to-end middleware though is that an operating system such as ROS must work within that existing design. If the design did not target a relevant-use case or is not flexible, it might be necessary to work around the design – defeating the object of using it in the first place. Publish/subscribe The DDS interface is well-defined to allow middleware such as ROS to provide modular services 25 AutonomouStuff Lincoln MKZ | Dossier algorithms from AS and an Ethernet output back to the compute engine. This is used to provide a reference image that can be superimposed on top of the Lidar point cloud to check that the two sets of data match. One of the cameras is placed near a front window pillar to monitor a lane, while one in the back of the car provides a reference image for the laser scanners when changing lanes. AS is also working on its own camera- based lane modelling algorithms that would provide longer range and a wider FOV than the Mobileye system. Radar There are five Delphi ESR 76 GHz radar sensors on the MKZ to provide object detection. One is mounted on the centre of the front bumper and has a range of up to 175 m with a narrow FOV of 10 º, or a wider FOV of 45 º but a shorter range of 60 m. Then there are 70 m range sensors on each corner of the vehicle with a 150 º FOV to cover any blind spots and detect any cross-traffic. The front radar is used in the highway scenario to track a car in front, where you only need to know what’s in the lane ahead and in the next lane. The side detection radars on the bumpers need a focused beam but the FOV covers the blind spots and overlaps with the front sensor to give a 360 º ‘radar bubble’ around the car. “The idea is to have multiple sensor FOV with multiple modalities,” says Hambrick. The radar sensors use both Ethernet and CAN. “We get the raw detection data out of the radar, which requires filtering and processing, and this gives a lot of reflectivity data, like a point cloud. “Then there is track data, which is the grouping of the detections. For example, you may get ten detections on the back of a car but the track will turn it into one detection point,” he says. “We use both but fuse the higher level output of different sensors such as the object data from the Lidar and the Unmanned Systems Technology | April/May 2017