SLAM – simultaneous localisation and mapping – is a technique robots use to build a map of their environment, ascertain where they are within the map and assess whether they’ve been to a particular location before. It’s an incredibly important technique within navigation research, so important in fact that we currently have a team focusing on the topic.
We’ve interviewed two PAL Robotics interns, who are spending their six month placements helping to develop and test new approaches to SLAM. First up is Robotics Masters student from University of Genoa (Italy), Elena Rampone, who has kindly volunteered to answer our questions and explain how her research into loop closure and place recognition may help a robot recognise its environment.
This may be an impossible task, but how would you describe SLAM in less than 50 words?
A robot navigating in an environment needs to keep track of its movements by estimating its current position, building an internal map of the environment and recognising if it has already visited a particular place. This is SLAM. The estimations are performed by analysing data acquired by the robot’s sensors.
What exactly are you working on?
My project can be broken down into two steps. The first consists of adapting the company’s place-recognition framework, based on the Bag of Words algorithm, to 3D point clouds. Point clouds provide a 3D depiction of the world and are characterised by a compact descriptor, which allows us to efficiently compare them.
The whole system is ROS-based and tested using the Kitti dataset, a widely-used outdoor dataset of point clouds acquired by a Velodyne sensor.