Mapping cities using a modified autonomous smart car

Citation

(2007), "Mapping cities using a modified autonomous smart car", Industrial Robot, Vol. 34 No. 5. https://doi.org/10.1108/ir.2007.04934eab.005

Mapping cities using a modified autonomous smart car

Mapping cities using a modified autonomous smart car

A joint project between the University of Freiburg, Germany, and ETH Zurich, Switzerland, is on a mission to create large-scale maps of villages and cities using an autonomous driving car. The problem is an instance of the fundamental simultaneous localization and mapping (SLAM) problem but at a much larger scale in an outdoor environment. The team has outfitted a Smart car with a 5 SICK LMS laser range finder sensors, differential GPS, inertial measurement unit (IMU), and optical gyroscope (Figure 1); data from all these are integrated in a stochastic framework for performing SLAM. The stochastic framework used is based on the theory of information filters (IFs) that is closely related to the Kalman Filter (KF) often preferred by the SLAM research community. The information filter outperforms the KF when needing to combine information from multiple sensors such as in this case.

Figure 1 The modified smart car

Other than the localization and mapping software, the autonomous vehicle must also drive itself; as a result, it must process the information in the map learnt in order to decide which areas are drivable or not. The maps are augmented with traversability information in order to aid in this task. The resulting map is called a multi-level surface map (MLS).

The team recently presented results of their mapping algorithm with the car operating in an unstructured urban environment for long periods of time. Data was collected by driving the vehicle around the EPFL campus for a total distance of 2.3km. The resulting MLS map constructed with a resolution of 50cm was derived from the measurement of nearly 70,000,000 data points. The area mapped spanned 300 x 200m; the final MLS map required 55MB of computer disk storage space. Processing the data are not real-time but only 15min was needed to process the data.

The team is preparing the autonomous Smart car for an entry to the European Land-Robot Trial (ELROB) competition which is the European version of the DARPA Grand Challenge.

Courtesy Artificial Robotics and Intelligence (http://smart-machines.blogspot.com)