Online re-planning may be necessary to guarantee safe autonomous navigation of quadrotors in unknown environments. Gaussian processes (GPs), which are machine learning models used in tasks such as regression and classification, have been proposed in robotics to solve the planning problem.
GPs are an interesting tool for motion planning since they give a probabilistic perspective on the problem by formulating it as probabilistic inference. Such inference can be performed quickly, which allows to efficiently and incrementally re-plan the current trajectory. In this project, we will use GPs for planning trajectories for landing of a quadrotor platform in scenarios where safety is a key requirement (e.g., in proximity to humans, on hazardous surfaces) and investigate the advantages and disadvantages of such models compared to polynomial trajectories.
Requirements: - Experience in Machine learning preferable - Experience in motion planning in robotics preferable - Programming experience in C++ and Python.
Online re-planning may be necessary to guarantee safe autonomous navigation of quadrotors in unknown environments. Gaussian processes (GPs), which are machine learning models used in tasks such as regression and classification, have been proposed in robotics to solve the planning problem. GPs are an interesting tool for motion planning since they give a probabilistic perspective on the problem by formulating it as probabilistic inference. Such inference can be performed quickly, which allows to efficiently and incrementally re-plan the current trajectory. In this project, we will use GPs for planning trajectories for landing of a quadrotor platform in scenarios where safety is a key requirement (e.g., in proximity to humans, on hazardous surfaces) and investigate the advantages and disadvantages of such models compared to polynomial trajectories. Requirements: - Experience in Machine learning preferable - Experience in motion planning in robotics preferable - Programming experience in C++ and Python.
The final goal of this thesis is to benchmark GPs motion planning against polynomial trajectories in simulation. If promising results are achieved, we will deploy the algorithm developed in this thesis on a real system.
The final goal of this thesis is to benchmark GPs motion planning against polynomial trajectories in simulation. If promising results are achieved, we will deploy the algorithm developed in this thesis on a real system.