TU Delft self-driving car anticipates pedestrian behaviour
A research team from the Department of Cognitive Robotics (CoR) has achieved a milestone in the area of autonomous driving in an urban environment. During the IEEE Intelligent Vehicles Symposium last month in Paris, they held a demo called ‘Interaction of Self-Driving Vehicle with Pedestrian’ in front of a large audience. They demonstrated that their self-driving car can anticipate the behaviour of a pedestrian: will the pedestrian continue to walk from the kerbside or decide to stand still? This happens by analysing visual signals such as the pedestrian’s head orientation and his/her distance to the kerbside. By using this contextual information, the self-driving car can initiate a useful manoeuvre (such as evasive action) as much as 1 second earlier compared to cases that only take into consideration the pedestrian’s position and speed.
Dariu Gavrila (head of the Intelligent Vehicles group): ‘We are proud that we, the Intelligent Vehicles group and the Learning and Autonomous Control research group, have managed to join forces and develop such a complex system within several months. It is a system that encompasses the entire processing chain from vehicle perception, situation analysis, planning to control. This has taken us a step closer to being in the position to also effectively introduce autonomous driving in a “tricky” busy urban environment.’
IEEE Intelligent Vehicles Symposium
The IEEE Intelligent Vehicles Symposium is organised every year by the IEEE Intelligent Transportation Systems Society and aims to bring together representatives from science, industry and government who work in the area of intelligent vehicles and provide updates on the latest developments in this field. Gavrila’s research team was well represented at the 30th edition in Paris: in addition to the above-mentioned demo, colleague Julian Kooij also co-organised a workshop entitled ‘Unsupervised Learning for Automated Driving’, and the group presented seven papers. Read more at IEEE IV.