Autonomous material handling with ZEUX

Motivation and Scope

The goal of this project is to extend and improve material handling with the ZEUX platform. This involves all the needed steps for filling the bucket while maintaining a good pile shape. Problems to be addressed are highly multidisciplinary as the load cycle consists of both control-oriented problems as well as perception and navigation aspects. Due to the complexity and the broad scope of this project, multiple thesis works is to be defined based on the applicant's interest.

Specific Tasks:

Essentially three capabilities are required that all are interconnected.

1) How to load the material. Given an attach pose, the goal is to scoop up as much material as possible. This is to a great extent a control problem where the actuation commands are the drive-wheels as well as the different joints reaching from the machine to the bucket. This task varies greatly depending on the material used, ranging from simple homogenous material such as gravel to more complex non-homogenous materials with boulders. To handle complex materials perception capabilities is very helpful.

2) Finding where to dig. Given a pile, this task is to find out where you want to dig. This requires perception to form a suitable representation of the surroundings to reason upon as well as other constraints on where to dig. The perception part is also useful for boulder detection of non-homogenous material as well as to model and measure what areas are drivable by the machine. This also includes to find the pose where to drop off the material (for example on a dump-truck). This part provides the machines relative pose to different interest points.

3) Navigating and local planning of the vehicle. This is to allow the vehicle to drive around in the loading area and possible unload the material on a machine. This task involves local motion planning and positioning of the vehicle. It can be seen as the decision-making module of what to do next. It can put constraints on the overall plan on where to dig and based on incoming request priorities and choose to do pile maintenance, that is to clean the ground from material spill.

Necessary Skills:

Good programming skills is required for all tasks. Depending on the problem at hand; knowledge about control, machine learning, perception related task such as point cloud processing, map representations, motion generation and planning.

Benefits:

You will have the possibility to work with the next generation autonomous platforms for construction equipment and obtain skills within a rapidly emerging industry sector.

Contact

Henrik Andreasson () and Sara Afshar ()