Situation Aware Perception for Safe Autonomous Robotics Systems

Volker Krueger, LU & Per-Erik Forssén, LiU

This project will address the areas of sensing, analytics and learning. We will develop a modular adaptable situation-aware approach for perception, such that autonomous robotic systems (ARS) such as autonomous cars or robots can decide in each situation how to sense. Consider the famous first accident of an autonomously driving Tesla, where the perception system missed a white trailer in front of a bright horizon because the sensing system was not able to adapt. Adaptability of sensing will make ARS safer, more robust to situation changes, and reduce costs and time for network training, programming and testing, thus making robots more accessible to SMEs. This will require advances in probabilistic modelling of perceptual outcomes, and will use latent space learning and latent space tracking to adapt the perception to the situation. Scientific progress will be evaluated in an agile manner under realistic scenarios on our robot platforms (robotic arms and mobile robots, autonomous model cars, in/outdoors).

Project number: C08