The SAVIE Project
The purpose of the SAVIE project is to develop the next generation navigation systems for AGVs that are both flexible and safe.
To acheive this the following scientific objects will be adressed:
The development of a new self-localization method that exploits non-artificial features of the environment. To achieve this objective, we need to address very challenging scientific issues such as data association and uncertainty handling. A previous result can bee seen at http://www.youtube.com/v/4iUgpse8XIQ from a previous project MALTA, where the truck navigates using 2D laser scanners to detect reels and pillars as features. The sensor modality that primarily will be investigated are cameras.
The development of a method for collision-free navigation at high speeds in the presence of static and moving objects. The method should ensure that the goal location is reached in a smooth and timely manner. Collision-free navigation covers issues related to vehicle modeling and control, obstacle detection and modeling, as well as path planning.
Detecting and tracking humans in the vicinity of the vehicle. Such information is to be used by the navigation system to ensure safe movement through adapting the behavior of the vehicle, such as reducing the cruise speed or even stopping the vehicle when humans are detected. Our main focus will be on modeling the environment and the movement of other objects using predictive models in order to navigate without collisions.
Generating paths using traffic information, such as loading/unloading areas, free navigation areas, one-way roads, crossings, etc. (see the figure below). The idea is to decrease the complexity of the path search process with the aim to obtain a method that is fast and yet produces paths of good quality (smooth and cost-aware).
Incorporating traffic information in monitoring the execution of paths. This includes 'right hand' rule in crossings, 'give way' rule when secondary roads meet main roads, etc. Special rules may apply in free navigation areas (see the figure below). Traffic rules are important to make the behavior of the vehicle predictable for humans present in the environment.
The project is a collaboration between the
Learning Systems Lab at
Kollmorgen Automation AB,
Linde Material Handling.
This project is co-sponsored by the KK-foundation.