For service robots to be employed in normal human environments, autonomous navigation is an important requirement. To enable autonomous navigation, the robot needs to be able to build a map of any unknown environment. The problem of letting a robot build a map and simultaneously localizing itself within the same map is known in robotics as the Simultaneous Localization and Mapping (SLAM) problem.
An important problem in SLAM is known as the loop closing problem. This problem occurs when the trajectory of the robot contains a large loop. After traversing this loop, the accumulated pose error causes the algorithm to fail to recognize the robot has returned to its original position.
In this thesis a novel algorithm is introduced to improve the loop closing behaviour of a widely used SLAM algorithm, called FastSLAM. FastSLAM uses a particle filter to estimate the pose of the robot. The resampling step in the particle filter algorithm causes particle depletion, which inhibits correct loop closures. The proposed algorithm, called Ant Colony Optimization (ACO-)SLAM, uses ACO to improve the resampling step of the FastSLAM algorithm. ACO-SLAM does this by incorporating a measure of the map consistency and a measure for correct loop closures into the resampling step.
Simulations are used to compare the performance of both algorithms with varying noise settings. Robby, the personal robot developed by the Delft Biorobotics lab, is used for experiments to test the performance of the algorithms in a realistic human environment.

to be announced

An increasing fraction of our electricity is produced by weather-dependent sources.
To maintain a stable electrical grid, embedded processing in its components can be used to autonomously schedule electrical loads and match the volatile supply.
This literature research gives an overview of proposed autonomous demand response methods. It is shown that due to their storing capacity thermostatically controlled loads (TCLs) such as refrigerators and electric water heater are especially well-suited to such programs. Dierent TCL models are formulated and example implementations of a model predictive control (MPC) approach are presented. This is extended to MPC of multiple loads. Centralized, decentralized and distributed MPC strategies are evaluated on their fundamental control properties as well as their ap-plicability to the demand response problem. The strategies are compared on scalability, stability, robustness and feasibility of implementation. It is concluded that distributed model predictive control is a promising research direction to control large-scale popula-tions of TCLs and is likely to yield viable results.

Not allowed to be published.

to be announced

to be announced

to be announced