At TNO Security, Defence and Safety I worked on path planning of a car-like autonomous vehicle, called the EyeRobot. In this project, the autonomous vehicle had to drive to a given location, but had to take into account the soil type is was passing: driving through sand requires more energy than driving over a road.
This problem is known as the Weighted Region Problem. This problem is different than standard path planning problems. For example, the shortest path between two points does not need to be straight line. I compared two algorithms for the weighted region problem.
Next to this, I developed a local obstacle avoidance algorithm that steers the vehicle past small obstacles such as garbage cans, while also taking into account the soil type. In the images, green areas are road, blue areas are fields and red areas are obstacles. Click here for the Master’s thesis.
In highly-constrained environments, motion synthesis techniques generally do not provide the desired level of control. The space of motions they can produce is limited by the data in a motion database. Procedural animation techniques generally do provide exact control.
|This is work done by Frank Baars, whom I supervised. The goal of this project was to use a motion planner (RRT) to generate an upper-body reach motion.
Unfortunately, the spine consists of many joints, heavily increasing the dimensionality of the motion planning problem. Therefore, we used a spine representation consisting of only three parameters. After planning, the individual joint orientations are determined using a linear model, resulting in an collision-free upper-body reaching motion.