Unmanned Systems Technology 005 | Selex ES Falco UAV | Sense and avoid systems | RCV Engines DF70 | DSEI show report | Fuel cells | CUAV Expo, InterDrone and CUAV Show reports | SLAM

80 vehicle’s probable position after a non- linear response to a command or that of a landmark after an observation, for example. To determine the probability that the true state is within a small region, the algorithm looks into that region and counts the samples. The more there are or the higher their weighting, the higher the probability that the true state lies within that area. One disadvantage with this approach though is that a very large number of samples might be needed to represent the probability distribution properly if there are large areas of uncertainty, which imposes high computational costs. As this extremely brief glance at different flavours of SLAM algorithm suggests, the best choice for a particular environment depends on several factors including hardware limitations, the size of the environment the autonomous system must map and navigate, and the time available for processing. Large areas with many landmarks still present problems, which can be addressed by making smaller sub-maps and stitching them together later, but this relies on robust data association, which is another important area for improvement, according to a recent academic review of current approaches to SLAM. Multi-system SLAM A lot of current r&d work is focused on multi-vehicle SLAM, in which pairs or larger teams collaborate to find their way around an unknown environment. For example, a team from the Autonomous Systems Laboratory at ETH Zurich, Switzerland, has developed collaborative navigation in which a small UAV helps a four-legged walking robot. In one demonstration, available on video, the walker must find and follow a safe path from the start area to the goal position. Before the quadrupedal robot takes a step, a hexacopter with a camera flies over the mission area to track visual features for SLAM, produce a consistent map of the landmarks and create a high-resolution elevation map of the environment. The walker then prepares to navigate with the help of the hexacopter data, analysing the elevation map for slope steepness, local roughness and maximum step height. Based on the resulting ‘combined traversability map’, the team’s navigation planner finds a path to the goal position. The quadruped produces accurate estimations of its state by fusing signals from position encoders in its joints and an inertial measurement unit, and processing image data for localisation. It also continuously updates the elevation map with measurements from its laser rangefinder, checks the route for safety, and senses and avoids new obstacles. A video of this shows the walker safely navigating along a narrow raised metal path, around tight turns, avoiding a box placed in its path after it started its journey and down a ramp to the goal position. Because the walker was provided with an initial map by the hexacopter, some observers might consider that it did not conduct a real SLAM task; however, the overall system formed by the pair of autonomous systems did. Dec 2015/Jan 2016 | Unmanned Systems Technology A multi-robot SLAM capability has been demonstrated in which a map created by this hexacopter helps this quadrupedal walker find its way along a path to a goal position (Courtesy of ETH Zurich) A lot of current r&d work is focused on multi- vehicle SLAM, in which pairs or larger teams collaborate in an environment